From 28929ad2132bc7c33124a6b2bbf3b9919911a718 Mon Sep 17 00:00:00 2001 From: Tim Angus Date: Sun, 13 Jul 2025 21:57:07 +0100 Subject: [PATCH] Update xiph.org internal libs --- Makefile | 9 +- .../include/ogg/config_types.h | 7 +- .../include/ogg/ogg.h | 1 - .../include/ogg/os_types.h | 20 +- .../src/bitwise.c | 61 +- code/libogg-1.3.6/src/crctable.h | 278 ++++ .../src/framing.c | 222 ++-- .../include/vorbis/codec.h | 2 +- .../include/vorbis/vorbisenc.h | 2 +- .../include/vorbis/vorbisfile.h | 2 +- .../lib/analysis.c | 2 +- .../lib/backends.h | 2 +- code/libvorbis-1.3.7/lib/barkmel.c | 63 + .../lib/bitrate.c | 2 +- .../lib/bitrate.h | 2 +- .../lib/block.c | 2 +- .../lib/books/coupled/res_books_51.h | 2 +- .../lib/books/coupled/res_books_stereo.h | 2 +- .../lib/books/floor/floor_books.h | 2 +- .../lib/books/uncoupled/res_books_uncoupled.h | 2 +- .../lib/codebook.c | 2 +- .../lib/codebook.h | 2 +- .../lib/codec_internal.h | 2 +- .../lib/envelope.c | 2 +- .../lib/envelope.h | 2 +- .../lib/floor0.c | 2 +- .../lib/floor1.c | 2 +- .../lib/highlevel.h | 2 +- .../lib/info.c | 22 +- .../lib/lookup.c | 2 +- .../lib/lookup.h | 2 +- .../lib/lookup_data.h | 2 +- .../lib/lpc.c | 2 +- .../lib/lpc.h | 2 +- .../lib/lsp.c | 5 +- .../lib/lsp.h | 2 +- .../lib/mapping0.c | 2 +- .../lib/masking.h | 2 +- .../lib/mdct.c | 2 +- .../lib/mdct.h | 2 +- .../lib/misc.h | 2 +- .../lib/modes/floor_all.h | 2 +- .../lib/modes/psych_11.h | 2 +- .../lib/modes/psych_16.h | 2 +- .../lib/modes/psych_44.h | 2 +- .../lib/modes/psych_8.h | 2 +- .../lib/modes/residue_16.h | 2 +- .../lib/modes/residue_44.h | 2 +- .../lib/modes/residue_44p51.h | 2 +- .../lib/modes/residue_44u.h | 2 +- .../lib/modes/residue_8.h | 2 +- .../lib/modes/setup_11.h | 2 +- .../lib/modes/setup_16.h | 2 +- .../lib/modes/setup_22.h | 2 +- .../lib/modes/setup_32.h | 2 +- .../lib/modes/setup_44.h | 2 +- .../lib/modes/setup_44p51.h | 2 +- .../lib/modes/setup_44u.h | 2 +- .../lib/modes/setup_8.h | 2 +- .../lib/modes/setup_X.h | 2 +- .../lib/os.h | 13 +- .../lib/psy.c | 26 +- .../lib/psy.h | 2 +- code/libvorbis-1.3.7/lib/psytune.c | 523 ++++++++ .../lib/registry.c | 2 +- .../lib/registry.h | 2 +- .../lib/res0.c | 5 +- .../lib/scales.h | 2 +- .../lib/sharedbook.c | 17 +- .../lib/smallft.c | 2 +- .../lib/smallft.h | 2 +- .../lib/synthesis.c | 2 +- code/libvorbis-1.3.7/lib/tone.c | 54 + .../lib/vorbisenc.c | 5 +- .../lib/vorbisfile.c | 27 +- .../lib/window.c | 2 +- .../lib/window.h | 2 +- code/opus-1.2.1/celt/arm/arm2gnu.pl | 353 ------ .../celt/arm/celt_pitch_xcorr_arm-gnu.S | 555 -------- .../celt/arm/celt_pitch_xcorr_arm.s | 551 -------- code/opus-1.2.1/celt/x86/x86cpu.h | 93 -- code/opus-1.2.1/silk/x86/VQ_WMat_EC_sse.c | 142 --- code/opus-1.2.1/silk/x86/main_sse.h | 248 ---- code/opus-1.2.1/src/mlp.c | 145 --- code/opus-1.2.1/src/mlp_data.c | 112 -- code/opus-1.2.1/src/tansig_table.h | 45 - .../celt/_kiss_fft_guts.h | 0 code/{opus-1.2.1 => opus-1.5.2}/celt/arch.h | 34 +- .../celt/arm/arm_celt_map.c | 31 +- .../celt/arm/armcpu.c | 112 +- .../celt/arm/armcpu.h | 13 + .../celt/arm/celt_fft_ne10.c} | 2 +- .../celt/arm/celt_mdct_ne10.c} | 2 +- .../celt/arm/celt_neon_intr.c | 85 +- .../celt/arm/fft_arm.h | 0 .../celt/arm/fixed_arm64.h | 0 .../celt/arm/fixed_armv4.h | 0 .../celt/arm/fixed_armv5e.h | 0 .../celt/arm/kiss_fft_armv4.h | 0 .../celt/arm/kiss_fft_armv5e.h | 0 .../celt/arm/mdct_arm.h | 0 .../celt/arm/pitch_arm.h | 8 +- .../celt/arm/pitch_neon_intr.c | 58 +- code/{opus-1.2.1 => opus-1.5.2}/celt/bands.c | 36 +- code/{opus-1.2.1 => opus-1.5.2}/celt/bands.h | 2 +- code/{opus-1.2.1 => opus-1.5.2}/celt/celt.c | 0 code/{opus-1.2.1 => opus-1.5.2}/celt/celt.h | 36 +- .../celt/celt_decoder.c | 489 +++++-- .../celt/celt_encoder.c | 208 ++- .../celt/celt_lpc.c | 103 +- .../celt/celt_lpc.h | 2 +- .../celt/cpu_support.h | 10 +- code/{opus-1.2.1 => opus-1.5.2}/celt/cwrs.c | 2 +- code/{opus-1.2.1 => opus-1.5.2}/celt/cwrs.h | 0 .../celt/ecintrin.h | 4 + .../{opus-1.2.1 => opus-1.5.2}/celt/entcode.c | 0 .../{opus-1.2.1 => opus-1.5.2}/celt/entcode.h | 4 +- code/{opus-1.2.1 => opus-1.5.2}/celt/entdec.c | 21 + code/{opus-1.2.1 => opus-1.5.2}/celt/entdec.h | 12 +- code/{opus-1.2.1 => opus-1.5.2}/celt/entenc.c | 11 + code/{opus-1.2.1 => opus-1.5.2}/celt/entenc.h | 11 +- .../celt/fixed_debug.h | 71 +- .../celt/fixed_generic.h | 14 +- .../celt/float_cast.h | 62 +- .../celt/kiss_fft.c | 0 .../celt/kiss_fft.h | 12 +- .../{opus-1.2.1 => opus-1.5.2}/celt/laplace.c | 101 ++ .../{opus-1.2.1 => opus-1.5.2}/celt/laplace.h | 9 + .../{opus-1.2.1 => opus-1.5.2}/celt/mathops.c | 5 +- .../{opus-1.2.1 => opus-1.5.2}/celt/mathops.h | 16 +- code/{opus-1.2.1 => opus-1.5.2}/celt/mdct.c | 0 code/{opus-1.2.1 => opus-1.5.2}/celt/mdct.h | 0 .../celt/mfrngcod.h | 0 .../celt/mips/celt_mipsr1.h | 7 +- .../celt/mips/fixed_generic_mipsr1.h | 0 .../celt/mips/kiss_fft_mipsr1.h | 0 .../celt/mips/mdct_mipsr1.h | 6 +- .../celt/mips/pitch_mipsr1.h | 0 .../celt/mips/vq_mipsr1.h | 14 +- code/{opus-1.2.1 => opus-1.5.2}/celt/modes.c | 3 + code/{opus-1.2.1 => opus-1.5.2}/celt/modes.h | 0 .../celt/opus_custom_demo.c | 0 .../celt/os_support.h | 15 +- code/{opus-1.2.1 => opus-1.5.2}/celt/pitch.c | 61 +- code/{opus-1.2.1 => opus-1.5.2}/celt/pitch.h | 11 + .../celt/quant_bands.c | 2 +- .../celt/quant_bands.h | 0 code/{opus-1.2.1 => opus-1.5.2}/celt/rate.c | 4 +- code/{opus-1.2.1 => opus-1.5.2}/celt/rate.h | 2 +- .../celt/stack_alloc.h | 6 +- .../celt/static_modes_fixed.h | 0 .../celt/static_modes_fixed_arm_ne10.h | 0 .../celt/static_modes_float.h | 0 .../celt/static_modes_float_arm_ne10.h | 0 code/opus-1.5.2/celt/tests/test_unit_cwrs32.c | 162 +++ code/opus-1.5.2/celt/tests/test_unit_dft.c | 181 +++ .../opus-1.5.2/celt/tests/test_unit_entropy.c | 383 ++++++ .../opus-1.5.2/celt/tests/test_unit_laplace.c | 94 ++ .../opus-1.5.2/celt/tests/test_unit_mathops.c | 266 ++++ code/opus-1.5.2/celt/tests/test_unit_mdct.c | 229 ++++ .../celt/tests/test_unit_rotation.c | 87 ++ .../celt/tests/test_unit_types.c} | 33 +- code/{opus-1.2.1 => opus-1.5.2}/celt/vq.c | 8 +- code/{opus-1.2.1 => opus-1.5.2}/celt/vq.h | 4 - .../celt/x86/celt_lpc_sse.h | 5 +- .../celt/x86/celt_lpc_sse4_1.c} | 13 +- code/opus-1.5.2/celt/x86/pitch_avx.c | 101 ++ .../celt/x86/pitch_sse.c | 0 .../celt/x86/pitch_sse.h | 48 +- .../celt/x86/pitch_sse2.c | 0 .../celt/x86/pitch_sse4_1.c | 51 +- .../celt/x86/vq_sse.h | 6 +- .../celt/x86/vq_sse2.c | 10 +- .../celt/x86/x86_arch_macros.h} | 32 +- .../celt/x86/x86_celt_map.c | 20 + .../celt/x86/x86cpu.c | 51 +- code/opus-1.5.2/celt/x86/x86cpu.h | 112 ++ .../{opus-1.2.1 => opus-1.5.2}/include/opus.h | 128 +- .../include/opus_custom.h | 7 +- .../include/opus_defines.h | 51 +- .../include/opus_multistream.h | 2 +- code/opus-1.5.2/include/opus_projection.h | 568 +++++++++ .../include/opus_types.h | 25 +- code/{opus-1.2.1 => opus-1.5.2}/silk/A2NLSF.c | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/API.h | 26 +- code/{opus-1.2.1 => opus-1.5.2}/silk/CNG.c | 6 +- .../silk/HP_variable_cutoff.c | 0 .../{opus-1.2.1 => opus-1.5.2}/silk/Inlines.h | 0 .../silk/LPC_analysis_filter.c | 8 +- .../{opus-1.2.1 => opus-1.5.2}/silk/LPC_fit.c | 3 +- .../silk/LPC_inv_pred_gain.c | 0 .../silk/LP_variable_cutoff.c | 0 .../silk/MacroCount.h | 2 +- .../silk/MacroDebug.h | 54 +- code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF2A.c | 2 +- .../{opus-1.2.1 => opus-1.5.2}/silk/NLSF_VQ.c | 2 +- .../silk/NLSF_VQ_weights_laroia.c | 4 +- .../silk/NLSF_decode.c | 0 .../silk/NLSF_del_dec_quant.c | 0 .../silk/NLSF_encode.c | 2 +- .../silk/NLSF_stabilize.c | 0 .../silk/NLSF_unpack.c | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/NSQ.c | 54 +- code/{opus-1.2.1 => opus-1.5.2}/silk/NSQ.h | 0 .../silk/NSQ_del_dec.c | 66 +- code/{opus-1.2.1 => opus-1.5.2}/silk/PLC.c | 71 +- code/{opus-1.2.1 => opus-1.5.2}/silk/PLC.h | 3 + .../silk/SigProc_FIX.h | 10 +- code/{opus-1.2.1 => opus-1.5.2}/silk/VAD.c | 22 +- .../silk/VQ_WMat_EC.c | 4 +- .../silk/ana_filt_bank_1.c | 0 .../silk/arm/LPC_inv_pred_gain_arm.h | 0 .../silk/arm/LPC_inv_pred_gain_neon_intr.c | 32 +- .../silk/arm/NSQ_del_dec_arm.h | 4 +- .../silk/arm/NSQ_del_dec_neon_intr.c | 28 +- .../silk/arm/NSQ_neon.c | 0 .../silk/arm/NSQ_neon.h | 4 +- .../silk/arm/SigProc_FIX_armv4.h | 0 .../silk/arm/SigProc_FIX_armv5e.h | 0 .../silk/arm/arm_silk_map.c | 7 +- .../silk/arm/biquad_alt_arm.h | 0 .../silk/arm/biquad_alt_neon_intr.c | 0 .../silk/arm/macros_arm64.h | 0 .../silk/arm/macros_armv4.h | 0 .../silk/arm/macros_armv5e.h | 0 .../silk/biquad_alt.c | 0 .../silk/bwexpander.c | 0 .../silk/bwexpander_32.c | 3 +- .../silk/check_control_input.c | 22 +- .../silk/code_signs.c | 0 .../{opus-1.2.1 => opus-1.5.2}/silk/control.h | 11 + code/opus-1.5.2/silk/control_SNR.c | 113 ++ .../silk/control_audio_bandwidth.c | 16 +- .../silk/control_codec.c | 22 +- code/{opus-1.2.1 => opus-1.5.2}/silk/debug.c | 12 +- code/{opus-1.2.1 => opus-1.5.2}/silk/debug.h | 25 +- .../{opus-1.2.1 => opus-1.5.2}/silk/dec_API.c | 75 +- .../silk/decode_core.c | 4 +- .../silk/decode_frame.c | 65 +- .../silk/decode_indices.c | 2 +- .../silk/decode_parameters.c | 0 .../silk/decode_pitch.c | 4 +- .../silk/decode_pulses.c | 2 +- .../silk/decoder_set_fs.c | 8 +- code/{opus-1.2.1 => opus-1.5.2}/silk/define.h | 7 +- .../{opus-1.2.1 => opus-1.5.2}/silk/enc_API.c | 58 +- .../silk/encode_indices.c | 6 +- .../silk/encode_pulses.c | 2 +- code/{opus-1.2.1 => opus-1.5.2}/silk/errors.h | 0 .../silk/fixed/LTP_analysis_filter_FIX.c | 0 .../silk/fixed/LTP_scale_ctrl_FIX.c | 11 +- .../silk/fixed/apply_sine_window_FIX.c | 8 +- .../arm/warped_autocorrelation_FIX_arm.h | 68 + .../warped_autocorrelation_FIX_neon_intr.c | 265 ++++ .../silk/fixed/autocorr_FIX.c | 0 .../silk/fixed/burg_modified_FIX.c | 10 +- .../silk/fixed/corrMatrix_FIX.c | 0 .../silk/fixed/encode_frame_FIX.c | 35 +- .../silk/fixed/find_LPC_FIX.c | 2 +- .../silk/fixed/find_LTP_FIX.c | 0 .../silk/fixed/find_pitch_lags_FIX.c | 2 +- .../silk/fixed/find_pred_coefs_FIX.c | 5 +- .../silk/fixed/k2a_FIX.c | 0 .../silk/fixed/k2a_Q16_FIX.c | 0 .../silk/fixed/main_FIX.h | 3 +- .../mips/noise_shape_analysis_FIX_mipsr1.h | 336 +++++ .../mips/warped_autocorrelation_FIX_mipsr1.h | 165 +++ .../silk/fixed/noise_shape_analysis_FIX.c | 0 .../silk/fixed/pitch_analysis_core_FIX.c | 40 +- .../silk/fixed/process_gains_FIX.c | 0 .../silk/fixed/regularize_correlations_FIX.c | 0 .../silk/fixed/residual_energy16_FIX.c | 8 +- .../silk/fixed/residual_energy_FIX.c | 2 +- .../silk/fixed/schur64_FIX.c | 2 +- .../silk/fixed/schur_FIX.c | 2 +- .../silk/fixed/structs_FIX.h | 0 .../silk/fixed/vector_ops_FIX.c | 2 +- .../silk/fixed/warped_autocorrelation_FIX.c | 4 +- .../silk/fixed/x86/burg_modified_FIX_sse4_1.c | 400 ++++++ .../silk/fixed/x86/vector_ops_FIX_sse4_1.c | 93 ++ .../silk/float/LPC_analysis_filter_FLP.c | 4 +- .../silk/float/LPC_inv_pred_gain_FLP.c | 0 .../silk/float/LTP_analysis_filter_FLP.c | 0 .../silk/float/LTP_scale_ctrl_FLP.c | 10 +- .../silk/float/SigProc_FLP.h | 14 +- .../silk/float/apply_sine_window_FLP.c | 4 +- .../silk/float/autocorrelation_FLP.c | 5 +- .../silk/float/burg_modified_FLP.c | 7 +- .../silk/float/bwexpander_FLP.c | 0 .../silk/float/corrMatrix_FLP.c | 10 +- .../silk/float/encode_frame_FLP.c | 34 +- .../silk/float/energy_FLP.c | 0 .../silk/float/find_LPC_FLP.c | 9 +- .../silk/float/find_LTP_FLP.c | 7 +- .../silk/float/find_pitch_lags_FLP.c | 4 +- .../silk/float/find_pred_coefs_FLP.c | 9 +- .../silk/float/inner_product_FLP.c | 2 +- .../silk/float/k2a_FLP.c | 0 .../silk/float/main_FLP.h | 15 +- .../silk/float/noise_shape_analysis_FLP.c | 2 +- .../silk/float/pitch_analysis_core_FLP.c | 36 +- .../silk/float/process_gains_FLP.c | 0 .../silk/float/regularize_correlations_FLP.c | 0 .../silk/float/residual_energy_FLP.c | 2 +- .../silk/float/scale_copy_vector_FLP.c | 0 .../silk/float/scale_vector_FLP.c | 0 .../silk/float/schur_FLP.c | 2 +- .../silk/float/sort_FLP.c | 6 +- .../silk/float/structs_FLP.h | 0 .../silk/float/warped_autocorrelation_FLP.c | 12 +- .../silk/float/wrappers_FLP.c | 10 +- .../silk/float/x86/inner_product_FLP_avx2.c} | 82 +- .../silk/gain_quant.c | 0 .../silk/init_decoder.c | 33 +- .../silk/init_encoder.c | 4 + .../silk/inner_prod_aligned.c | 0 .../silk/interpolate.c | 4 +- .../{opus-1.2.1 => opus-1.5.2}/silk/lin2log.c | 0 .../{opus-1.2.1 => opus-1.5.2}/silk/log2lin.c | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/macros.h | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/main.h | 70 +- .../silk/mips/NSQ_del_dec_mipsr1.h | 6 +- .../silk/mips/macros_mipsr1.h | 6 +- .../silk/mips/sigproc_fix_mipsr1.h | 0 .../silk/pitch_est_defines.h | 0 .../silk/pitch_est_tables.c | 0 .../silk/process_NLSFs.c | 6 +- .../silk/quant_LTP_gains.c | 0 .../silk/resampler.c | 10 +- .../silk/resampler_down2.c | 4 +- .../silk/resampler_down2_3.c | 0 .../silk/resampler_private.h | 0 .../silk/resampler_private_AR2.c | 0 .../silk/resampler_private_IIR_FIR.c | 0 .../silk/resampler_private_down_FIR.c | 2 +- .../silk/resampler_private_up2_HQ.c | 0 .../silk/resampler_rom.c | 0 .../silk/resampler_rom.h | 0 .../silk/resampler_structs.h | 0 .../silk/shell_coder.c | 0 .../silk/sigm_Q15.c | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/sort.c | 14 +- .../silk/stereo_LR_to_MS.c | 10 +- .../silk/stereo_MS_to_LR.c | 4 +- .../silk/stereo_decode_pred.c | 0 .../silk/stereo_encode_pred.c | 6 +- .../silk/stereo_find_predictor.c | 0 .../silk/stereo_quant_pred.c | 0 .../{opus-1.2.1 => opus-1.5.2}/silk/structs.h | 29 + .../silk/sum_sqr_shift.c | 0 .../silk/table_LSF_cos.c | 0 code/{opus-1.2.1 => opus-1.5.2}/silk/tables.h | 6 - .../silk/tables_LTP.c | 0 .../silk/tables_NLSF_CB_NB_MB.c | 0 .../silk/tables_NLSF_CB_WB.c | 0 .../silk/tables_gain.c | 0 .../silk/tables_other.c | 14 - .../silk/tables_pitch_lag.c | 0 .../silk/tables_pulses_per_block.c | 0 .../silk/tests/test_unit_LPC_inv_pred_gain.c | 130 ++ .../silk/tuning_parameters.h | 0 .../{opus-1.2.1 => opus-1.5.2}/silk/typedef.h | 3 + code/opus-1.5.2/silk/x86/NSQ_del_dec_avx2.c | 1074 ++++++++++++++++ .../silk/x86/NSQ_del_dec_sse4_1.c} | 210 +-- .../silk/x86/NSQ_sse4_1.c} | 249 ++-- .../silk/x86/SigProc_FIX_sse.h | 49 +- .../silk/x86/VAD_sse4_1.c} | 36 +- code/opus-1.5.2/silk/x86/VQ_WMat_EC_sse4_1.c | 173 +++ code/opus-1.5.2/silk/x86/main_sse.h | 299 +++++ .../silk/x86/x86_silk_map.c | 113 +- .../{opus-1.2.1 => opus-1.5.2}/src/analysis.c | 403 +++--- .../{opus-1.2.1 => opus-1.5.2}/src/analysis.h | 16 +- code/opus-1.5.2/src/extensions.c | 315 +++++ code/opus-1.5.2/src/mapping_matrix.c | 939 ++++++++++++++ code/opus-1.5.2/src/mapping_matrix.h | 145 +++ code/opus-1.5.2/src/mlp.c | 132 ++ code/opus-1.5.2/src/mlp.h | 60 + code/opus-1.5.2/src/mlp_data.c | 672 ++++++++++ code/{opus-1.2.1 => opus-1.5.2}/src/opus.c | 12 +- code/opus-1.5.2/src/opus_compare.c | 382 ++++++ .../src/opus_decoder.c | 621 ++++++++- code/opus-1.5.2/src/opus_demo.c | 1127 +++++++++++++++++ .../src/opus_encoder.c | 906 +++++++++---- .../src/opus_multistream.c | 0 .../src/opus_multistream_decoder.c | 98 +- .../src/opus_multistream_encoder.c | 177 +-- .../src/opus_private.h | 103 +- code/opus-1.5.2/src/opus_projection_decoder.c | 258 ++++ code/opus-1.5.2/src/opus_projection_encoder.c | 510 ++++++++ .../src/repacketizer.c | 148 ++- code/opus-1.5.2/src/repacketizer_demo.c | 247 ++++ .../include/opusfile.h | 16 +- .../src/http.c | 116 +- .../src/info.c | 14 +- .../src/internal.c | 4 +- .../src/internal.h | 4 +- .../src/opusfile.c | 71 +- .../src/stream.c | 71 +- .../src/wincerts.c | 6 +- .../src/winerrno.h | 4 +- 400 files changed, 16287 insertions(+), 4588 deletions(-) rename code/{libogg-1.3.3 => libogg-1.3.6}/include/ogg/config_types.h (74%) rename code/{libogg-1.3.3 => libogg-1.3.6}/include/ogg/ogg.h (99%) rename code/{libogg-1.3.3 => libogg-1.3.6}/include/ogg/os_types.h (87%) rename code/{libogg-1.3.3 => libogg-1.3.6}/src/bitwise.c (94%) create mode 100644 code/libogg-1.3.6/src/crctable.h rename code/{libogg-1.3.3 => libogg-1.3.6}/src/framing.c (92%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/include/vorbis/codec.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/include/vorbis/vorbisenc.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/include/vorbis/vorbisfile.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/analysis.c (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/backends.h (98%) create mode 100644 code/libvorbis-1.3.7/lib/barkmel.c rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/bitrate.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/bitrate.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/block.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/books/coupled/res_books_51.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/books/coupled/res_books_stereo.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/books/floor/floor_books.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/books/uncoupled/res_books_uncoupled.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/codebook.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/codebook.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/codec_internal.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/envelope.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/envelope.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/floor0.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/floor1.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/highlevel.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/info.c (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lookup.c (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lookup.h (94%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lookup_data.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lpc.c (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lpc.h (94%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lsp.c (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/lsp.h (94%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/mapping0.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/masking.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/mdct.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/mdct.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/misc.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/floor_all.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/psych_11.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/psych_16.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/psych_44.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/psych_8.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/residue_16.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/residue_44.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/residue_44p51.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/residue_44u.h (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/residue_8.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_11.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_16.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_22.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_32.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_44.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_44p51.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_44u.h (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_8.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/modes/setup_X.h (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/os.h (93%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/psy.c (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/psy.h (98%) create mode 100644 code/libvorbis-1.3.7/lib/psytune.c rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/registry.c (96%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/registry.h (94%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/res0.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/scales.h (97%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/sharedbook.c (98%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/smallft.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/smallft.h (94%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/synthesis.c (98%) create mode 100644 code/libvorbis-1.3.7/lib/tone.c rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/vorbisenc.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/vorbisfile.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/window.c (99%) rename code/{libvorbis-1.3.6 => libvorbis-1.3.7}/lib/window.h (94%) delete mode 100755 code/opus-1.2.1/celt/arm/arm2gnu.pl delete mode 100644 code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm-gnu.S delete mode 100644 code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm.s delete mode 100644 code/opus-1.2.1/celt/x86/x86cpu.h delete mode 100644 code/opus-1.2.1/silk/x86/VQ_WMat_EC_sse.c delete mode 100644 code/opus-1.2.1/silk/x86/main_sse.h delete mode 100644 code/opus-1.2.1/src/mlp.c delete mode 100644 code/opus-1.2.1/src/mlp_data.c delete mode 100644 code/opus-1.2.1/src/tansig_table.h rename code/{opus-1.2.1 => opus-1.5.2}/celt/_kiss_fft_guts.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arch.h (88%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/arm_celt_map.c (85%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/armcpu.c (64%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/armcpu.h (87%) rename code/{opus-1.2.1/celt/arm/celt_ne10_fft.c => opus-1.5.2/celt/arm/celt_fft_ne10.c} (99%) rename code/{opus-1.2.1/celt/arm/celt_ne10_mdct.c => opus-1.5.2/celt/arm/celt_mdct_ne10.c} (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/celt_neon_intr.c (71%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/fft_arm.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/fixed_arm64.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/fixed_armv4.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/fixed_armv5e.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/kiss_fft_armv4.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/kiss_fft_armv5e.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/mdct_arm.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/pitch_arm.h (95%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/arm/pitch_neon_intr.c (88%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/bands.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/bands.h (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt.h (91%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt_decoder.c (72%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt_encoder.c (92%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt_lpc.c (71%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/celt_lpc.h (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/cpu_support.h (88%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/cwrs.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/cwrs.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/ecintrin.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entcode.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entcode.h (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entdec.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entdec.h (89%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entenc.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/entenc.h (90%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/fixed_debug.h (91%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/fixed_generic.h (91%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/float_cast.h (89%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/kiss_fft.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/kiss_fft.h (93%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/laplace.c (64%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/laplace.h (91%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mathops.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mathops.h (97%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mdct.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mdct.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mfrngcod.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/celt_mipsr1.h (97%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/fixed_generic_mipsr1.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/kiss_fft_mipsr1.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/mdct_mipsr1.h (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/pitch_mipsr1.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/mips/vq_mipsr1.h (90%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/modes.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/modes.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/opus_custom_demo.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/os_support.h (85%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/pitch.c (91%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/pitch.h (94%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/quant_bands.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/quant_bands.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/rate.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/rate.h (95%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/stack_alloc.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/static_modes_fixed.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/static_modes_fixed_arm_ne10.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/static_modes_float.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/static_modes_float_arm_ne10.h (100%) create mode 100644 code/opus-1.5.2/celt/tests/test_unit_cwrs32.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_dft.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_entropy.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_laplace.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_mathops.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_mdct.c create mode 100644 code/opus-1.5.2/celt/tests/test_unit_rotation.c rename code/{opus-1.2.1/celt/arm/armopts.s.in => opus-1.5.2/celt/tests/test_unit_types.c} (74%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/vq.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/vq.h (98%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/celt_lpc_sse.h (97%) rename code/{opus-1.2.1/celt/x86/celt_lpc_sse.c => opus-1.5.2/celt/x86/celt_lpc_sse4_1.c} (90%) create mode 100644 code/opus-1.5.2/celt/x86/pitch_avx.c rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/pitch_sse.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/pitch_sse.h (81%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/pitch_sse2.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/pitch_sse4_1.c (82%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/vq_sse.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/vq_sse2.c (96%) rename code/{opus-1.2.1/src/mlp.h => opus-1.5.2/celt/x86/x86_arch_macros.h} (80%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/x86_celt_map.c (92%) rename code/{opus-1.2.1 => opus-1.5.2}/celt/x86/x86cpu.c (68%) create mode 100644 code/opus-1.5.2/celt/x86/x86cpu.h rename code/{opus-1.2.1 => opus-1.5.2}/include/opus.h (87%) rename code/{opus-1.2.1 => opus-1.5.2}/include/opus_custom.h (98%) rename code/{opus-1.2.1 => opus-1.5.2}/include/opus_defines.h (93%) rename code/{opus-1.2.1 => opus-1.5.2}/include/opus_multistream.h (99%) create mode 100644 code/opus-1.5.2/include/opus_projection.h rename code/{opus-1.2.1 => opus-1.5.2}/include/opus_types.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/A2NLSF.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/API.h (86%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/CNG.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/HP_variable_cutoff.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/Inlines.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/LPC_analysis_filter.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/LPC_fit.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/LPC_inv_pred_gain.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/LP_variable_cutoff.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/MacroCount.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/MacroDebug.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF2A.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_VQ.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_VQ_weights_laroia.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_decode.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_del_dec_quant.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_encode.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_stabilize.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NLSF_unpack.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NSQ.c (92%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NSQ.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/NSQ_del_dec.c (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/PLC.c (90%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/PLC.h (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/SigProc_FIX.h (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/VAD.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/VQ_WMat_EC.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/ana_filt_bank_1.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/LPC_inv_pred_gain_arm.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/LPC_inv_pred_gain_neon_intr.c (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/NSQ_del_dec_arm.h (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/NSQ_del_dec_neon_intr.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/NSQ_neon.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/NSQ_neon.h (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/SigProc_FIX_armv4.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/SigProc_FIX_armv5e.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/arm_silk_map.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/biquad_alt_arm.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/biquad_alt_neon_intr.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/macros_arm64.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/macros_armv4.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/arm/macros_armv5e.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/biquad_alt.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/bwexpander.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/bwexpander_32.c (94%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/check_control_input.c (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/code_signs.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/control.h (95%) create mode 100644 code/opus-1.5.2/silk/control_SNR.c rename code/{opus-1.2.1 => opus-1.5.2}/silk/control_audio_bandwidth.c (91%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/control_codec.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/debug.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/debug.h (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/dec_API.c (90%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_core.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_frame.c (73%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_indices.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_parameters.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_pitch.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decode_pulses.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/decoder_set_fs.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/define.h (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/enc_API.c (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/encode_indices.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/encode_pulses.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/errors.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/LTP_analysis_filter_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/LTP_scale_ctrl_FIX.c (78%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/apply_sine_window_FIX.c (96%) create mode 100644 code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_arm.h create mode 100644 code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/autocorr_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/burg_modified_FIX.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/corrMatrix_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/encode_frame_FIX.c (94%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/find_LPC_FIX.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/find_LTP_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/find_pitch_lags_FIX.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/find_pred_coefs_FIX.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/k2a_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/k2a_Q16_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/main_FIX.h (98%) create mode 100644 code/opus-1.5.2/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h create mode 100644 code/opus-1.5.2/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/noise_shape_analysis_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/pitch_analysis_core_FIX.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/process_gains_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/regularize_correlations_FIX.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/residual_energy16_FIX.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/residual_energy_FIX.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/schur64_FIX.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/schur_FIX.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/structs_FIX.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/vector_ops_FIX.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/fixed/warped_autocorrelation_FIX.c (96%) create mode 100644 code/opus-1.5.2/silk/fixed/x86/burg_modified_FIX_sse4_1.c create mode 100644 code/opus-1.5.2/silk/fixed/x86/vector_ops_FIX_sse4_1.c rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/LPC_analysis_filter_FLP.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/LPC_inv_pred_gain_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/LTP_analysis_filter_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/LTP_scale_ctrl_FLP.c (78%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/SigProc_FLP.h (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/apply_sine_window_FLP.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/autocorrelation_FLP.c (92%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/burg_modified_FLP.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/bwexpander_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/corrMatrix_FLP.c (94%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/encode_frame_FLP.c (94%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/energy_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/find_LPC_FLP.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/find_LTP_FLP.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/find_pitch_lags_FLP.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/find_pred_coefs_FLP.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/inner_product_FLP.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/k2a_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/main_FLP.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/noise_shape_analysis_FLP.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/pitch_analysis_core_FLP.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/process_gains_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/regularize_correlations_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/residual_energy_FLP.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/scale_copy_vector_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/scale_vector_FLP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/schur_FLP.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/sort_FLP.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/structs_FLP.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/warped_autocorrelation_FLP.c (86%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/float/wrappers_FLP.c (98%) rename code/{opus-1.2.1/silk/control_SNR.c => opus-1.5.2/silk/float/x86/inner_product_FLP_avx2.c} (50%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/gain_quant.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/init_decoder.c (77%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/init_encoder.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/inner_prod_aligned.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/interpolate.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/lin2log.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/log2lin.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/macros.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/main.h (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/mips/NSQ_del_dec_mipsr1.h (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/mips/macros_mipsr1.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/mips/sigproc_fix_mipsr1.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/pitch_est_defines.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/pitch_est_tables.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/process_NLSFs.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/quant_LTP_gains.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler.c (98%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_down2.c (97%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_down2_3.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_private.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_private_AR2.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_private_IIR_FIR.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_private_down_FIR.c (99%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_private_up2_HQ.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_rom.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_rom.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/resampler_structs.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/shell_coder.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/sigm_Q15.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/sort.c (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_LR_to_MS.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_MS_to_LR.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_decode_pred.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_encode_pred.c (95%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_find_predictor.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/stereo_quant_pred.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/structs.h (96%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/sum_sqr_shift.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/table_LSF_cos.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables.h (93%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_LTP.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_NLSF_CB_NB_MB.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_NLSF_CB_WB.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_gain.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_other.c (87%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_pitch_lag.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/tables_pulses_per_block.c (100%) create mode 100644 code/opus-1.5.2/silk/tests/test_unit_LPC_inv_pred_gain.c rename code/{opus-1.2.1 => opus-1.5.2}/silk/tuning_parameters.h (100%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/typedef.h (97%) create mode 100644 code/opus-1.5.2/silk/x86/NSQ_del_dec_avx2.c rename code/{opus-1.2.1/silk/x86/NSQ_del_dec_sse.c => opus-1.5.2/silk/x86/NSQ_del_dec_sse4_1.c} (86%) rename code/{opus-1.2.1/silk/x86/NSQ_sse.c => opus-1.5.2/silk/x86/NSQ_sse4_1.c} (77%) rename code/{opus-1.2.1 => opus-1.5.2}/silk/x86/SigProc_FIX_sse.h (75%) rename code/{opus-1.2.1/silk/x86/VAD_sse.c => opus-1.5.2/silk/x86/VAD_sse4_1.c} (92%) create mode 100644 code/opus-1.5.2/silk/x86/VQ_WMat_EC_sse4_1.c create mode 100644 code/opus-1.5.2/silk/x86/main_sse.h rename code/{opus-1.2.1 => opus-1.5.2}/silk/x86/x86_silk_map.c (64%) rename code/{opus-1.2.1 => opus-1.5.2}/src/analysis.c (77%) rename code/{opus-1.2.1 => opus-1.5.2}/src/analysis.h (86%) create mode 100644 code/opus-1.5.2/src/extensions.c create mode 100644 code/opus-1.5.2/src/mapping_matrix.c create mode 100644 code/opus-1.5.2/src/mapping_matrix.h create mode 100644 code/opus-1.5.2/src/mlp.c create mode 100644 code/opus-1.5.2/src/mlp.h create mode 100644 code/opus-1.5.2/src/mlp_data.c rename code/{opus-1.2.1 => opus-1.5.2}/src/opus.c (97%) create mode 100644 code/opus-1.5.2/src/opus_compare.c rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_decoder.c (62%) create mode 100644 code/opus-1.5.2/src/opus_demo.c rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_encoder.c (78%) rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_multistream.c (100%) rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_multistream_decoder.c (91%) rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_multistream_encoder.c (90%) rename code/{opus-1.2.1 => opus-1.5.2}/src/opus_private.h (63%) create mode 100644 code/opus-1.5.2/src/opus_projection_decoder.c create mode 100644 code/opus-1.5.2/src/opus_projection_encoder.c rename code/{opus-1.2.1 => opus-1.5.2}/src/repacketizer.c (68%) create mode 100644 code/opus-1.5.2/src/repacketizer_demo.c rename code/{opusfile-0.9 => opusfile-0.12}/include/opusfile.h (99%) rename code/{opusfile-0.9 => opusfile-0.12}/src/http.c (97%) rename code/{opusfile-0.9 => opusfile-0.12}/src/info.c (98%) rename code/{opusfile-0.9 => opusfile-0.12}/src/internal.c (90%) rename code/{opusfile-0.9 => opusfile-0.12}/src/internal.h (98%) rename code/{opusfile-0.9 => opusfile-0.12}/src/opusfile.c (98%) rename code/{opusfile-0.9 => opusfile-0.12}/src/stream.c (81%) rename code/{opusfile-0.9 => opusfile-0.12}/src/wincerts.c (96%) rename code/{opusfile-0.9 => opusfile-0.12}/src/winerrno.h (96%) diff --git a/Makefile b/Makefile index 1dc0abd2..de2e1eec 100644 --- a/Makefile +++ b/Makefile @@ -275,10 +275,10 @@ NDIR=$(MOUNT_DIR)/null UIDIR=$(MOUNT_DIR)/ui Q3UIDIR=$(MOUNT_DIR)/q3_ui JPDIR=$(MOUNT_DIR)/jpeg-8c -OGGDIR=$(MOUNT_DIR)/libogg-1.3.3 -VORBISDIR=$(MOUNT_DIR)/libvorbis-1.3.6 -OPUSDIR=$(MOUNT_DIR)/opus-1.2.1 -OPUSFILEDIR=$(MOUNT_DIR)/opusfile-0.9 +OGGDIR=$(MOUNT_DIR)/libogg-1.3.6 +VORBISDIR=$(MOUNT_DIR)/libvorbis-1.3.7 +OPUSDIR=$(MOUNT_DIR)/opus-1.5.2 +OPUSFILEDIR=$(MOUNT_DIR)/opusfile-0.12 ZDIR=$(MOUNT_DIR)/zlib TOOLSDIR=$(MOUNT_DIR)/tools Q3ASMDIR=$(MOUNT_DIR)/tools/asm @@ -2194,6 +2194,7 @@ Q3OBJ += \ $(B)/client/opus/opus_multistream_encoder.o \ $(B)/client/opus/opus_multistream_decoder.o \ $(B)/client/opus/repacketizer.o \ + $(B)/client/opus/extensions.o \ \ $(B)/client/opus/bands.o \ $(B)/client/opus/celt.o \ diff --git a/code/libogg-1.3.3/include/ogg/config_types.h b/code/libogg-1.3.6/include/ogg/config_types.h similarity index 74% rename from code/libogg-1.3.3/include/ogg/config_types.h rename to code/libogg-1.3.6/include/ogg/config_types.h index 28288279..1a87df64 100644 --- a/code/libogg-1.3.3/include/ogg/config_types.h +++ b/code/libogg-1.3.6/include/ogg/config_types.h @@ -1,10 +1,10 @@ #ifndef __CONFIG_TYPES_H__ #define __CONFIG_TYPES_H__ -/* these are filled in by configure */ -/* #define INCLUDE_INTTYPES_H 1 */ +/* these are filled in by configure or cmake*/ +#define INCLUDE_INTTYPES_H 1 #define INCLUDE_STDINT_H 1 -/* #define INCLUDE_SYS_TYPES_H 1 */ +#define INCLUDE_SYS_TYPES_H 1 #if INCLUDE_INTTYPES_H # include @@ -21,5 +21,6 @@ typedef uint16_t ogg_uint16_t; typedef int32_t ogg_int32_t; typedef uint32_t ogg_uint32_t; typedef int64_t ogg_int64_t; +typedef uint64_t ogg_uint64_t; #endif diff --git a/code/libogg-1.3.3/include/ogg/ogg.h b/code/libogg-1.3.6/include/ogg/ogg.h similarity index 99% rename from code/libogg-1.3.3/include/ogg/ogg.h rename to code/libogg-1.3.6/include/ogg/ogg.h index 7609fc24..c4325aa7 100644 --- a/code/libogg-1.3.3/include/ogg/ogg.h +++ b/code/libogg-1.3.6/include/ogg/ogg.h @@ -11,7 +11,6 @@ ******************************************************************** function: toplevel libogg include - last mod: $Id$ ********************************************************************/ #ifndef _OGG_H diff --git a/code/libogg-1.3.3/include/ogg/os_types.h b/code/libogg-1.3.6/include/ogg/os_types.h similarity index 87% rename from code/libogg-1.3.3/include/ogg/os_types.h rename to code/libogg-1.3.6/include/ogg/os_types.h index b8f56308..e655a1d6 100644 --- a/code/libogg-1.3.3/include/ogg/os_types.h +++ b/code/libogg-1.3.6/include/ogg/os_types.h @@ -10,8 +10,7 @@ * * ******************************************************************** - function: #ifdef jail to whip a few platforms into the UNIX ideal. - last mod: $Id$ + function: Define a consistent set of types on each platform. ********************************************************************/ #ifndef _OS_TYPES_H @@ -44,6 +43,7 @@ typedef unsigned long long ogg_uint64_t; # elif defined(__MWERKS__) typedef long long ogg_int64_t; + typedef unsigned long long ogg_uint64_t; typedef int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef short ogg_int16_t; @@ -62,6 +62,7 @@ typedef __int64 ogg_int64_t; typedef __int32 ogg_int32_t; typedef unsigned __int32 ogg_uint32_t; + typedef unsigned __int64 ogg_uint64_t; typedef __int16 ogg_int16_t; typedef unsigned __int16 ogg_uint16_t; # endif @@ -69,12 +70,13 @@ #elif (defined(__APPLE__) && defined(__MACH__)) /* MacOS X Framework build */ -# include +# include typedef int16_t ogg_int16_t; - typedef uint16_t ogg_uint16_t; + typedef u_int16_t ogg_uint16_t; typedef int32_t ogg_int32_t; - typedef uint32_t ogg_uint32_t; + typedef u_int32_t ogg_uint32_t; typedef int64_t ogg_int64_t; + typedef u_int64_t ogg_uint64_t; #elif defined(__HAIKU__) @@ -85,6 +87,7 @@ typedef int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef long long ogg_int64_t; + typedef unsigned long long ogg_uint64_t; #elif defined(__BEOS__) @@ -95,6 +98,7 @@ typedef int32_t ogg_int32_t; typedef uint32_t ogg_uint32_t; typedef int64_t ogg_int64_t; + typedef uint64_t ogg_uint64_t; #elif defined (__EMX__) @@ -104,6 +108,8 @@ typedef int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef long long ogg_int64_t; + typedef unsigned long long ogg_uint64_t; + #elif defined (DJGPP) @@ -112,11 +118,13 @@ typedef int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef long long ogg_int64_t; + typedef unsigned long long ogg_uint64_t; #elif defined(R5900) /* PS2 EE */ typedef long ogg_int64_t; + typedef unsigned long ogg_uint64_t; typedef int ogg_int32_t; typedef unsigned ogg_uint32_t; typedef short ogg_int16_t; @@ -129,6 +137,7 @@ typedef signed int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef long long int ogg_int64_t; + typedef unsigned long long int ogg_uint64_t; #elif defined(__TMS320C6X__) @@ -138,6 +147,7 @@ typedef signed int ogg_int32_t; typedef unsigned int ogg_uint32_t; typedef long long int ogg_int64_t; + typedef unsigned long long int ogg_uint64_t; #else diff --git a/code/libogg-1.3.3/src/bitwise.c b/code/libogg-1.3.6/src/bitwise.c similarity index 94% rename from code/libogg-1.3.3/src/bitwise.c rename to code/libogg-1.3.6/src/bitwise.c index fa2b5720..75b1f0ed 100644 --- a/code/libogg-1.3.3/src/bitwise.c +++ b/code/libogg-1.3.6/src/bitwise.c @@ -11,7 +11,6 @@ ******************************************************************** function: packing variable sized words into an octet stream - last mod: $Id$ ********************************************************************/ @@ -40,6 +39,8 @@ static const unsigned int mask8B[]= void oggpack_writeinit(oggpack_buffer *b){ memset(b,0,sizeof(*b)); b->ptr=b->buffer=_ogg_malloc(BUFFER_INCREMENT); + if (!b->buffer) + return; b->buffer[0]='\0'; b->storage=BUFFER_INCREMENT; } @@ -285,13 +286,13 @@ long oggpack_look(oggpack_buffer *b,int bits){ ret=b->ptr[0]>>b->endbit; if(bits>8){ - ret|=b->ptr[1]<<(8-b->endbit); + ret|=(unsigned long)b->ptr[1]<<(8-b->endbit); if(bits>16){ - ret|=b->ptr[2]<<(16-b->endbit); + ret|=(unsigned long)b->ptr[2]<<(16-b->endbit); if(bits>24){ - ret|=b->ptr[3]<<(24-b->endbit); + ret|=(unsigned long)b->ptr[3]<<(24-b->endbit); if(bits>32 && b->endbit) - ret|=b->ptr[4]<<(32-b->endbit); + ret|=(unsigned long)b->ptr[4]<<(32-b->endbit); } } } @@ -314,13 +315,13 @@ long oggpackB_look(oggpack_buffer *b,int bits){ else if(!bits)return(0L); } - ret=b->ptr[0]<<(24+b->endbit); + ret=(unsigned long)b->ptr[0]<<(24+b->endbit); if(bits>8){ - ret|=b->ptr[1]<<(16+b->endbit); + ret|=(unsigned long)b->ptr[1]<<(16+b->endbit); if(bits>16){ - ret|=b->ptr[2]<<(8+b->endbit); + ret|=(unsigned long)b->ptr[2]<<(8+b->endbit); if(bits>24){ - ret|=b->ptr[3]<<(b->endbit); + ret|=(unsigned long)b->ptr[3]<<(b->endbit); if(bits>32 && b->endbit) ret|=b->ptr[4]>>(8-b->endbit); } @@ -390,13 +391,13 @@ long oggpack_read(oggpack_buffer *b,int bits){ ret=b->ptr[0]>>b->endbit; if(bits>8){ - ret|=b->ptr[1]<<(8-b->endbit); + ret|=(unsigned long)b->ptr[1]<<(8-b->endbit); if(bits>16){ - ret|=b->ptr[2]<<(16-b->endbit); + ret|=(unsigned long)b->ptr[2]<<(16-b->endbit); if(bits>24){ - ret|=b->ptr[3]<<(24-b->endbit); + ret|=(unsigned long)b->ptr[3]<<(24-b->endbit); if(bits>32 && b->endbit){ - ret|=b->ptr[4]<<(32-b->endbit); + ret|=(unsigned long)b->ptr[4]<<(32-b->endbit); } } } @@ -431,13 +432,13 @@ long oggpackB_read(oggpack_buffer *b,int bits){ else if(!bits)return(0L); } - ret=b->ptr[0]<<(24+b->endbit); + ret=(unsigned long)b->ptr[0]<<(24+b->endbit); if(bits>8){ - ret|=b->ptr[1]<<(16+b->endbit); + ret|=(unsigned long)b->ptr[1]<<(16+b->endbit); if(bits>16){ - ret|=b->ptr[2]<<(8+b->endbit); + ret|=(unsigned long)b->ptr[2]<<(8+b->endbit); if(bits>24){ - ret|=b->ptr[3]<<(b->endbit); + ret|=(unsigned long)b->ptr[3]<<(b->endbit); if(bits>32 && b->endbit) ret|=b->ptr[4]>>(8-b->endbit); } @@ -566,17 +567,17 @@ void cliptest(unsigned long *b,int vals,int bits,int *comp,int compsize){ int tbit=bits?bits:ilog(b[i]); if(oggpack_look(&r,tbit)==-1) report("out of data!\n"); - if(oggpack_look(&r,tbit)!=(b[i]&mask[tbit])) + if((unsigned long)oggpack_look(&r,tbit)!=(b[i]&mask[tbit])) report("looked at incorrect value!\n"); if(tbit==1) - if(oggpack_look1(&r)!=(b[i]&mask[tbit])) + if((unsigned long)oggpack_look1(&r)!=(b[i]&mask[tbit])) report("looked at single bit incorrect value!\n"); if(tbit==1){ - if(oggpack_read1(&r)!=(b[i]&mask[tbit])) + if((unsigned long)oggpack_read1(&r)!=(b[i]&mask[tbit])) report("read incorrect single bit value!\n"); }else{ - if(oggpack_read(&r,tbit)!=(b[i]&mask[tbit])) - report("read incorrect value!\n"); + if((unsigned long)oggpack_read(&r,tbit)!=(b[i]&mask[tbit])) + report("read incorrect value!\n"); } } if(oggpack_bytes(&r)!=bytes)report("leftover bytes after read!\n"); @@ -601,16 +602,16 @@ void cliptestB(unsigned long *b,int vals,int bits,int *comp,int compsize){ int tbit=bits?bits:ilog(b[i]); if(oggpackB_look(&r,tbit)==-1) report("out of data!\n"); - if(oggpackB_look(&r,tbit)!=(b[i]&mask[tbit])) + if((unsigned long)oggpackB_look(&r,tbit)!=(b[i]&mask[tbit])) report("looked at incorrect value!\n"); if(tbit==1) - if(oggpackB_look1(&r)!=(b[i]&mask[tbit])) + if((unsigned long)oggpackB_look1(&r)!=(b[i]&mask[tbit])) report("looked at single bit incorrect value!\n"); if(tbit==1){ - if(oggpackB_read1(&r)!=(b[i]&mask[tbit])) + if((unsigned long)oggpackB_read1(&r)!=(b[i]&mask[tbit])) report("read incorrect single bit value!\n"); }else{ - if(oggpackB_read(&r,tbit)!=(b[i]&mask[tbit])) + if((unsigned long)oggpackB_read(&r,tbit)!=(b[i]&mask[tbit])) report("read incorrect value!\n"); } } @@ -889,8 +890,8 @@ int main(void){ oggpack_readinit(&r,buffer,bytes); for(i=0;i + +static const ogg_uint32_t crc_lookup[8][256]={ +{0x00000000,0x04c11db7,0x09823b6e,0x0d4326d9,0x130476dc,0x17c56b6b,0x1a864db2,0x1e475005, + 0x2608edb8,0x22c9f00f,0x2f8ad6d6,0x2b4bcb61,0x350c9b64,0x31cd86d3,0x3c8ea00a,0x384fbdbd, + 0x4c11db70,0x48d0c6c7,0x4593e01e,0x4152fda9,0x5f15adac,0x5bd4b01b,0x569796c2,0x52568b75, + 0x6a1936c8,0x6ed82b7f,0x639b0da6,0x675a1011,0x791d4014,0x7ddc5da3,0x709f7b7a,0x745e66cd, + 0x9823b6e0,0x9ce2ab57,0x91a18d8e,0x95609039,0x8b27c03c,0x8fe6dd8b,0x82a5fb52,0x8664e6e5, + 0xbe2b5b58,0xbaea46ef,0xb7a96036,0xb3687d81,0xad2f2d84,0xa9ee3033,0xa4ad16ea,0xa06c0b5d, + 0xd4326d90,0xd0f37027,0xddb056fe,0xd9714b49,0xc7361b4c,0xc3f706fb,0xceb42022,0xca753d95, + 0xf23a8028,0xf6fb9d9f,0xfbb8bb46,0xff79a6f1,0xe13ef6f4,0xe5ffeb43,0xe8bccd9a,0xec7dd02d, + 0x34867077,0x30476dc0,0x3d044b19,0x39c556ae,0x278206ab,0x23431b1c,0x2e003dc5,0x2ac12072, + 0x128e9dcf,0x164f8078,0x1b0ca6a1,0x1fcdbb16,0x018aeb13,0x054bf6a4,0x0808d07d,0x0cc9cdca, + 0x7897ab07,0x7c56b6b0,0x71159069,0x75d48dde,0x6b93dddb,0x6f52c06c,0x6211e6b5,0x66d0fb02, + 0x5e9f46bf,0x5a5e5b08,0x571d7dd1,0x53dc6066,0x4d9b3063,0x495a2dd4,0x44190b0d,0x40d816ba, + 0xaca5c697,0xa864db20,0xa527fdf9,0xa1e6e04e,0xbfa1b04b,0xbb60adfc,0xb6238b25,0xb2e29692, + 0x8aad2b2f,0x8e6c3698,0x832f1041,0x87ee0df6,0x99a95df3,0x9d684044,0x902b669d,0x94ea7b2a, + 0xe0b41de7,0xe4750050,0xe9362689,0xedf73b3e,0xf3b06b3b,0xf771768c,0xfa325055,0xfef34de2, + 0xc6bcf05f,0xc27dede8,0xcf3ecb31,0xcbffd686,0xd5b88683,0xd1799b34,0xdc3abded,0xd8fba05a, + 0x690ce0ee,0x6dcdfd59,0x608edb80,0x644fc637,0x7a089632,0x7ec98b85,0x738aad5c,0x774bb0eb, + 0x4f040d56,0x4bc510e1,0x46863638,0x42472b8f,0x5c007b8a,0x58c1663d,0x558240e4,0x51435d53, + 0x251d3b9e,0x21dc2629,0x2c9f00f0,0x285e1d47,0x36194d42,0x32d850f5,0x3f9b762c,0x3b5a6b9b, + 0x0315d626,0x07d4cb91,0x0a97ed48,0x0e56f0ff,0x1011a0fa,0x14d0bd4d,0x19939b94,0x1d528623, + 0xf12f560e,0xf5ee4bb9,0xf8ad6d60,0xfc6c70d7,0xe22b20d2,0xe6ea3d65,0xeba91bbc,0xef68060b, + 0xd727bbb6,0xd3e6a601,0xdea580d8,0xda649d6f,0xc423cd6a,0xc0e2d0dd,0xcda1f604,0xc960ebb3, + 0xbd3e8d7e,0xb9ff90c9,0xb4bcb610,0xb07daba7,0xae3afba2,0xaafbe615,0xa7b8c0cc,0xa379dd7b, + 0x9b3660c6,0x9ff77d71,0x92b45ba8,0x9675461f,0x8832161a,0x8cf30bad,0x81b02d74,0x857130c3, + 0x5d8a9099,0x594b8d2e,0x5408abf7,0x50c9b640,0x4e8ee645,0x4a4ffbf2,0x470cdd2b,0x43cdc09c, + 0x7b827d21,0x7f436096,0x7200464f,0x76c15bf8,0x68860bfd,0x6c47164a,0x61043093,0x65c52d24, + 0x119b4be9,0x155a565e,0x18197087,0x1cd86d30,0x029f3d35,0x065e2082,0x0b1d065b,0x0fdc1bec, + 0x3793a651,0x3352bbe6,0x3e119d3f,0x3ad08088,0x2497d08d,0x2056cd3a,0x2d15ebe3,0x29d4f654, + 0xc5a92679,0xc1683bce,0xcc2b1d17,0xc8ea00a0,0xd6ad50a5,0xd26c4d12,0xdf2f6bcb,0xdbee767c, + 0xe3a1cbc1,0xe760d676,0xea23f0af,0xeee2ed18,0xf0a5bd1d,0xf464a0aa,0xf9278673,0xfde69bc4, + 0x89b8fd09,0x8d79e0be,0x803ac667,0x84fbdbd0,0x9abc8bd5,0x9e7d9662,0x933eb0bb,0x97ffad0c, + 0xafb010b1,0xab710d06,0xa6322bdf,0xa2f33668,0xbcb4666d,0xb8757bda,0xb5365d03,0xb1f740b4}, + +{0x00000000,0xd219c1dc,0xa0f29e0f,0x72eb5fd3,0x452421a9,0x973de075,0xe5d6bfa6,0x37cf7e7a, + 0x8a484352,0x5851828e,0x2abadd5d,0xf8a31c81,0xcf6c62fb,0x1d75a327,0x6f9efcf4,0xbd873d28, + 0x10519b13,0xc2485acf,0xb0a3051c,0x62bac4c0,0x5575baba,0x876c7b66,0xf58724b5,0x279ee569, + 0x9a19d841,0x4800199d,0x3aeb464e,0xe8f28792,0xdf3df9e8,0x0d243834,0x7fcf67e7,0xadd6a63b, + 0x20a33626,0xf2baf7fa,0x8051a829,0x524869f5,0x6587178f,0xb79ed653,0xc5758980,0x176c485c, + 0xaaeb7574,0x78f2b4a8,0x0a19eb7b,0xd8002aa7,0xefcf54dd,0x3dd69501,0x4f3dcad2,0x9d240b0e, + 0x30f2ad35,0xe2eb6ce9,0x9000333a,0x4219f2e6,0x75d68c9c,0xa7cf4d40,0xd5241293,0x073dd34f, + 0xbabaee67,0x68a32fbb,0x1a487068,0xc851b1b4,0xff9ecfce,0x2d870e12,0x5f6c51c1,0x8d75901d, + 0x41466c4c,0x935fad90,0xe1b4f243,0x33ad339f,0x04624de5,0xd67b8c39,0xa490d3ea,0x76891236, + 0xcb0e2f1e,0x1917eec2,0x6bfcb111,0xb9e570cd,0x8e2a0eb7,0x5c33cf6b,0x2ed890b8,0xfcc15164, + 0x5117f75f,0x830e3683,0xf1e56950,0x23fca88c,0x1433d6f6,0xc62a172a,0xb4c148f9,0x66d88925, + 0xdb5fb40d,0x094675d1,0x7bad2a02,0xa9b4ebde,0x9e7b95a4,0x4c625478,0x3e890bab,0xec90ca77, + 0x61e55a6a,0xb3fc9bb6,0xc117c465,0x130e05b9,0x24c17bc3,0xf6d8ba1f,0x8433e5cc,0x562a2410, + 0xebad1938,0x39b4d8e4,0x4b5f8737,0x994646eb,0xae893891,0x7c90f94d,0x0e7ba69e,0xdc626742, + 0x71b4c179,0xa3ad00a5,0xd1465f76,0x035f9eaa,0x3490e0d0,0xe689210c,0x94627edf,0x467bbf03, + 0xfbfc822b,0x29e543f7,0x5b0e1c24,0x8917ddf8,0xbed8a382,0x6cc1625e,0x1e2a3d8d,0xcc33fc51, + 0x828cd898,0x50951944,0x227e4697,0xf067874b,0xc7a8f931,0x15b138ed,0x675a673e,0xb543a6e2, + 0x08c49bca,0xdadd5a16,0xa83605c5,0x7a2fc419,0x4de0ba63,0x9ff97bbf,0xed12246c,0x3f0be5b0, + 0x92dd438b,0x40c48257,0x322fdd84,0xe0361c58,0xd7f96222,0x05e0a3fe,0x770bfc2d,0xa5123df1, + 0x189500d9,0xca8cc105,0xb8679ed6,0x6a7e5f0a,0x5db12170,0x8fa8e0ac,0xfd43bf7f,0x2f5a7ea3, + 0xa22feebe,0x70362f62,0x02dd70b1,0xd0c4b16d,0xe70bcf17,0x35120ecb,0x47f95118,0x95e090c4, + 0x2867adec,0xfa7e6c30,0x889533e3,0x5a8cf23f,0x6d438c45,0xbf5a4d99,0xcdb1124a,0x1fa8d396, + 0xb27e75ad,0x6067b471,0x128ceba2,0xc0952a7e,0xf75a5404,0x254395d8,0x57a8ca0b,0x85b10bd7, + 0x383636ff,0xea2ff723,0x98c4a8f0,0x4add692c,0x7d121756,0xaf0bd68a,0xdde08959,0x0ff94885, + 0xc3cab4d4,0x11d37508,0x63382adb,0xb121eb07,0x86ee957d,0x54f754a1,0x261c0b72,0xf405caae, + 0x4982f786,0x9b9b365a,0xe9706989,0x3b69a855,0x0ca6d62f,0xdebf17f3,0xac544820,0x7e4d89fc, + 0xd39b2fc7,0x0182ee1b,0x7369b1c8,0xa1707014,0x96bf0e6e,0x44a6cfb2,0x364d9061,0xe45451bd, + 0x59d36c95,0x8bcaad49,0xf921f29a,0x2b383346,0x1cf74d3c,0xceee8ce0,0xbc05d333,0x6e1c12ef, + 0xe36982f2,0x3170432e,0x439b1cfd,0x9182dd21,0xa64da35b,0x74546287,0x06bf3d54,0xd4a6fc88, + 0x6921c1a0,0xbb38007c,0xc9d35faf,0x1bca9e73,0x2c05e009,0xfe1c21d5,0x8cf77e06,0x5eeebfda, + 0xf33819e1,0x2121d83d,0x53ca87ee,0x81d34632,0xb61c3848,0x6405f994,0x16eea647,0xc4f7679b, + 0x79705ab3,0xab699b6f,0xd982c4bc,0x0b9b0560,0x3c547b1a,0xee4dbac6,0x9ca6e515,0x4ebf24c9}, + +{0x00000000,0x01d8ac87,0x03b1590e,0x0269f589,0x0762b21c,0x06ba1e9b,0x04d3eb12,0x050b4795, + 0x0ec56438,0x0f1dc8bf,0x0d743d36,0x0cac91b1,0x09a7d624,0x087f7aa3,0x0a168f2a,0x0bce23ad, + 0x1d8ac870,0x1c5264f7,0x1e3b917e,0x1fe33df9,0x1ae87a6c,0x1b30d6eb,0x19592362,0x18818fe5, + 0x134fac48,0x129700cf,0x10fef546,0x112659c1,0x142d1e54,0x15f5b2d3,0x179c475a,0x1644ebdd, + 0x3b1590e0,0x3acd3c67,0x38a4c9ee,0x397c6569,0x3c7722fc,0x3daf8e7b,0x3fc67bf2,0x3e1ed775, + 0x35d0f4d8,0x3408585f,0x3661add6,0x37b90151,0x32b246c4,0x336aea43,0x31031fca,0x30dbb34d, + 0x269f5890,0x2747f417,0x252e019e,0x24f6ad19,0x21fdea8c,0x2025460b,0x224cb382,0x23941f05, + 0x285a3ca8,0x2982902f,0x2beb65a6,0x2a33c921,0x2f388eb4,0x2ee02233,0x2c89d7ba,0x2d517b3d, + 0x762b21c0,0x77f38d47,0x759a78ce,0x7442d449,0x714993dc,0x70913f5b,0x72f8cad2,0x73206655, + 0x78ee45f8,0x7936e97f,0x7b5f1cf6,0x7a87b071,0x7f8cf7e4,0x7e545b63,0x7c3daeea,0x7de5026d, + 0x6ba1e9b0,0x6a794537,0x6810b0be,0x69c81c39,0x6cc35bac,0x6d1bf72b,0x6f7202a2,0x6eaaae25, + 0x65648d88,0x64bc210f,0x66d5d486,0x670d7801,0x62063f94,0x63de9313,0x61b7669a,0x606fca1d, + 0x4d3eb120,0x4ce61da7,0x4e8fe82e,0x4f5744a9,0x4a5c033c,0x4b84afbb,0x49ed5a32,0x4835f6b5, + 0x43fbd518,0x4223799f,0x404a8c16,0x41922091,0x44996704,0x4541cb83,0x47283e0a,0x46f0928d, + 0x50b47950,0x516cd5d7,0x5305205e,0x52dd8cd9,0x57d6cb4c,0x560e67cb,0x54679242,0x55bf3ec5, + 0x5e711d68,0x5fa9b1ef,0x5dc04466,0x5c18e8e1,0x5913af74,0x58cb03f3,0x5aa2f67a,0x5b7a5afd, + 0xec564380,0xed8eef07,0xefe71a8e,0xee3fb609,0xeb34f19c,0xeaec5d1b,0xe885a892,0xe95d0415, + 0xe29327b8,0xe34b8b3f,0xe1227eb6,0xe0fad231,0xe5f195a4,0xe4293923,0xe640ccaa,0xe798602d, + 0xf1dc8bf0,0xf0042777,0xf26dd2fe,0xf3b57e79,0xf6be39ec,0xf766956b,0xf50f60e2,0xf4d7cc65, + 0xff19efc8,0xfec1434f,0xfca8b6c6,0xfd701a41,0xf87b5dd4,0xf9a3f153,0xfbca04da,0xfa12a85d, + 0xd743d360,0xd69b7fe7,0xd4f28a6e,0xd52a26e9,0xd021617c,0xd1f9cdfb,0xd3903872,0xd24894f5, + 0xd986b758,0xd85e1bdf,0xda37ee56,0xdbef42d1,0xdee40544,0xdf3ca9c3,0xdd555c4a,0xdc8df0cd, + 0xcac91b10,0xcb11b797,0xc978421e,0xc8a0ee99,0xcdaba90c,0xcc73058b,0xce1af002,0xcfc25c85, + 0xc40c7f28,0xc5d4d3af,0xc7bd2626,0xc6658aa1,0xc36ecd34,0xc2b661b3,0xc0df943a,0xc10738bd, + 0x9a7d6240,0x9ba5cec7,0x99cc3b4e,0x981497c9,0x9d1fd05c,0x9cc77cdb,0x9eae8952,0x9f7625d5, + 0x94b80678,0x9560aaff,0x97095f76,0x96d1f3f1,0x93dab464,0x920218e3,0x906bed6a,0x91b341ed, + 0x87f7aa30,0x862f06b7,0x8446f33e,0x859e5fb9,0x8095182c,0x814db4ab,0x83244122,0x82fceda5, + 0x8932ce08,0x88ea628f,0x8a839706,0x8b5b3b81,0x8e507c14,0x8f88d093,0x8de1251a,0x8c39899d, + 0xa168f2a0,0xa0b05e27,0xa2d9abae,0xa3010729,0xa60a40bc,0xa7d2ec3b,0xa5bb19b2,0xa463b535, + 0xafad9698,0xae753a1f,0xac1ccf96,0xadc46311,0xa8cf2484,0xa9178803,0xab7e7d8a,0xaaa6d10d, + 0xbce23ad0,0xbd3a9657,0xbf5363de,0xbe8bcf59,0xbb8088cc,0xba58244b,0xb831d1c2,0xb9e97d45, + 0xb2275ee8,0xb3fff26f,0xb19607e6,0xb04eab61,0xb545ecf4,0xb49d4073,0xb6f4b5fa,0xb72c197d}, + +{0x00000000,0xdc6d9ab7,0xbc1a28d9,0x6077b26e,0x7cf54c05,0xa098d6b2,0xc0ef64dc,0x1c82fe6b, + 0xf9ea980a,0x258702bd,0x45f0b0d3,0x999d2a64,0x851fd40f,0x59724eb8,0x3905fcd6,0xe5686661, + 0xf7142da3,0x2b79b714,0x4b0e057a,0x97639fcd,0x8be161a6,0x578cfb11,0x37fb497f,0xeb96d3c8, + 0x0efeb5a9,0xd2932f1e,0xb2e49d70,0x6e8907c7,0x720bf9ac,0xae66631b,0xce11d175,0x127c4bc2, + 0xeae946f1,0x3684dc46,0x56f36e28,0x8a9ef49f,0x961c0af4,0x4a719043,0x2a06222d,0xf66bb89a, + 0x1303defb,0xcf6e444c,0xaf19f622,0x73746c95,0x6ff692fe,0xb39b0849,0xd3ecba27,0x0f812090, + 0x1dfd6b52,0xc190f1e5,0xa1e7438b,0x7d8ad93c,0x61082757,0xbd65bde0,0xdd120f8e,0x017f9539, + 0xe417f358,0x387a69ef,0x580ddb81,0x84604136,0x98e2bf5d,0x448f25ea,0x24f89784,0xf8950d33, + 0xd1139055,0x0d7e0ae2,0x6d09b88c,0xb164223b,0xade6dc50,0x718b46e7,0x11fcf489,0xcd916e3e, + 0x28f9085f,0xf49492e8,0x94e32086,0x488eba31,0x540c445a,0x8861deed,0xe8166c83,0x347bf634, + 0x2607bdf6,0xfa6a2741,0x9a1d952f,0x46700f98,0x5af2f1f3,0x869f6b44,0xe6e8d92a,0x3a85439d, + 0xdfed25fc,0x0380bf4b,0x63f70d25,0xbf9a9792,0xa31869f9,0x7f75f34e,0x1f024120,0xc36fdb97, + 0x3bfad6a4,0xe7974c13,0x87e0fe7d,0x5b8d64ca,0x470f9aa1,0x9b620016,0xfb15b278,0x277828cf, + 0xc2104eae,0x1e7dd419,0x7e0a6677,0xa267fcc0,0xbee502ab,0x6288981c,0x02ff2a72,0xde92b0c5, + 0xcceefb07,0x108361b0,0x70f4d3de,0xac994969,0xb01bb702,0x6c762db5,0x0c019fdb,0xd06c056c, + 0x3504630d,0xe969f9ba,0x891e4bd4,0x5573d163,0x49f12f08,0x959cb5bf,0xf5eb07d1,0x29869d66, + 0xa6e63d1d,0x7a8ba7aa,0x1afc15c4,0xc6918f73,0xda137118,0x067eebaf,0x660959c1,0xba64c376, + 0x5f0ca517,0x83613fa0,0xe3168dce,0x3f7b1779,0x23f9e912,0xff9473a5,0x9fe3c1cb,0x438e5b7c, + 0x51f210be,0x8d9f8a09,0xede83867,0x3185a2d0,0x2d075cbb,0xf16ac60c,0x911d7462,0x4d70eed5, + 0xa81888b4,0x74751203,0x1402a06d,0xc86f3ada,0xd4edc4b1,0x08805e06,0x68f7ec68,0xb49a76df, + 0x4c0f7bec,0x9062e15b,0xf0155335,0x2c78c982,0x30fa37e9,0xec97ad5e,0x8ce01f30,0x508d8587, + 0xb5e5e3e6,0x69887951,0x09ffcb3f,0xd5925188,0xc910afe3,0x157d3554,0x750a873a,0xa9671d8d, + 0xbb1b564f,0x6776ccf8,0x07017e96,0xdb6ce421,0xc7ee1a4a,0x1b8380fd,0x7bf43293,0xa799a824, + 0x42f1ce45,0x9e9c54f2,0xfeebe69c,0x22867c2b,0x3e048240,0xe26918f7,0x821eaa99,0x5e73302e, + 0x77f5ad48,0xab9837ff,0xcbef8591,0x17821f26,0x0b00e14d,0xd76d7bfa,0xb71ac994,0x6b775323, + 0x8e1f3542,0x5272aff5,0x32051d9b,0xee68872c,0xf2ea7947,0x2e87e3f0,0x4ef0519e,0x929dcb29, + 0x80e180eb,0x5c8c1a5c,0x3cfba832,0xe0963285,0xfc14ccee,0x20795659,0x400ee437,0x9c637e80, + 0x790b18e1,0xa5668256,0xc5113038,0x197caa8f,0x05fe54e4,0xd993ce53,0xb9e47c3d,0x6589e68a, + 0x9d1cebb9,0x4171710e,0x2106c360,0xfd6b59d7,0xe1e9a7bc,0x3d843d0b,0x5df38f65,0x819e15d2, + 0x64f673b3,0xb89be904,0xd8ec5b6a,0x0481c1dd,0x18033fb6,0xc46ea501,0xa419176f,0x78748dd8, + 0x6a08c61a,0xb6655cad,0xd612eec3,0x0a7f7474,0x16fd8a1f,0xca9010a8,0xaae7a2c6,0x768a3871, + 0x93e25e10,0x4f8fc4a7,0x2ff876c9,0xf395ec7e,0xef171215,0x337a88a2,0x530d3acc,0x8f60a07b}, + +{0x00000000,0x490d678d,0x921acf1a,0xdb17a897,0x20f48383,0x69f9e40e,0xb2ee4c99,0xfbe32b14, + 0x41e90706,0x08e4608b,0xd3f3c81c,0x9afeaf91,0x611d8485,0x2810e308,0xf3074b9f,0xba0a2c12, + 0x83d20e0c,0xcadf6981,0x11c8c116,0x58c5a69b,0xa3268d8f,0xea2bea02,0x313c4295,0x78312518, + 0xc23b090a,0x8b366e87,0x5021c610,0x192ca19d,0xe2cf8a89,0xabc2ed04,0x70d54593,0x39d8221e, + 0x036501af,0x4a686622,0x917fceb5,0xd872a938,0x2391822c,0x6a9ce5a1,0xb18b4d36,0xf8862abb, + 0x428c06a9,0x0b816124,0xd096c9b3,0x999bae3e,0x6278852a,0x2b75e2a7,0xf0624a30,0xb96f2dbd, + 0x80b70fa3,0xc9ba682e,0x12adc0b9,0x5ba0a734,0xa0438c20,0xe94eebad,0x3259433a,0x7b5424b7, + 0xc15e08a5,0x88536f28,0x5344c7bf,0x1a49a032,0xe1aa8b26,0xa8a7ecab,0x73b0443c,0x3abd23b1, + 0x06ca035e,0x4fc764d3,0x94d0cc44,0xddddabc9,0x263e80dd,0x6f33e750,0xb4244fc7,0xfd29284a, + 0x47230458,0x0e2e63d5,0xd539cb42,0x9c34accf,0x67d787db,0x2edae056,0xf5cd48c1,0xbcc02f4c, + 0x85180d52,0xcc156adf,0x1702c248,0x5e0fa5c5,0xa5ec8ed1,0xece1e95c,0x37f641cb,0x7efb2646, + 0xc4f10a54,0x8dfc6dd9,0x56ebc54e,0x1fe6a2c3,0xe40589d7,0xad08ee5a,0x761f46cd,0x3f122140, + 0x05af02f1,0x4ca2657c,0x97b5cdeb,0xdeb8aa66,0x255b8172,0x6c56e6ff,0xb7414e68,0xfe4c29e5, + 0x444605f7,0x0d4b627a,0xd65ccaed,0x9f51ad60,0x64b28674,0x2dbfe1f9,0xf6a8496e,0xbfa52ee3, + 0x867d0cfd,0xcf706b70,0x1467c3e7,0x5d6aa46a,0xa6898f7e,0xef84e8f3,0x34934064,0x7d9e27e9, + 0xc7940bfb,0x8e996c76,0x558ec4e1,0x1c83a36c,0xe7608878,0xae6deff5,0x757a4762,0x3c7720ef, + 0x0d9406bc,0x44996131,0x9f8ec9a6,0xd683ae2b,0x2d60853f,0x646de2b2,0xbf7a4a25,0xf6772da8, + 0x4c7d01ba,0x05706637,0xde67cea0,0x976aa92d,0x6c898239,0x2584e5b4,0xfe934d23,0xb79e2aae, + 0x8e4608b0,0xc74b6f3d,0x1c5cc7aa,0x5551a027,0xaeb28b33,0xe7bfecbe,0x3ca84429,0x75a523a4, + 0xcfaf0fb6,0x86a2683b,0x5db5c0ac,0x14b8a721,0xef5b8c35,0xa656ebb8,0x7d41432f,0x344c24a2, + 0x0ef10713,0x47fc609e,0x9cebc809,0xd5e6af84,0x2e058490,0x6708e31d,0xbc1f4b8a,0xf5122c07, + 0x4f180015,0x06156798,0xdd02cf0f,0x940fa882,0x6fec8396,0x26e1e41b,0xfdf64c8c,0xb4fb2b01, + 0x8d23091f,0xc42e6e92,0x1f39c605,0x5634a188,0xadd78a9c,0xe4daed11,0x3fcd4586,0x76c0220b, + 0xccca0e19,0x85c76994,0x5ed0c103,0x17dda68e,0xec3e8d9a,0xa533ea17,0x7e244280,0x3729250d, + 0x0b5e05e2,0x4253626f,0x9944caf8,0xd049ad75,0x2baa8661,0x62a7e1ec,0xb9b0497b,0xf0bd2ef6, + 0x4ab702e4,0x03ba6569,0xd8adcdfe,0x91a0aa73,0x6a438167,0x234ee6ea,0xf8594e7d,0xb15429f0, + 0x888c0bee,0xc1816c63,0x1a96c4f4,0x539ba379,0xa878886d,0xe175efe0,0x3a624777,0x736f20fa, + 0xc9650ce8,0x80686b65,0x5b7fc3f2,0x1272a47f,0xe9918f6b,0xa09ce8e6,0x7b8b4071,0x328627fc, + 0x083b044d,0x413663c0,0x9a21cb57,0xd32cacda,0x28cf87ce,0x61c2e043,0xbad548d4,0xf3d82f59, + 0x49d2034b,0x00df64c6,0xdbc8cc51,0x92c5abdc,0x692680c8,0x202be745,0xfb3c4fd2,0xb231285f, + 0x8be90a41,0xc2e46dcc,0x19f3c55b,0x50fea2d6,0xab1d89c2,0xe210ee4f,0x390746d8,0x700a2155, + 0xca000d47,0x830d6aca,0x581ac25d,0x1117a5d0,0xeaf48ec4,0xa3f9e949,0x78ee41de,0x31e32653}, + +{0x00000000,0x1b280d78,0x36501af0,0x2d781788,0x6ca035e0,0x77883898,0x5af02f10,0x41d82268, + 0xd9406bc0,0xc26866b8,0xef107130,0xf4387c48,0xb5e05e20,0xaec85358,0x83b044d0,0x989849a8, + 0xb641ca37,0xad69c74f,0x8011d0c7,0x9b39ddbf,0xdae1ffd7,0xc1c9f2af,0xecb1e527,0xf799e85f, + 0x6f01a1f7,0x7429ac8f,0x5951bb07,0x4279b67f,0x03a19417,0x1889996f,0x35f18ee7,0x2ed9839f, + 0x684289d9,0x736a84a1,0x5e129329,0x453a9e51,0x04e2bc39,0x1fcab141,0x32b2a6c9,0x299aabb1, + 0xb102e219,0xaa2aef61,0x8752f8e9,0x9c7af591,0xdda2d7f9,0xc68ada81,0xebf2cd09,0xf0dac071, + 0xde0343ee,0xc52b4e96,0xe853591e,0xf37b5466,0xb2a3760e,0xa98b7b76,0x84f36cfe,0x9fdb6186, + 0x0743282e,0x1c6b2556,0x311332de,0x2a3b3fa6,0x6be31dce,0x70cb10b6,0x5db3073e,0x469b0a46, + 0xd08513b2,0xcbad1eca,0xe6d50942,0xfdfd043a,0xbc252652,0xa70d2b2a,0x8a753ca2,0x915d31da, + 0x09c57872,0x12ed750a,0x3f956282,0x24bd6ffa,0x65654d92,0x7e4d40ea,0x53355762,0x481d5a1a, + 0x66c4d985,0x7decd4fd,0x5094c375,0x4bbcce0d,0x0a64ec65,0x114ce11d,0x3c34f695,0x271cfbed, + 0xbf84b245,0xa4acbf3d,0x89d4a8b5,0x92fca5cd,0xd32487a5,0xc80c8add,0xe5749d55,0xfe5c902d, + 0xb8c79a6b,0xa3ef9713,0x8e97809b,0x95bf8de3,0xd467af8b,0xcf4fa2f3,0xe237b57b,0xf91fb803, + 0x6187f1ab,0x7aaffcd3,0x57d7eb5b,0x4cffe623,0x0d27c44b,0x160fc933,0x3b77debb,0x205fd3c3, + 0x0e86505c,0x15ae5d24,0x38d64aac,0x23fe47d4,0x622665bc,0x790e68c4,0x54767f4c,0x4f5e7234, + 0xd7c63b9c,0xccee36e4,0xe196216c,0xfabe2c14,0xbb660e7c,0xa04e0304,0x8d36148c,0x961e19f4, + 0xa5cb3ad3,0xbee337ab,0x939b2023,0x88b32d5b,0xc96b0f33,0xd243024b,0xff3b15c3,0xe41318bb, + 0x7c8b5113,0x67a35c6b,0x4adb4be3,0x51f3469b,0x102b64f3,0x0b03698b,0x267b7e03,0x3d53737b, + 0x138af0e4,0x08a2fd9c,0x25daea14,0x3ef2e76c,0x7f2ac504,0x6402c87c,0x497adff4,0x5252d28c, + 0xcaca9b24,0xd1e2965c,0xfc9a81d4,0xe7b28cac,0xa66aaec4,0xbd42a3bc,0x903ab434,0x8b12b94c, + 0xcd89b30a,0xd6a1be72,0xfbd9a9fa,0xe0f1a482,0xa12986ea,0xba018b92,0x97799c1a,0x8c519162, + 0x14c9d8ca,0x0fe1d5b2,0x2299c23a,0x39b1cf42,0x7869ed2a,0x6341e052,0x4e39f7da,0x5511faa2, + 0x7bc8793d,0x60e07445,0x4d9863cd,0x56b06eb5,0x17684cdd,0x0c4041a5,0x2138562d,0x3a105b55, + 0xa28812fd,0xb9a01f85,0x94d8080d,0x8ff00575,0xce28271d,0xd5002a65,0xf8783ded,0xe3503095, + 0x754e2961,0x6e662419,0x431e3391,0x58363ee9,0x19ee1c81,0x02c611f9,0x2fbe0671,0x34960b09, + 0xac0e42a1,0xb7264fd9,0x9a5e5851,0x81765529,0xc0ae7741,0xdb867a39,0xf6fe6db1,0xedd660c9, + 0xc30fe356,0xd827ee2e,0xf55ff9a6,0xee77f4de,0xafafd6b6,0xb487dbce,0x99ffcc46,0x82d7c13e, + 0x1a4f8896,0x016785ee,0x2c1f9266,0x37379f1e,0x76efbd76,0x6dc7b00e,0x40bfa786,0x5b97aafe, + 0x1d0ca0b8,0x0624adc0,0x2b5cba48,0x3074b730,0x71ac9558,0x6a849820,0x47fc8fa8,0x5cd482d0, + 0xc44ccb78,0xdf64c600,0xf21cd188,0xe934dcf0,0xa8ecfe98,0xb3c4f3e0,0x9ebce468,0x8594e910, + 0xab4d6a8f,0xb06567f7,0x9d1d707f,0x86357d07,0xc7ed5f6f,0xdcc55217,0xf1bd459f,0xea9548e7, + 0x720d014f,0x69250c37,0x445d1bbf,0x5f7516c7,0x1ead34af,0x058539d7,0x28fd2e5f,0x33d52327}, + +{0x00000000,0x4f576811,0x9eaed022,0xd1f9b833,0x399cbdf3,0x76cbd5e2,0xa7326dd1,0xe86505c0, + 0x73397be6,0x3c6e13f7,0xed97abc4,0xa2c0c3d5,0x4aa5c615,0x05f2ae04,0xd40b1637,0x9b5c7e26, + 0xe672f7cc,0xa9259fdd,0x78dc27ee,0x378b4fff,0xdfee4a3f,0x90b9222e,0x41409a1d,0x0e17f20c, + 0x954b8c2a,0xda1ce43b,0x0be55c08,0x44b23419,0xacd731d9,0xe38059c8,0x3279e1fb,0x7d2e89ea, + 0xc824f22f,0x87739a3e,0x568a220d,0x19dd4a1c,0xf1b84fdc,0xbeef27cd,0x6f169ffe,0x2041f7ef, + 0xbb1d89c9,0xf44ae1d8,0x25b359eb,0x6ae431fa,0x8281343a,0xcdd65c2b,0x1c2fe418,0x53788c09, + 0x2e5605e3,0x61016df2,0xb0f8d5c1,0xffafbdd0,0x17cab810,0x589dd001,0x89646832,0xc6330023, + 0x5d6f7e05,0x12381614,0xc3c1ae27,0x8c96c636,0x64f3c3f6,0x2ba4abe7,0xfa5d13d4,0xb50a7bc5, + 0x9488f9e9,0xdbdf91f8,0x0a2629cb,0x457141da,0xad14441a,0xe2432c0b,0x33ba9438,0x7cedfc29, + 0xe7b1820f,0xa8e6ea1e,0x791f522d,0x36483a3c,0xde2d3ffc,0x917a57ed,0x4083efde,0x0fd487cf, + 0x72fa0e25,0x3dad6634,0xec54de07,0xa303b616,0x4b66b3d6,0x0431dbc7,0xd5c863f4,0x9a9f0be5, + 0x01c375c3,0x4e941dd2,0x9f6da5e1,0xd03acdf0,0x385fc830,0x7708a021,0xa6f11812,0xe9a67003, + 0x5cac0bc6,0x13fb63d7,0xc202dbe4,0x8d55b3f5,0x6530b635,0x2a67de24,0xfb9e6617,0xb4c90e06, + 0x2f957020,0x60c21831,0xb13ba002,0xfe6cc813,0x1609cdd3,0x595ea5c2,0x88a71df1,0xc7f075e0, + 0xbadefc0a,0xf589941b,0x24702c28,0x6b274439,0x834241f9,0xcc1529e8,0x1dec91db,0x52bbf9ca, + 0xc9e787ec,0x86b0effd,0x574957ce,0x181e3fdf,0xf07b3a1f,0xbf2c520e,0x6ed5ea3d,0x2182822c, + 0x2dd0ee65,0x62878674,0xb37e3e47,0xfc295656,0x144c5396,0x5b1b3b87,0x8ae283b4,0xc5b5eba5, + 0x5ee99583,0x11befd92,0xc04745a1,0x8f102db0,0x67752870,0x28224061,0xf9dbf852,0xb68c9043, + 0xcba219a9,0x84f571b8,0x550cc98b,0x1a5ba19a,0xf23ea45a,0xbd69cc4b,0x6c907478,0x23c71c69, + 0xb89b624f,0xf7cc0a5e,0x2635b26d,0x6962da7c,0x8107dfbc,0xce50b7ad,0x1fa90f9e,0x50fe678f, + 0xe5f41c4a,0xaaa3745b,0x7b5acc68,0x340da479,0xdc68a1b9,0x933fc9a8,0x42c6719b,0x0d91198a, + 0x96cd67ac,0xd99a0fbd,0x0863b78e,0x4734df9f,0xaf51da5f,0xe006b24e,0x31ff0a7d,0x7ea8626c, + 0x0386eb86,0x4cd18397,0x9d283ba4,0xd27f53b5,0x3a1a5675,0x754d3e64,0xa4b48657,0xebe3ee46, + 0x70bf9060,0x3fe8f871,0xee114042,0xa1462853,0x49232d93,0x06744582,0xd78dfdb1,0x98da95a0, + 0xb958178c,0xf60f7f9d,0x27f6c7ae,0x68a1afbf,0x80c4aa7f,0xcf93c26e,0x1e6a7a5d,0x513d124c, + 0xca616c6a,0x8536047b,0x54cfbc48,0x1b98d459,0xf3fdd199,0xbcaab988,0x6d5301bb,0x220469aa, + 0x5f2ae040,0x107d8851,0xc1843062,0x8ed35873,0x66b65db3,0x29e135a2,0xf8188d91,0xb74fe580, + 0x2c139ba6,0x6344f3b7,0xb2bd4b84,0xfdea2395,0x158f2655,0x5ad84e44,0x8b21f677,0xc4769e66, + 0x717ce5a3,0x3e2b8db2,0xefd23581,0xa0855d90,0x48e05850,0x07b73041,0xd64e8872,0x9919e063, + 0x02459e45,0x4d12f654,0x9ceb4e67,0xd3bc2676,0x3bd923b6,0x748e4ba7,0xa577f394,0xea209b85, + 0x970e126f,0xd8597a7e,0x09a0c24d,0x46f7aa5c,0xae92af9c,0xe1c5c78d,0x303c7fbe,0x7f6b17af, + 0xe4376989,0xab600198,0x7a99b9ab,0x35ced1ba,0xddabd47a,0x92fcbc6b,0x43050458,0x0c526c49}, + +{0x00000000,0x5ba1dcca,0xb743b994,0xece2655e,0x6a466e9f,0x31e7b255,0xdd05d70b,0x86a40bc1, + 0xd48cdd3e,0x8f2d01f4,0x63cf64aa,0x386eb860,0xbecab3a1,0xe56b6f6b,0x09890a35,0x5228d6ff, + 0xadd8a7cb,0xf6797b01,0x1a9b1e5f,0x413ac295,0xc79ec954,0x9c3f159e,0x70dd70c0,0x2b7cac0a, + 0x79547af5,0x22f5a63f,0xce17c361,0x95b61fab,0x1312146a,0x48b3c8a0,0xa451adfe,0xfff07134, + 0x5f705221,0x04d18eeb,0xe833ebb5,0xb392377f,0x35363cbe,0x6e97e074,0x8275852a,0xd9d459e0, + 0x8bfc8f1f,0xd05d53d5,0x3cbf368b,0x671eea41,0xe1bae180,0xba1b3d4a,0x56f95814,0x0d5884de, + 0xf2a8f5ea,0xa9092920,0x45eb4c7e,0x1e4a90b4,0x98ee9b75,0xc34f47bf,0x2fad22e1,0x740cfe2b, + 0x262428d4,0x7d85f41e,0x91679140,0xcac64d8a,0x4c62464b,0x17c39a81,0xfb21ffdf,0xa0802315, + 0xbee0a442,0xe5417888,0x09a31dd6,0x5202c11c,0xd4a6cadd,0x8f071617,0x63e57349,0x3844af83, + 0x6a6c797c,0x31cda5b6,0xdd2fc0e8,0x868e1c22,0x002a17e3,0x5b8bcb29,0xb769ae77,0xecc872bd, + 0x13380389,0x4899df43,0xa47bba1d,0xffda66d7,0x797e6d16,0x22dfb1dc,0xce3dd482,0x959c0848, + 0xc7b4deb7,0x9c15027d,0x70f76723,0x2b56bbe9,0xadf2b028,0xf6536ce2,0x1ab109bc,0x4110d576, + 0xe190f663,0xba312aa9,0x56d34ff7,0x0d72933d,0x8bd698fc,0xd0774436,0x3c952168,0x6734fda2, + 0x351c2b5d,0x6ebdf797,0x825f92c9,0xd9fe4e03,0x5f5a45c2,0x04fb9908,0xe819fc56,0xb3b8209c, + 0x4c4851a8,0x17e98d62,0xfb0be83c,0xa0aa34f6,0x260e3f37,0x7dafe3fd,0x914d86a3,0xcaec5a69, + 0x98c48c96,0xc365505c,0x2f873502,0x7426e9c8,0xf282e209,0xa9233ec3,0x45c15b9d,0x1e608757, + 0x79005533,0x22a189f9,0xce43eca7,0x95e2306d,0x13463bac,0x48e7e766,0xa4058238,0xffa45ef2, + 0xad8c880d,0xf62d54c7,0x1acf3199,0x416eed53,0xc7cae692,0x9c6b3a58,0x70895f06,0x2b2883cc, + 0xd4d8f2f8,0x8f792e32,0x639b4b6c,0x383a97a6,0xbe9e9c67,0xe53f40ad,0x09dd25f3,0x527cf939, + 0x00542fc6,0x5bf5f30c,0xb7179652,0xecb64a98,0x6a124159,0x31b39d93,0xdd51f8cd,0x86f02407, + 0x26700712,0x7dd1dbd8,0x9133be86,0xca92624c,0x4c36698d,0x1797b547,0xfb75d019,0xa0d40cd3, + 0xf2fcda2c,0xa95d06e6,0x45bf63b8,0x1e1ebf72,0x98bab4b3,0xc31b6879,0x2ff90d27,0x7458d1ed, + 0x8ba8a0d9,0xd0097c13,0x3ceb194d,0x674ac587,0xe1eece46,0xba4f128c,0x56ad77d2,0x0d0cab18, + 0x5f247de7,0x0485a12d,0xe867c473,0xb3c618b9,0x35621378,0x6ec3cfb2,0x8221aaec,0xd9807626, + 0xc7e0f171,0x9c412dbb,0x70a348e5,0x2b02942f,0xada69fee,0xf6074324,0x1ae5267a,0x4144fab0, + 0x136c2c4f,0x48cdf085,0xa42f95db,0xff8e4911,0x792a42d0,0x228b9e1a,0xce69fb44,0x95c8278e, + 0x6a3856ba,0x31998a70,0xdd7bef2e,0x86da33e4,0x007e3825,0x5bdfe4ef,0xb73d81b1,0xec9c5d7b, + 0xbeb48b84,0xe515574e,0x09f73210,0x5256eeda,0xd4f2e51b,0x8f5339d1,0x63b15c8f,0x38108045, + 0x9890a350,0xc3317f9a,0x2fd31ac4,0x7472c60e,0xf2d6cdcf,0xa9771105,0x4595745b,0x1e34a891, + 0x4c1c7e6e,0x17bda2a4,0xfb5fc7fa,0xa0fe1b30,0x265a10f1,0x7dfbcc3b,0x9119a965,0xcab875af, + 0x3548049b,0x6ee9d851,0x820bbd0f,0xd9aa61c5,0x5f0e6a04,0x04afb6ce,0xe84dd390,0xb3ec0f5a, + 0xe1c4d9a5,0xba65056f,0x56876031,0x0d26bcfb,0x8b82b73a,0xd0236bf0,0x3cc10eae,0x6760d264}}; diff --git a/code/libogg-1.3.3/src/framing.c b/code/libogg-1.3.6/src/framing.c similarity index 92% rename from code/libogg-1.3.3/src/framing.c rename to code/libogg-1.3.6/src/framing.c index 79fc715c..0ddf0fcf 100644 --- a/code/libogg-1.3.3/src/framing.c +++ b/code/libogg-1.3.6/src/framing.c @@ -5,14 +5,13 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * + * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2018 * * by the Xiph.Org Foundation http://www.xiph.org/ * * * ******************************************************************** function: code raw packets into framed OggSquish stream and decode Ogg streams back into raw packets - last mod: $Id$ note: The CRC code is directly derived from public domain code by Ross Williams (ross@guest.adelaide.edu.au). See docs/framing.html @@ -20,6 +19,10 @@ ********************************************************************/ +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + #include #include #include @@ -45,7 +48,7 @@ int ogg_page_eos(const ogg_page *og){ ogg_int64_t ogg_page_granulepos(const ogg_page *og){ unsigned char *page=og->header; - ogg_int64_t granulepos=page[13]&(0xff); + ogg_uint64_t granulepos=page[13]&(0xff); granulepos= (granulepos<<8)|(page[12]&0xff); granulepos= (granulepos<<8)|(page[11]&0xff); granulepos= (granulepos<<8)|(page[10]&0xff); @@ -53,21 +56,21 @@ ogg_int64_t ogg_page_granulepos(const ogg_page *og){ granulepos= (granulepos<<8)|(page[8]&0xff); granulepos= (granulepos<<8)|(page[7]&0xff); granulepos= (granulepos<<8)|(page[6]&0xff); - return(granulepos); + return((ogg_int64_t)granulepos); } int ogg_page_serialno(const ogg_page *og){ - return(og->header[14] | - (og->header[15]<<8) | - (og->header[16]<<16) | - (og->header[17]<<24)); + return((int)((ogg_uint32_t)og->header[14]) | + ((ogg_uint32_t)og->header[15]<<8) | + ((ogg_uint32_t)og->header[16]<<16) | + ((ogg_uint32_t)og->header[17]<<24)); } long ogg_page_pageno(const ogg_page *og){ - return(og->header[18] | - (og->header[19]<<8) | - (og->header[20]<<16) | - (og->header[21]<<24)); + return((long)((ogg_uint32_t)og->header[18]) | + ((ogg_uint32_t)og->header[19]<<8) | + ((ogg_uint32_t)og->header[20]<<16) | + ((ogg_uint32_t)og->header[21]<<24)); } @@ -99,90 +102,31 @@ int ogg_page_packets(const ogg_page *og){ #if 0 /* helper to initialize lookup for direct-table CRC (illustrative; we - use the static init below) */ + use the static init in crctable.h) */ -static ogg_uint32_t _ogg_crc_entry(unsigned long index){ - int i; - unsigned long r; +static void _ogg_crc_init(){ + int i, j; + ogg_uint32_t polynomial, crc; + polynomial = 0x04c11db7; /* The same as the ethernet generator + polynomial, although we use an + unreflected alg and an init/final + of 0, not 0xffffffff */ + for (i = 0; i <= 0xFF; i++){ + crc = i << 24; - r = index << 24; - for (i=0; i<8; i++) - if (r & 0x80000000UL) - r = (r << 1) ^ 0x04c11db7; /* The same as the ethernet generator - polynomial, although we use an - unreflected alg and an init/final - of 0, not 0xffffffff */ - else - r<<=1; - return (r & 0xffffffffUL); + for (j = 0; j < 8; j++) + crc = (crc << 1) ^ (crc & (1 << 31) ? polynomial : 0); + + crc_lookup[0][i] = crc; + } + + for (i = 0; i <= 0xFF; i++) + for (j = 1; j < 8; j++) + crc_lookup[j][i] = crc_lookup[0][(crc_lookup[j - 1][i] >> 24) & 0xFF] ^ (crc_lookup[j - 1][i] << 8); } #endif -static const ogg_uint32_t crc_lookup[256]={ - 0x00000000,0x04c11db7,0x09823b6e,0x0d4326d9, - 0x130476dc,0x17c56b6b,0x1a864db2,0x1e475005, - 0x2608edb8,0x22c9f00f,0x2f8ad6d6,0x2b4bcb61, - 0x350c9b64,0x31cd86d3,0x3c8ea00a,0x384fbdbd, - 0x4c11db70,0x48d0c6c7,0x4593e01e,0x4152fda9, - 0x5f15adac,0x5bd4b01b,0x569796c2,0x52568b75, - 0x6a1936c8,0x6ed82b7f,0x639b0da6,0x675a1011, - 0x791d4014,0x7ddc5da3,0x709f7b7a,0x745e66cd, - 0x9823b6e0,0x9ce2ab57,0x91a18d8e,0x95609039, - 0x8b27c03c,0x8fe6dd8b,0x82a5fb52,0x8664e6e5, - 0xbe2b5b58,0xbaea46ef,0xb7a96036,0xb3687d81, - 0xad2f2d84,0xa9ee3033,0xa4ad16ea,0xa06c0b5d, - 0xd4326d90,0xd0f37027,0xddb056fe,0xd9714b49, - 0xc7361b4c,0xc3f706fb,0xceb42022,0xca753d95, - 0xf23a8028,0xf6fb9d9f,0xfbb8bb46,0xff79a6f1, - 0xe13ef6f4,0xe5ffeb43,0xe8bccd9a,0xec7dd02d, - 0x34867077,0x30476dc0,0x3d044b19,0x39c556ae, - 0x278206ab,0x23431b1c,0x2e003dc5,0x2ac12072, - 0x128e9dcf,0x164f8078,0x1b0ca6a1,0x1fcdbb16, - 0x018aeb13,0x054bf6a4,0x0808d07d,0x0cc9cdca, - 0x7897ab07,0x7c56b6b0,0x71159069,0x75d48dde, - 0x6b93dddb,0x6f52c06c,0x6211e6b5,0x66d0fb02, - 0x5e9f46bf,0x5a5e5b08,0x571d7dd1,0x53dc6066, - 0x4d9b3063,0x495a2dd4,0x44190b0d,0x40d816ba, - 0xaca5c697,0xa864db20,0xa527fdf9,0xa1e6e04e, - 0xbfa1b04b,0xbb60adfc,0xb6238b25,0xb2e29692, - 0x8aad2b2f,0x8e6c3698,0x832f1041,0x87ee0df6, - 0x99a95df3,0x9d684044,0x902b669d,0x94ea7b2a, - 0xe0b41de7,0xe4750050,0xe9362689,0xedf73b3e, - 0xf3b06b3b,0xf771768c,0xfa325055,0xfef34de2, - 0xc6bcf05f,0xc27dede8,0xcf3ecb31,0xcbffd686, - 0xd5b88683,0xd1799b34,0xdc3abded,0xd8fba05a, - 0x690ce0ee,0x6dcdfd59,0x608edb80,0x644fc637, - 0x7a089632,0x7ec98b85,0x738aad5c,0x774bb0eb, - 0x4f040d56,0x4bc510e1,0x46863638,0x42472b8f, - 0x5c007b8a,0x58c1663d,0x558240e4,0x51435d53, - 0x251d3b9e,0x21dc2629,0x2c9f00f0,0x285e1d47, - 0x36194d42,0x32d850f5,0x3f9b762c,0x3b5a6b9b, - 0x0315d626,0x07d4cb91,0x0a97ed48,0x0e56f0ff, - 0x1011a0fa,0x14d0bd4d,0x19939b94,0x1d528623, - 0xf12f560e,0xf5ee4bb9,0xf8ad6d60,0xfc6c70d7, - 0xe22b20d2,0xe6ea3d65,0xeba91bbc,0xef68060b, - 0xd727bbb6,0xd3e6a601,0xdea580d8,0xda649d6f, - 0xc423cd6a,0xc0e2d0dd,0xcda1f604,0xc960ebb3, - 0xbd3e8d7e,0xb9ff90c9,0xb4bcb610,0xb07daba7, - 0xae3afba2,0xaafbe615,0xa7b8c0cc,0xa379dd7b, - 0x9b3660c6,0x9ff77d71,0x92b45ba8,0x9675461f, - 0x8832161a,0x8cf30bad,0x81b02d74,0x857130c3, - 0x5d8a9099,0x594b8d2e,0x5408abf7,0x50c9b640, - 0x4e8ee645,0x4a4ffbf2,0x470cdd2b,0x43cdc09c, - 0x7b827d21,0x7f436096,0x7200464f,0x76c15bf8, - 0x68860bfd,0x6c47164a,0x61043093,0x65c52d24, - 0x119b4be9,0x155a565e,0x18197087,0x1cd86d30, - 0x029f3d35,0x065e2082,0x0b1d065b,0x0fdc1bec, - 0x3793a651,0x3352bbe6,0x3e119d3f,0x3ad08088, - 0x2497d08d,0x2056cd3a,0x2d15ebe3,0x29d4f654, - 0xc5a92679,0xc1683bce,0xcc2b1d17,0xc8ea00a0, - 0xd6ad50a5,0xd26c4d12,0xdf2f6bcb,0xdbee767c, - 0xe3a1cbc1,0xe760d676,0xea23f0af,0xeee2ed18, - 0xf0a5bd1d,0xf464a0aa,0xf9278673,0xfde69bc4, - 0x89b8fd09,0x8d79e0be,0x803ac667,0x84fbdbd0, - 0x9abc8bd5,0x9e7d9662,0x933eb0bb,0x97ffad0c, - 0xafb010b1,0xab710d06,0xa6322bdf,0xa2f33668, - 0xbcb4666d,0xb8757bda,0xb5365d03,0xb1f740b4}; +#include "crctable.h" /* init the encode/decode logical stream state */ @@ -290,10 +234,27 @@ static int _os_lacing_expand(ogg_stream_state *os,long needed){ /* Direct table CRC; note that this will be faster in the future if we perform the checksum simultaneously with other copies */ +static ogg_uint32_t _os_update_crc(ogg_uint32_t crc, unsigned char *buffer, int size){ + while (size>=8){ + crc^=((ogg_uint32_t)buffer[0]<<24)|((ogg_uint32_t)buffer[1]<<16)|((ogg_uint32_t)buffer[2]<<8)|((ogg_uint32_t)buffer[3]); + + crc=crc_lookup[7][ crc>>24 ]^crc_lookup[6][(crc>>16)&0xFF]^ + crc_lookup[5][(crc>> 8)&0xFF]^crc_lookup[4][ crc &0xFF]^ + crc_lookup[3][buffer[4] ]^crc_lookup[2][buffer[5] ]^ + crc_lookup[1][buffer[6] ]^crc_lookup[0][buffer[7] ]; + + buffer+=8; + size-=8; + } + + while (size--) + crc=(crc<<8)^crc_lookup[0][((crc >> 24)&0xff)^*buffer++]; + return crc; +} + void ogg_page_checksum_set(ogg_page *og){ if(og){ ogg_uint32_t crc_reg=0; - int i; /* safety; needed for API behavior, but not framing code */ og->header[22]=0; @@ -301,10 +262,8 @@ void ogg_page_checksum_set(ogg_page *og){ og->header[24]=0; og->header[25]=0; - for(i=0;iheader_len;i++) - crc_reg=(crc_reg<<8)^crc_lookup[((crc_reg >> 24)&0xff)^og->header[i]]; - for(i=0;ibody_len;i++) - crc_reg=(crc_reg<<8)^crc_lookup[((crc_reg >> 24)&0xff)^og->body[i]]; + crc_reg=_os_update_crc(crc_reg,og->header,og->header_len); + crc_reg=_os_update_crc(crc_reg,og->body,og->body_len); og->header[22]=(unsigned char)(crc_reg&0xff); og->header[23]=(unsigned char)((crc_reg>>8)&0xff); @@ -390,12 +349,13 @@ int ogg_stream_packetin(ogg_stream_state *os,ogg_packet *op){ static int ogg_stream_flush_i(ogg_stream_state *os,ogg_page *og, int force, int nfill){ int i; int vals=0; - int maxvals=(os->lacing_fill>255?255:os->lacing_fill); + int maxvals; int bytes=0; long acc=0; ogg_int64_t granule_pos=-1; if(ogg_stream_check(os)) return(0); + maxvals=(os->lacing_fill>255?255:os->lacing_fill); if(maxvals==0) return(0); /* construct a page */ @@ -414,9 +374,9 @@ static int ogg_stream_flush_i(ogg_stream_state *os,ogg_page *og, int force, int }else{ /* The extra packets_done, packet_just_done logic here attempts to do two things: - 1) Don't unneccessarily span pages. + 1) Don't unnecessarily span pages. 2) Unless necessary, don't flush pages if there are less than four packets on - them; this expands page size to reduce unneccessary overhead if incoming packets + them; this expands page size to reduce unnecessary overhead if incoming packets are large. These are not necessary behaviors, just 'always better than naive flushing' without requiring an application to explicitly request a specific optimized @@ -638,9 +598,14 @@ char *ogg_sync_buffer(ogg_sync_state *oy, long size){ if(size>oy->storage-oy->fill){ /* We need to extend the internal buffer */ - long newsize=size+oy->fill+4096; /* an extra page to be nice */ + long newsize; void *ret; + if(size>INT_MAX-4096-oy->fill){ + ogg_sync_clear(oy); + return NULL; + } + newsize=size+oy->fill+4096; /* an extra page to be nice */ if(oy->data) ret=_ogg_realloc(oy->data,newsize); else @@ -675,12 +640,15 @@ int ogg_sync_wrote(ogg_sync_state *oy, long bytes){ */ long ogg_sync_pageseek(ogg_sync_state *oy,ogg_page *og){ - unsigned char *page=oy->data+oy->returned; + unsigned char *page; unsigned char *next; - long bytes=oy->fill-oy->returned; + long bytes; if(ogg_sync_check(oy))return 0; + page=oy->data+oy->returned; + bytes=oy->fill-oy->returned; + if(oy->headerbytes==0){ int headerbytes,i; if(bytes<27)return(0); /* not enough for a header */ @@ -723,16 +691,15 @@ long ogg_sync_pageseek(ogg_sync_state *oy,ogg_page *og){ /* replace the computed checksum with the one actually read in */ memcpy(page+22,chksum,4); +#ifndef DISABLE_CRC /* Bad checksum. Lose sync */ goto sync_fail; +#endif } } /* yes, have a whole page all ready to go */ { - unsigned char *page=oy->data+oy->returned; - long bytes; - if(og){ og->header=page; og->header_len=oy->headerbytes; @@ -1123,11 +1090,11 @@ void print_header(ogg_page *og){ (int)og->header[4],(int)og->header[5]); fprintf(stderr," granulepos: %d serialno: %d pageno: %ld\n", - (og->header[9]<<24)|(og->header[8]<<16)| + ((unsigned)og->header[9]<<24)|(og->header[8]<<16)| (og->header[7]<<8)|og->header[6], - (og->header[17]<<24)|(og->header[16]<<16)| + ((unsigned)og->header[17]<<24)|(og->header[16]<<16)| (og->header[15]<<8)|og->header[14], - ((long)(og->header[21])<<24)|(og->header[20]<<16)| + ((long)((unsigned)og->header[21])<<24)|(og->header[20]<<16)| (og->header[19]<<8)|og->header[18]); fprintf(stderr," checksum: %02x:%02x:%02x:%02x\n segments: %d (", @@ -1140,19 +1107,26 @@ void print_header(ogg_page *og){ fprintf(stderr,")\n\n"); } -void copy_page(ogg_page *og){ +static int copy_page(ogg_page *og){ unsigned char *temp=_ogg_malloc(og->header_len); + if (!temp) + return -1; memcpy(temp,og->header,og->header_len); og->header=temp; temp=_ogg_malloc(og->body_len); + if (!temp) + return -1; memcpy(temp,og->body,og->body_len); og->body=temp; + return 0; } -void free_page(ogg_page *og){ +static void free_page(ogg_page *og){ _ogg_free (og->header); + og->header=NULL; _ogg_free (og->body); + og->body=NULL; } void error(void){ @@ -1535,6 +1509,11 @@ void test_pack(const int *pl, const int **headers, int byteskip, int byteskipcount=0; + if (!data) { + fprintf(stderr,"unable to allocate requried data buffer!\n"); + exit(1); + } + ogg_stream_reset(&os_en); ogg_stream_reset(&os_de); ogg_sync_reset(&oy); @@ -1606,7 +1585,7 @@ void test_pack(const int *pl, const int **headers, int byteskip, byteskipcount=byteskip; } - ogg_sync_wrote(&oy,next-buf); + ogg_sync_wrote(&oy,(long)(next-buf)); while(1){ int ret=ogg_sync_pageout(&oy,&og_de); @@ -1814,6 +1793,7 @@ int main(void){ test_pack(packets,headret,0,0,0); } +#ifndef DISABLE_CRC { /* test for the libogg 1.1.1 resync in large continuation bug found by Josh Coalson) */ @@ -1823,6 +1803,9 @@ int main(void){ fprintf(stderr,"testing continuation resync in very large packets... "); test_pack(packets,headret,100,2,3); } +#else + fprintf(stderr,"Skipping continuation resync test due to --disable-crc\n"); +#endif { /* term only page. why not? */ @@ -1842,6 +1825,11 @@ int main(void){ int inptr=0,i,j; ogg_page og[5]; + if (!data) { + fprintf(stderr,"unable to allocate requried packet data buffer!\n"); + exit(1); + } + ogg_stream_reset(&os_en); for(i=0;pl[i]!=-1;i++){ @@ -1865,7 +1853,10 @@ int main(void){ fprintf(stderr,"Too few pages output building sync tests!\n"); exit(1); } - copy_page(&og[i]); + if (-1 == copy_page(&og[i])) { + fprintf(stderr,"unable to copy page building sync tests!\n"); + exit(1); + } } /* Test lost pages on pagein/packetout: no rollback */ @@ -2084,6 +2075,7 @@ int main(void){ fprintf(stderr,"ok.\n"); } +#ifndef DISABLE_CRC /* Test recapture: page + garbage + page */ { ogg_page og_de; @@ -2125,6 +2117,9 @@ int main(void){ fprintf(stderr,"ok.\n"); } +#else + fprintf(stderr,"Skipping recapture test due to --disable-crc\n"); +#endif /* Free page data that was previously copied */ { @@ -2133,6 +2128,9 @@ int main(void){ } } } + ogg_sync_clear(&oy); + ogg_stream_clear(&os_en); + ogg_stream_clear(&os_de); return(0); } diff --git a/code/libvorbis-1.3.6/include/vorbis/codec.h b/code/libvorbis-1.3.7/include/vorbis/codec.h similarity index 99% rename from code/libvorbis-1.3.6/include/vorbis/codec.h rename to code/libvorbis-1.3.7/include/vorbis/codec.h index 42aa2913..f8a912bc 100644 --- a/code/libvorbis-1.3.6/include/vorbis/codec.h +++ b/code/libvorbis-1.3.7/include/vorbis/codec.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2001 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * ******************************************************************** diff --git a/code/libvorbis-1.3.6/include/vorbis/vorbisenc.h b/code/libvorbis-1.3.7/include/vorbis/vorbisenc.h similarity index 99% rename from code/libvorbis-1.3.6/include/vorbis/vorbisenc.h rename to code/libvorbis-1.3.7/include/vorbis/vorbisenc.h index 55f3b4a6..085b15e6 100644 --- a/code/libvorbis-1.3.6/include/vorbis/vorbisenc.h +++ b/code/libvorbis-1.3.7/include/vorbis/vorbisenc.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2001 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/include/vorbis/vorbisfile.h b/code/libvorbis-1.3.7/include/vorbis/vorbisfile.h similarity index 99% rename from code/libvorbis-1.3.6/include/vorbis/vorbisfile.h rename to code/libvorbis-1.3.7/include/vorbis/vorbisfile.h index 56626119..3d65393f 100644 --- a/code/libvorbis-1.3.6/include/vorbis/vorbisfile.h +++ b/code/libvorbis-1.3.7/include/vorbis/vorbisfile.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/analysis.c b/code/libvorbis-1.3.7/lib/analysis.c similarity index 97% rename from code/libvorbis-1.3.6/lib/analysis.c rename to code/libvorbis-1.3.7/lib/analysis.c index 0e11a167..14919737 100644 --- a/code/libvorbis-1.3.6/lib/analysis.c +++ b/code/libvorbis-1.3.7/lib/analysis.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/backends.h b/code/libvorbis-1.3.7/lib/backends.h similarity index 98% rename from code/libvorbis-1.3.6/lib/backends.h rename to code/libvorbis-1.3.7/lib/backends.h index 22809d46..670b0b90 100644 --- a/code/libvorbis-1.3.6/lib/backends.h +++ b/code/libvorbis-1.3.7/lib/backends.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.7/lib/barkmel.c b/code/libvorbis-1.3.7/lib/barkmel.c new file mode 100644 index 00000000..f833c302 --- /dev/null +++ b/code/libvorbis-1.3.7/lib/barkmel.c @@ -0,0 +1,63 @@ +/******************************************************************** + * * + * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE. * + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * + * * + * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * + * by the Xiph.Org Foundation https://xiph.org/ * + * * + ******************************************************************** + + function: bark scale utility + + ********************************************************************/ + +#include +#include "scales.h" +int main(){ + int i; + double rate; + for(i=64;i<32000;i*=2){ + rate=48000.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=44100.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=32000.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=22050.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=16000.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=11025.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + rate=8000.f; + fprintf(stderr,"rate=%gHz, block=%d, f(1)=%.2gHz bark(1)=%.2g (of %.2g)\n\n", + rate,i,rate/2 / (i/2),toBARK(rate/2 /(i/2)),toBARK(rate/2)); + + + } + { + float i; + int j; + for(i=0.,j=0;i<28;i+=1,j++){ + fprintf(stderr,"(%d) bark=%f %gHz (%d of 128)\n", + j,i,fromBARK(i),(int)(fromBARK(i)/22050.*128.)); + } + } + return(0); +} + diff --git a/code/libvorbis-1.3.6/lib/bitrate.c b/code/libvorbis-1.3.7/lib/bitrate.c similarity index 99% rename from code/libvorbis-1.3.6/lib/bitrate.c rename to code/libvorbis-1.3.7/lib/bitrate.c index 96055140..132553cb 100644 --- a/code/libvorbis-1.3.6/lib/bitrate.c +++ b/code/libvorbis-1.3.7/lib/bitrate.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/bitrate.h b/code/libvorbis-1.3.7/lib/bitrate.h similarity index 96% rename from code/libvorbis-1.3.6/lib/bitrate.h rename to code/libvorbis-1.3.7/lib/bitrate.h index 655a68cc..48fa1505 100644 --- a/code/libvorbis-1.3.6/lib/bitrate.h +++ b/code/libvorbis-1.3.7/lib/bitrate.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/block.c b/code/libvorbis-1.3.7/lib/block.c similarity index 99% rename from code/libvorbis-1.3.6/lib/block.c rename to code/libvorbis-1.3.7/lib/block.c index db245b3e..6a50da08 100644 --- a/code/libvorbis-1.3.6/lib/block.c +++ b/code/libvorbis-1.3.7/lib/block.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/books/coupled/res_books_51.h b/code/libvorbis-1.3.7/lib/books/coupled/res_books_51.h similarity index 99% rename from code/libvorbis-1.3.6/lib/books/coupled/res_books_51.h rename to code/libvorbis-1.3.7/lib/books/coupled/res_books_51.h index 47df4b22..eb569c6f 100644 --- a/code/libvorbis-1.3.6/lib/books/coupled/res_books_51.h +++ b/code/libvorbis-1.3.7/lib/books/coupled/res_books_51.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** * diff --git a/code/libvorbis-1.3.6/lib/books/coupled/res_books_stereo.h b/code/libvorbis-1.3.7/lib/books/coupled/res_books_stereo.h similarity index 99% rename from code/libvorbis-1.3.6/lib/books/coupled/res_books_stereo.h rename to code/libvorbis-1.3.7/lib/books/coupled/res_books_stereo.h index 61d93404..7b53cb97 100644 --- a/code/libvorbis-1.3.6/lib/books/coupled/res_books_stereo.h +++ b/code/libvorbis-1.3.7/lib/books/coupled/res_books_stereo.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/books/floor/floor_books.h b/code/libvorbis-1.3.7/lib/books/floor/floor_books.h similarity index 99% rename from code/libvorbis-1.3.6/lib/books/floor/floor_books.h rename to code/libvorbis-1.3.7/lib/books/floor/floor_books.h index 67d5f31a..d26664f7 100644 --- a/code/libvorbis-1.3.6/lib/books/floor/floor_books.h +++ b/code/libvorbis-1.3.7/lib/books/floor/floor_books.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/books/uncoupled/res_books_uncoupled.h b/code/libvorbis-1.3.7/lib/books/uncoupled/res_books_uncoupled.h similarity index 99% rename from code/libvorbis-1.3.6/lib/books/uncoupled/res_books_uncoupled.h rename to code/libvorbis-1.3.7/lib/books/uncoupled/res_books_uncoupled.h index 3d658ec4..107e22f9 100644 --- a/code/libvorbis-1.3.6/lib/books/uncoupled/res_books_uncoupled.h +++ b/code/libvorbis-1.3.7/lib/books/uncoupled/res_books_uncoupled.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/codebook.c b/code/libvorbis-1.3.7/lib/codebook.c similarity index 99% rename from code/libvorbis-1.3.6/lib/codebook.c rename to code/libvorbis-1.3.7/lib/codebook.c index 78672e22..7a0c2067 100644 --- a/code/libvorbis-1.3.6/lib/codebook.c +++ b/code/libvorbis-1.3.7/lib/codebook.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/codebook.h b/code/libvorbis-1.3.7/lib/codebook.h similarity index 98% rename from code/libvorbis-1.3.6/lib/codebook.h rename to code/libvorbis-1.3.7/lib/codebook.h index 08440c69..7d4e2aae 100644 --- a/code/libvorbis-1.3.6/lib/codebook.h +++ b/code/libvorbis-1.3.7/lib/codebook.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/codec_internal.h b/code/libvorbis-1.3.7/lib/codec_internal.h similarity index 98% rename from code/libvorbis-1.3.6/lib/codec_internal.h rename to code/libvorbis-1.3.7/lib/codec_internal.h index e522be18..2ecf5e5c 100644 --- a/code/libvorbis-1.3.6/lib/codec_internal.h +++ b/code/libvorbis-1.3.7/lib/codec_internal.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/envelope.c b/code/libvorbis-1.3.7/lib/envelope.c similarity index 99% rename from code/libvorbis-1.3.6/lib/envelope.c rename to code/libvorbis-1.3.7/lib/envelope.c index da752375..22d39aa6 100644 --- a/code/libvorbis-1.3.6/lib/envelope.c +++ b/code/libvorbis-1.3.7/lib/envelope.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/envelope.h b/code/libvorbis-1.3.7/lib/envelope.h similarity index 96% rename from code/libvorbis-1.3.6/lib/envelope.h rename to code/libvorbis-1.3.7/lib/envelope.h index f466efde..2ef60a82 100644 --- a/code/libvorbis-1.3.6/lib/envelope.h +++ b/code/libvorbis-1.3.7/lib/envelope.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/floor0.c b/code/libvorbis-1.3.7/lib/floor0.c similarity index 99% rename from code/libvorbis-1.3.6/lib/floor0.c rename to code/libvorbis-1.3.7/lib/floor0.c index 443c0e5a..f4a6d4d5 100644 --- a/code/libvorbis-1.3.6/lib/floor0.c +++ b/code/libvorbis-1.3.7/lib/floor0.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/floor1.c b/code/libvorbis-1.3.7/lib/floor1.c similarity index 99% rename from code/libvorbis-1.3.6/lib/floor1.c rename to code/libvorbis-1.3.7/lib/floor1.c index 673e954c..c4fe3ea7 100644 --- a/code/libvorbis-1.3.6/lib/floor1.c +++ b/code/libvorbis-1.3.7/lib/floor1.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/highlevel.h b/code/libvorbis-1.3.7/lib/highlevel.h similarity index 96% rename from code/libvorbis-1.3.6/lib/highlevel.h rename to code/libvorbis-1.3.7/lib/highlevel.h index 337b75bf..7690e3eb 100644 --- a/code/libvorbis-1.3.6/lib/highlevel.h +++ b/code/libvorbis-1.3.7/lib/highlevel.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/info.c b/code/libvorbis-1.3.7/lib/info.c similarity index 97% rename from code/libvorbis-1.3.6/lib/info.c rename to code/libvorbis-1.3.7/lib/info.c index 3fbb7c75..f2e39e38 100644 --- a/code/libvorbis-1.3.6/lib/info.c +++ b/code/libvorbis-1.3.7/lib/info.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -19,7 +19,6 @@ #include #include -#include #include #include "vorbis/codec.h" #include "codec_internal.h" @@ -30,8 +29,8 @@ #include "misc.h" #include "os.h" -#define GENERAL_VENDOR_STRING "Xiph.Org libVorbis 1.3.6" -#define ENCODE_VENDOR_STRING "Xiph.Org libVorbis I 20180316 (Now 100% fewer shells)" +#define GENERAL_VENDOR_STRING "Xiph.Org libVorbis 1.3.7" +#define ENCODE_VENDOR_STRING "Xiph.Org libVorbis I 20200704 (Reducing Environment)" /* helpers */ static void _v_writestring(oggpack_buffer *o,const char *s, int bytes){ @@ -47,6 +46,10 @@ static void _v_readstring(oggpack_buffer *o,char *buf,int bytes){ } } +static int _v_toupper(int c) { + return (c >= 'a' && c <= 'z') ? (c & ~('a' - 'A')) : c; +} + void vorbis_comment_init(vorbis_comment *vc){ memset(vc,0,sizeof(*vc)); } @@ -78,7 +81,7 @@ void vorbis_comment_add_tag(vorbis_comment *vc, const char *tag, const char *con static int tagcompare(const char *s1, const char *s2, int n){ int c=0; while(c < n){ - if(toupper(s1[c]) != toupper(s2[c])) + if(_v_toupper(s1[c]) != _v_toupper(s2[c])) return !0; c++; } @@ -203,6 +206,7 @@ void vorbis_info_clear(vorbis_info *vi){ static int _vorbis_unpack_info(vorbis_info *vi,oggpack_buffer *opb){ codec_setup_info *ci=vi->codec_setup; + int bs; if(!ci)return(OV_EFAULT); vi->version=oggpack_read(opb,32); @@ -215,8 +219,12 @@ static int _vorbis_unpack_info(vorbis_info *vi,oggpack_buffer *opb){ vi->bitrate_nominal=(ogg_int32_t)oggpack_read(opb,32); vi->bitrate_lower=(ogg_int32_t)oggpack_read(opb,32); - ci->blocksizes[0]=1<blocksizes[1]=1<blocksizes[0]=1<blocksizes[1]=1<rate<1)goto err_out; if(vi->channels<1)goto err_out; diff --git a/code/libvorbis-1.3.6/lib/lookup.c b/code/libvorbis-1.3.7/lib/lookup.c similarity index 98% rename from code/libvorbis-1.3.6/lib/lookup.c rename to code/libvorbis-1.3.7/lib/lookup.c index 1cc1f88e..7cd01a44 100644 --- a/code/libvorbis-1.3.6/lib/lookup.c +++ b/code/libvorbis-1.3.7/lib/lookup.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/lookup.h b/code/libvorbis-1.3.7/lib/lookup.h similarity index 94% rename from code/libvorbis-1.3.6/lib/lookup.h rename to code/libvorbis-1.3.7/lib/lookup.h index 4bc0f3a2..ec05014f 100644 --- a/code/libvorbis-1.3.6/lib/lookup.h +++ b/code/libvorbis-1.3.7/lib/lookup.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/lookup_data.h b/code/libvorbis-1.3.7/lib/lookup_data.h similarity index 99% rename from code/libvorbis-1.3.6/lib/lookup_data.h rename to code/libvorbis-1.3.7/lib/lookup_data.h index 5de3cfdc..7935715a 100644 --- a/code/libvorbis-1.3.6/lib/lookup_data.h +++ b/code/libvorbis-1.3.7/lib/lookup_data.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/lpc.c b/code/libvorbis-1.3.7/lib/lpc.c similarity index 98% rename from code/libvorbis-1.3.6/lib/lpc.c rename to code/libvorbis-1.3.7/lib/lpc.c index 798f4cf0..877da47f 100644 --- a/code/libvorbis-1.3.6/lib/lpc.c +++ b/code/libvorbis-1.3.7/lib/lpc.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/lpc.h b/code/libvorbis-1.3.7/lib/lpc.h similarity index 94% rename from code/libvorbis-1.3.6/lib/lpc.h rename to code/libvorbis-1.3.7/lib/lpc.h index 9cc79451..4f59e6d3 100644 --- a/code/libvorbis-1.3.6/lib/lpc.h +++ b/code/libvorbis-1.3.7/lib/lpc.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/lsp.c b/code/libvorbis-1.3.7/lib/lsp.c similarity index 98% rename from code/libvorbis-1.3.6/lib/lsp.c rename to code/libvorbis-1.3.7/lib/lsp.c index 85880545..8afa305f 100644 --- a/code/libvorbis-1.3.6/lib/lsp.c +++ b/code/libvorbis-1.3.7/lib/lsp.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -15,9 +15,10 @@ The LSP generation code is taken (with minimal modification and a few bugfixes) from "On the Computation of the LSP Frequencies" by Joseph Rothweiler (see http://www.rothweiler.us for contact info). + The paper is available at: - http://www.myown1.com/joe/lsf + https://web.archive.org/web/20110810174000/http://home.myfairpoint.net/vzenxj75/myown1/joe/lsf/index.html ********************************************************************/ diff --git a/code/libvorbis-1.3.6/lib/lsp.h b/code/libvorbis-1.3.7/lib/lsp.h similarity index 94% rename from code/libvorbis-1.3.6/lib/lsp.h rename to code/libvorbis-1.3.7/lib/lsp.h index 8a8d10e9..68b38daf 100644 --- a/code/libvorbis-1.3.6/lib/lsp.h +++ b/code/libvorbis-1.3.7/lib/lsp.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/mapping0.c b/code/libvorbis-1.3.7/lib/mapping0.c similarity index 99% rename from code/libvorbis-1.3.6/lib/mapping0.c rename to code/libvorbis-1.3.7/lib/mapping0.c index ccb4493d..efa0fbcd 100644 --- a/code/libvorbis-1.3.6/lib/mapping0.c +++ b/code/libvorbis-1.3.7/lib/mapping0.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/masking.h b/code/libvorbis-1.3.7/lib/masking.h similarity index 99% rename from code/libvorbis-1.3.6/lib/masking.h rename to code/libvorbis-1.3.7/lib/masking.h index 955e18c7..7a196a37 100644 --- a/code/libvorbis-1.3.6/lib/masking.h +++ b/code/libvorbis-1.3.7/lib/masking.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/mdct.c b/code/libvorbis-1.3.7/lib/mdct.c similarity index 99% rename from code/libvorbis-1.3.6/lib/mdct.c rename to code/libvorbis-1.3.7/lib/mdct.c index f3f1ed80..2a0ff8d0 100644 --- a/code/libvorbis-1.3.6/lib/mdct.c +++ b/code/libvorbis-1.3.7/lib/mdct.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/mdct.h b/code/libvorbis-1.3.7/lib/mdct.h similarity index 96% rename from code/libvorbis-1.3.6/lib/mdct.h rename to code/libvorbis-1.3.7/lib/mdct.h index 3b8c9ba4..ceaea617 100644 --- a/code/libvorbis-1.3.6/lib/mdct.h +++ b/code/libvorbis-1.3.7/lib/mdct.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/misc.h b/code/libvorbis-1.3.7/lib/misc.h similarity index 96% rename from code/libvorbis-1.3.6/lib/misc.h rename to code/libvorbis-1.3.7/lib/misc.h index 13788445..eac5160e 100644 --- a/code/libvorbis-1.3.6/lib/misc.h +++ b/code/libvorbis-1.3.7/lib/misc.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/floor_all.h b/code/libvorbis-1.3.7/lib/modes/floor_all.h similarity index 99% rename from code/libvorbis-1.3.6/lib/modes/floor_all.h rename to code/libvorbis-1.3.7/lib/modes/floor_all.h index 20928aac..2e3d4a50 100644 --- a/code/libvorbis-1.3.6/lib/modes/floor_all.h +++ b/code/libvorbis-1.3.7/lib/modes/floor_all.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/psych_11.h b/code/libvorbis-1.3.7/lib/modes/psych_11.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/psych_11.h rename to code/libvorbis-1.3.7/lib/modes/psych_11.h index cc5eea24..9d8ed357 100644 --- a/code/libvorbis-1.3.6/lib/modes/psych_11.h +++ b/code/libvorbis-1.3.7/lib/modes/psych_11.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/psych_16.h b/code/libvorbis-1.3.7/lib/modes/psych_16.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/psych_16.h rename to code/libvorbis-1.3.7/lib/modes/psych_16.h index 477cb4d9..49cbf7c4 100644 --- a/code/libvorbis-1.3.6/lib/modes/psych_16.h +++ b/code/libvorbis-1.3.7/lib/modes/psych_16.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/psych_44.h b/code/libvorbis-1.3.7/lib/modes/psych_44.h similarity index 99% rename from code/libvorbis-1.3.6/lib/modes/psych_44.h rename to code/libvorbis-1.3.7/lib/modes/psych_44.h index 6c9eaa4e..d15509b7 100644 --- a/code/libvorbis-1.3.6/lib/modes/psych_44.h +++ b/code/libvorbis-1.3.7/lib/modes/psych_44.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/psych_8.h b/code/libvorbis-1.3.7/lib/modes/psych_8.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/psych_8.h rename to code/libvorbis-1.3.7/lib/modes/psych_8.h index 277db843..a19817f7 100644 --- a/code/libvorbis-1.3.6/lib/modes/psych_8.h +++ b/code/libvorbis-1.3.7/lib/modes/psych_8.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/residue_16.h b/code/libvorbis-1.3.7/lib/modes/residue_16.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/residue_16.h rename to code/libvorbis-1.3.7/lib/modes/residue_16.h index 3e05471c..15e161c8 100644 --- a/code/libvorbis-1.3.6/lib/modes/residue_16.h +++ b/code/libvorbis-1.3.7/lib/modes/residue_16.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/residue_44.h b/code/libvorbis-1.3.7/lib/modes/residue_44.h similarity index 99% rename from code/libvorbis-1.3.6/lib/modes/residue_44.h rename to code/libvorbis-1.3.7/lib/modes/residue_44.h index e89bc0e4..3f982695 100644 --- a/code/libvorbis-1.3.6/lib/modes/residue_44.h +++ b/code/libvorbis-1.3.7/lib/modes/residue_44.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/residue_44p51.h b/code/libvorbis-1.3.7/lib/modes/residue_44p51.h similarity index 99% rename from code/libvorbis-1.3.6/lib/modes/residue_44p51.h rename to code/libvorbis-1.3.7/lib/modes/residue_44p51.h index 7f33e250..8ac5f65e 100644 --- a/code/libvorbis-1.3.6/lib/modes/residue_44p51.h +++ b/code/libvorbis-1.3.7/lib/modes/residue_44p51.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/residue_44u.h b/code/libvorbis-1.3.7/lib/modes/residue_44u.h similarity index 99% rename from code/libvorbis-1.3.6/lib/modes/residue_44u.h rename to code/libvorbis-1.3.7/lib/modes/residue_44u.h index e55ac125..2f3595e4 100644 --- a/code/libvorbis-1.3.6/lib/modes/residue_44u.h +++ b/code/libvorbis-1.3.7/lib/modes/residue_44u.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/residue_8.h b/code/libvorbis-1.3.7/lib/modes/residue_8.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/residue_8.h rename to code/libvorbis-1.3.7/lib/modes/residue_8.h index ae123a27..b836f79c 100644 --- a/code/libvorbis-1.3.6/lib/modes/residue_8.h +++ b/code/libvorbis-1.3.7/lib/modes/residue_8.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_11.h b/code/libvorbis-1.3.7/lib/modes/setup_11.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/setup_11.h rename to code/libvorbis-1.3.7/lib/modes/setup_11.h index 0cbcaafc..5ade5dd1 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_11.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_11.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_16.h b/code/libvorbis-1.3.7/lib/modes/setup_16.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/setup_16.h rename to code/libvorbis-1.3.7/lib/modes/setup_16.h index d59ad70d..8b2daafa 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_16.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_16.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_22.h b/code/libvorbis-1.3.7/lib/modes/setup_22.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/setup_22.h rename to code/libvorbis-1.3.7/lib/modes/setup_22.h index bc38af96..eef5a4e7 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_22.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_22.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_32.h b/code/libvorbis-1.3.7/lib/modes/setup_32.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/setup_32.h rename to code/libvorbis-1.3.7/lib/modes/setup_32.h index f66a0bcd..f87cb767 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_32.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_32.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_44.h b/code/libvorbis-1.3.7/lib/modes/setup_44.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/setup_44.h rename to code/libvorbis-1.3.7/lib/modes/setup_44.h index a189b5fb..12d59280 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_44.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_44.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_44p51.h b/code/libvorbis-1.3.7/lib/modes/setup_44p51.h similarity index 96% rename from code/libvorbis-1.3.6/lib/modes/setup_44p51.h rename to code/libvorbis-1.3.7/lib/modes/setup_44p51.h index 3bde7b34..4d49173f 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_44p51.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_44p51.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_44u.h b/code/libvorbis-1.3.7/lib/modes/setup_44u.h similarity index 96% rename from code/libvorbis-1.3.6/lib/modes/setup_44u.h rename to code/libvorbis-1.3.7/lib/modes/setup_44u.h index 7ae3af6b..2dd8bf70 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_44u.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_44u.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_8.h b/code/libvorbis-1.3.7/lib/modes/setup_8.h similarity index 97% rename from code/libvorbis-1.3.6/lib/modes/setup_8.h rename to code/libvorbis-1.3.7/lib/modes/setup_8.h index 75025568..16b02e01 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_8.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_8.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/modes/setup_X.h b/code/libvorbis-1.3.7/lib/modes/setup_X.h similarity index 98% rename from code/libvorbis-1.3.6/lib/modes/setup_X.h rename to code/libvorbis-1.3.7/lib/modes/setup_X.h index 2229a5ef..27807c10 100644 --- a/code/libvorbis-1.3.6/lib/modes/setup_X.h +++ b/code/libvorbis-1.3.7/lib/modes/setup_X.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/os.h b/code/libvorbis-1.3.7/lib/os.h similarity index 93% rename from code/libvorbis-1.3.6/lib/os.h rename to code/libvorbis-1.3.7/lib/os.h index 416a401d..9ded7358 100644 --- a/code/libvorbis-1.3.6/lib/os.h +++ b/code/libvorbis-1.3.7/lib/os.h @@ -8,7 +8,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -60,7 +60,7 @@ void *_alloca(size_t size); # define FAST_HYPOT hypot #endif -#endif +#endif /* _V_IFDEFJAIL_H_ */ #ifdef HAVE_ALLOCA_H # include @@ -80,7 +80,7 @@ void *_alloca(size_t size); /* Special i386 GCC implementation */ -#if defined(__i386__) && defined(__GNUC__) && !defined(__BEOS__) +#if defined(__i386__) && defined(__GNUC__) && !defined(__BEOS__) && !defined(__SSE2_MATH__) # define VORBIS_FPU_CONTROL /* both GCC and MSVC are kinda stupid about rounding/casting to int. Because of encapsulation constraints (GCC can't see inside the asm @@ -119,8 +119,7 @@ static inline int vorbis_ftoi(double f){ /* yes, double! Otherwise, /* MSVC inline assembly. 32 bit only; inline ASM isn't implemented in the * 64 bit compiler and doesn't work on arm. */ -#if defined(_MSC_VER) && !defined(_WIN64) && \ - !defined(_WIN32_WCE) && !defined(_M_ARM) +#if defined(_MSC_VER) && defined(_M_IX86) && !defined(_WIN32_WCE) # define VORBIS_FPU_CONTROL typedef ogg_int16_t vorbis_fpu_control; @@ -147,7 +146,7 @@ static __inline void vorbis_fpu_restore(vorbis_fpu_control fpu){ /* Optimized code path for x86_64 builds. Uses SSE2 intrinsics. This can be done safely because all x86_64 CPUs supports SSE2. */ -#if (defined(_MSC_VER) && defined(_WIN64)) || (defined(__GNUC__) && defined (__x86_64__)) +#if (defined(_MSC_VER) && defined(_M_X64)) || (defined(__GNUC__) && defined (__SSE2_MATH__)) # define VORBIS_FPU_CONTROL typedef ogg_int16_t vorbis_fpu_control; @@ -174,7 +173,7 @@ static __inline void vorbis_fpu_restore(vorbis_fpu_control fpu){ typedef int vorbis_fpu_control; -static int vorbis_ftoi(double f){ +STIN int vorbis_ftoi(double f){ /* Note: MSVC and GCC (at least on some systems) round towards zero, thus, the floor() call is required to ensure correct roudning of negative numbers */ diff --git a/code/libvorbis-1.3.6/lib/psy.c b/code/libvorbis-1.3.7/lib/psy.c similarity index 98% rename from code/libvorbis-1.3.6/lib/psy.c rename to code/libvorbis-1.3.7/lib/psy.c index 422c6f1e..036b094a 100644 --- a/code/libvorbis-1.3.6/lib/psy.c +++ b/code/libvorbis-1.3.7/lib/psy.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2010 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -599,11 +599,12 @@ static void bark_noise_hybridmp(int n,const long *b, XY[i] = tXY; } - for (i = 0, x = 0.f;; i++, x += 1.f) { + for (i = 0, x = 0.f; i < n; i++, x += 1.f) { lo = b[i] >> 16; - if( lo>=0 ) break; hi = b[i] & 0xffff; + if( lo>=0 || -lo>=n ) break; + if( hi>=n ) break; tN = N[hi] + N[-lo]; tX = X[hi] - X[-lo]; @@ -615,17 +616,17 @@ static void bark_noise_hybridmp(int n,const long *b, B = tN * tXY - tX * tY; D = tN * tXX - tX * tX; R = (A + x * B) / D; - if (R < 0.f) - R = 0.f; + if (R < 0.f) R = 0.f; noise[i] = R - offset; } - for ( ;; i++, x += 1.f) { + for ( ; i < n; i++, x += 1.f) { lo = b[i] >> 16; hi = b[i] & 0xffff; - if(hi>=n)break; + if( lo<0 || lo>=n ) break; + if( hi>=n ) break; tN = N[hi] - N[lo]; tX = X[hi] - X[lo]; @@ -641,6 +642,7 @@ static void bark_noise_hybridmp(int n,const long *b, noise[i] = R - offset; } + for ( ; i < n; i++, x += 1.f) { R = (A + x * B) / D; @@ -651,10 +653,11 @@ static void bark_noise_hybridmp(int n,const long *b, if (fixed <= 0) return; - for (i = 0, x = 0.f;; i++, x += 1.f) { + for (i = 0, x = 0.f; i < n; i++, x += 1.f) { hi = i + fixed / 2; lo = hi - fixed; - if(lo>=0)break; + if ( hi>=n ) break; + if ( lo>=0 ) break; tN = N[hi] + N[-lo]; tX = X[hi] - X[-lo]; @@ -670,11 +673,12 @@ static void bark_noise_hybridmp(int n,const long *b, if (R - offset < noise[i]) noise[i] = R - offset; } - for ( ;; i++, x += 1.f) { + for ( ; i < n; i++, x += 1.f) { hi = i + fixed / 2; lo = hi - fixed; - if(hi>=n)break; + if ( hi>=n ) break; + if ( lo<0 ) break; tN = N[hi] - N[lo]; tX = X[hi] - X[lo]; diff --git a/code/libvorbis-1.3.6/lib/psy.h b/code/libvorbis-1.3.7/lib/psy.h similarity index 98% rename from code/libvorbis-1.3.6/lib/psy.h rename to code/libvorbis-1.3.7/lib/psy.h index ab2534db..d9a04e8b 100644 --- a/code/libvorbis-1.3.6/lib/psy.h +++ b/code/libvorbis-1.3.7/lib/psy.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.7/lib/psytune.c b/code/libvorbis-1.3.7/lib/psytune.c new file mode 100644 index 00000000..67223e51 --- /dev/null +++ b/code/libvorbis-1.3.7/lib/psytune.c @@ -0,0 +1,523 @@ +/******************************************************************** + * * + * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE. * + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * + * * + * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * + * by the Xiph.Org Foundation https://xiph.org/ * + * * + ******************************************************************** + + function: simple utility that runs audio through the psychoacoustics + without encoding + + ********************************************************************/ + +/* NB: this is dead code, retained purely for doc and reference value + don't try to compile it */ + +#include +#include +#include +#include + +#include "vorbis/codec.h" +#include "codec_internal.h" +#include "os.h" +#include "misc.h" +#include "psy.h" +#include "mdct.h" +#include "smallft.h" +#include "window.h" +#include "scales.h" +#include "lpc.h" +#include "lsp.h" +#include "masking.h" +#include "registry.h" + +static vorbis_info_psy_global _psy_set0G={ + 0, /* decaydBpms */ + 8, /* lines per eighth octave */ + + /* thresh sample period, preecho clamp trigger threshhold, range, minenergy */ + 256, {26.f,26.f,26.f,30.f}, {-90.f,-90.f,-90.f,-90.f}, -90.f, + -6.f, + + 0, + + 0., + 0., +}; + +static vp_part _vp_part0[]={ + { 1,9e10f, 9e10f, 1.f,9999.f}, + { 9999, .75f, 9e10f, .5f,9999.f}, +/*{ 9999, 1.5f, 9e10f, .5f,9999.f},*/ + { 18,9e10f, 9e10f, .5f, 30.f}, + { 9999,9e10f, 9e10f, .5f, 30.f} +}; + +static vp_couple _vp_couple0[]={ + { 1, {9e10f,9e10f,0}, { 0.f, 0.f,0}, { 0.f, 0.f,0}, {0.f,0.f,0}}, + { 18, {9e10f,9e10f,0}, { 0.f, 0.f,0}, { 0.f, 0.f,0}, {0.f,0.f,0}}, + { 9999, {9e10f,9e10f,0}, { 0.f, 9e10f,0}, { 0.f,22.f,1}, {0.f,0.f,0}} +}; + +static vorbis_info_psy _psy_set0={ + ATH_Bark_dB_lineaggressive, + + -100.f, + -140.f, + 6.f, /* floor master att */ + + /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 */ + /* x: 63 88 125 175 250 350 500 700 1k 1.4k 2k 2.8k 4k 5.6k 8k 11.5k 16k Hz */ + /* y: 0 10 20 30 40 50 60 70 80 90 100 dB */ + 1, /* tonemaskp */ + 0.f, /* tone master att */ + /* 0 10 20 30 40 50 60 70 80 90 100 */ + { + {-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f}, /*63*/ + {-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f}, /*88*/ + {-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f,-999.f}, /*125*/ + + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*175*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*250*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*350*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*500*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*700*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*1000*/ + {-30.f,-30.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*1400*/ + {-40.f,-40.f,-40.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*2000*/ + {-40.f,-40.f,-40.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*2800*/ + {-40.f,-40.f,-40.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*4000*/ + + {-30.f,-35.f,-35.f,-40.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*5600*/ + + {-30.f,-30.f,-33.f,-35.f,-40.f,-50.f,-60.f,-70.f,-80.f,-90.f,-100.f}, /*8000*/ + {-30.f,-30.f,-33.f,-35.f,-40.f,-45.f,-50.f,-60.f,-70.f,-85.f,-100.f}, /*11500*/ + {-24.f,-24.f,-26.f,-32.f,-32.f,-42.f,-50.f,-60.f,-70.f,-85.f,-100.f}, /*16000*/ + + }, + + 1,/* peakattp */ + {{-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*63*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*88*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*125*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*175*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*250*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*350*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*500*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*700*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*1000*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*1400*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*2000*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*2800*/ + {-14.f,-20.f,-20.f,-20.f,-26.f,-32.f,-40.f,-40.f,-40.f,-40.f,-40.f},/*4000*/ + {-10.f,-12.f,-14.f,-16.f,-16.f,-20.f,-24.f,-30.f,-32.f,-40.f,-40.f},/*5600*/ + {-10.f,-12.f,-14.f,-16.f,-16.f,-20.f,-24.f,-30.f,-32.f,-40.f,-40.f},/*8000*/ + {-10.f,-10.f,-10.f,-12.f,-14.f,-18.f,-22.f,-28.f,-32.f,-40.f,-40.f},/*11500*/ + {-10.f,-10.f,-10.f,-12.f,-14.f,-18.f,-22.f,-28.f,-32.f,-40.f,-40.f},/*16000*/ + }, + + 1,/*noisemaskp */ + -10.f, /* suppress any noise curve over maxspec+n */ + .5f, /* low window */ + .5f, /* high window */ + 10, + 10, + 25, + {.000f, 0.f, /*63*/ + .000f, 0.f, /*88*/ + .000f, 0.f, /*125*/ + .000f, 0.f, /*175*/ + .000f, 0.f, /*250*/ + .000f, 0.f, /*350*/ + .000f, 0.f, /*500*/ + .000f, 0.f, /*700*/ + .000f, 0.f, /*1000*/ + .300f, 0.f, /*1400*/ + .300f, 0.f, /*2000*/ + .300f, 0.f, /*2800*/ + .500f, 0.f, /*4000*/ + .700f, 0.f, /*5600*/ + .850f, 0.f, /*8000*/ + .900f, 0.f, /*11500*/ + .900f, 1.f, /*16000*/ + }, + + 95.f, /* even decade + 5 is important; saves an rint() later in a + tight loop) */ + -44., + + 32, + _vp_part0,_vp_couple0 +}; + +static vorbis_info_floor1 _floor_set0={1, + {0}, + + {32}, + {0}, + {0}, + {{-1}}, + + 2, + {0,1024, + + 88,31,243, + + 14,54,143,460, + + 6,3,10, 22,18,26, 41,36,47, + 69,61,78, 112,99,126, 185,162,211, + 329,282,387, 672,553,825 + }, + + 60,30,400, + 20,8,1,18., + 20,600, + 960}; + + +static vorbis_info_mapping0 mapping_info={1,{0,1},{0},{0},{0},0, 1, {0},{1}}; +static codec_setup_info codec_setup0={ {0,0}, + 1,1,1,1,1,0,1, + {NULL}, + {0},{&mapping_info}, + {0},{NULL}, + {1},{&_floor_set0}, + {2},{NULL}, + {NULL}, + {&_psy_set0}, + &_psy_set0G}; + +static int noisy=0; +void analysis(char *base,int i,float *v,int n,int bark,int dB){ + if(noisy){ + int j; + FILE *of; + char buffer[80]; + sprintf(buffer,"%s_%d.m",base,i); + of=fopen(buffer,"w"); + + for(j=0;jlook(NULL,NULL,&_floor_set0); + + /* we cheat on the WAV header; we just bypass 44 bytes and never + verify that it matches 16bit/stereo/44.1kHz. */ + + fread(buffer,1,44,stdin); + fwrite(buffer,1,44,stdout); + memset(buffer,0,framesize*2); + + analysis("window",0,window,framesize,0,0); + + fprintf(stderr,"Processing for frame size %d...\n",framesize); + + while(!eos){ + long bytes=fread(buffer2,1,framesize*2,stdin); + if(bytes>1]=todB(&temp); + if(temp>local_ampmax[i])local_ampmax[i]=temp; + } + if(local_ampmax[i]>ampmax)ampmax=local_ampmax[i]; + + mdct_forward(&m_look,pcm[i],mdct); + for(j=0;jforward(&vb,floor_look, + mdct, + logmdct, + mask, + logmax, + + flr[i]); + } + + _vp_remove_floor(&p_look, + pg_look, + logmdct, + mdct, + flr[i], + pcm[i], + local_ampmax[i]); + + for(j=0;j1500) + fprintf(stderr,"%ld ",frameno+i); + + analysis("res",frameno+i,pcm[i],framesize/2,1,0); + analysis("codedflr",frameno+i,flr[i],framesize/2,1,1); + } + + /* residue prequantization */ + _vp_partition_prequant(&p_look, + &vi, + pcm, + nonzero); + + for(i=0;i<2;i++) + analysis("quant",frameno+i,pcm[i],framesize/2,1,0); + + /* channel coupling / stereo quantization */ + + _vp_couple(&p_look, + &mapping_info, + pcm, + nonzero); + + for(i=0;i<2;i++) + analysis("coupled",frameno+i,pcm[i],framesize/2,1,0); + + /* decoupling */ + for(i=mapping_info.coupling_steps-1;i>=0;i--){ + float *pcmM=pcm[mapping_info.coupling_mag[i]]; + float *pcmA=pcm[mapping_info.coupling_ang[i]]; + + for(j=0;j0) + if(ang>0){ + pcmM[j]=mag; + pcmA[j]=mag-ang; + }else{ + pcmA[j]=mag; + pcmM[j]=mag+ang; + } + else + if(ang>0){ + pcmM[j]=mag; + pcmA[j]=mag+ang; + }else{ + pcmA[j]=mag; + pcmM[j]=mag-ang; + } + } + } + + for(i=0;i<2;i++) + analysis("decoupled",frameno+i,pcm[i],framesize/2,1,0); + + for(i=0;i<2;i++){ + float amp; + + for(j=0;j32767){ + if(!flag)fprintf(stderr,"clipping in frame %ld ",frameno+i); + flag=1; + val=32767; + } + if(val<-32768){ + if(!flag)fprintf(stderr,"clipping in frame %ld ",frameno+i); + flag=1; + val=-32768; + } + ptr[0]=val&0xff; + ptr[1]=(val>>8)&0xff; + ptr+=4; + } + } + + fprintf(stderr,"*"); + fwrite(buffer,1,framesize*2,stdout); + memmove(buffer,buffer2,framesize*2); + + for(i=0;i<2;i++){ + for(j=0,k=framesize/2;j #endif diff --git a/code/libvorbis-1.3.6/lib/scales.h b/code/libvorbis-1.3.7/lib/scales.h similarity index 97% rename from code/libvorbis-1.3.6/lib/scales.h rename to code/libvorbis-1.3.7/lib/scales.h index 18bc4e75..3c2ae48d 100644 --- a/code/libvorbis-1.3.6/lib/scales.h +++ b/code/libvorbis-1.3.7/lib/scales.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/sharedbook.c b/code/libvorbis-1.3.7/lib/sharedbook.c similarity index 98% rename from code/libvorbis-1.3.6/lib/sharedbook.c rename to code/libvorbis-1.3.7/lib/sharedbook.c index 4545d4f4..62a9a00a 100644 --- a/code/libvorbis-1.3.6/lib/sharedbook.c +++ b/code/libvorbis-1.3.7/lib/sharedbook.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -50,7 +50,7 @@ long _float32_pack(float val){ sign=0x80000000; val= -val; } - exp= floor(log(val)/log(2.f)+.001); //+epsilon + exp= floor(log(val)/log(2.f)+.001); /* +epsilon */ mant=rint(ldexp(val,(VQ_FMAN-1)-exp)); exp=(exp+VQ_FEXP_BIAS)<>VQ_FMAN; if(sign)mant= -mant; - return(ldexp(mant,exp-(VQ_FMAN-1)-VQ_FEXP_BIAS)); + exp=exp-(VQ_FMAN-1)-VQ_FEXP_BIAS; + /* clamp excessive exponent values */ + if (exp>63){ + exp=63; + } + if (exp<-63){ + exp=-63; + } + return(ldexp(mant,exp)); } /* given a list of word lengths, generate a list of codewords. Works @@ -294,7 +302,7 @@ int vorbis_book_init_encode(codebook *c,const static_codebook *s){ c->used_entries=s->entries; c->dim=s->dim; c->codelist=_make_words(s->lengthlist,s->entries,0); - //c->valuelist=_book_unquantize(s,s->entries,NULL); + /* c->valuelist=_book_unquantize(s,s->entries,NULL); */ c->quantvals=_book_maptype1_quantvals(s); c->minval=(int)rint(_float32_unpack(s->q_min)); c->delta=(int)rint(_float32_unpack(s->q_delta)); @@ -573,6 +581,7 @@ void run_test(static_codebook *b,float *comp){ exit(1); } } + free(out); } int main(){ diff --git a/code/libvorbis-1.3.6/lib/smallft.c b/code/libvorbis-1.3.7/lib/smallft.c similarity index 99% rename from code/libvorbis-1.3.6/lib/smallft.c rename to code/libvorbis-1.3.7/lib/smallft.c index 6d528af4..4ffabab4 100644 --- a/code/libvorbis-1.3.6/lib/smallft.c +++ b/code/libvorbis-1.3.7/lib/smallft.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/smallft.h b/code/libvorbis-1.3.7/lib/smallft.h similarity index 94% rename from code/libvorbis-1.3.6/lib/smallft.h rename to code/libvorbis-1.3.7/lib/smallft.h index 9e867c67..02fe8f9c 100644 --- a/code/libvorbis-1.3.6/lib/smallft.h +++ b/code/libvorbis-1.3.7/lib/smallft.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/synthesis.c b/code/libvorbis-1.3.7/lib/synthesis.c similarity index 98% rename from code/libvorbis-1.3.6/lib/synthesis.c rename to code/libvorbis-1.3.7/lib/synthesis.c index 5f6092c3..3e2d6812 100644 --- a/code/libvorbis-1.3.6/lib/synthesis.c +++ b/code/libvorbis-1.3.7/lib/synthesis.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.7/lib/tone.c b/code/libvorbis-1.3.7/lib/tone.c new file mode 100644 index 00000000..5b8b0206 --- /dev/null +++ b/code/libvorbis-1.3.7/lib/tone.c @@ -0,0 +1,54 @@ +#include +#include +#include +#include + +void usage(){ + fprintf(stderr,"tone ,[] [,[]...]\n"); + exit(1); +} + +int main (int argc,char *argv[]){ + int i,j; + double *f; + double *amp; + + if(argc<2)usage(); + + f=alloca(sizeof(*f)*(argc-1)); + amp=alloca(sizeof(*amp)*(argc-1)); + + i=0; + while(argv[i+1]){ + char *pos=strchr(argv[i+1],','); + + f[i]=atof(argv[i+1]); + if(pos) + amp[i]=atof(pos+1)*32767.f; + else + amp[i]=32767.f; + + fprintf(stderr,"%g Hz, %g amp\n",f[i],amp[i]); + + i++; + } + + for(i=0;i<44100*10;i++){ + float val=0; + int ival; + for(j=0;j32767.f)ival=32767.f; + if(ival<-32768.f)ival=-32768.f; + + fprintf(stdout,"%c%c%c%c", + (char)(ival&0xff), + (char)((ival>>8)&0xff), + (char)(ival&0xff), + (char)((ival>>8)&0xff)); + } + return(0); +} + diff --git a/code/libvorbis-1.3.6/lib/vorbisenc.c b/code/libvorbis-1.3.7/lib/vorbisenc.c similarity index 99% rename from code/libvorbis-1.3.6/lib/vorbisenc.c rename to code/libvorbis-1.3.7/lib/vorbisenc.c index 4a4607cb..cf3806a6 100644 --- a/code/libvorbis-1.3.6/lib/vorbisenc.c +++ b/code/libvorbis-1.3.7/lib/vorbisenc.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -684,6 +684,7 @@ int vorbis_encode_setup_init(vorbis_info *vi){ highlevel_encode_setup *hi=&ci->hi; if(ci==NULL)return(OV_EINVAL); + if(vi->channels<1||vi->channels>255)return(OV_EINVAL); if(!hi->impulse_block_p)i0=1; /* too low/high an ATH floater is nonsensical, but doesn't break anything */ @@ -1210,7 +1211,7 @@ int vorbis_encode_ctl(vorbis_info *vi,int number,void *arg){ hi->req, hi->managed, &new_base); - if(!hi->setup)return OV_EIMPL; + if(!new_template)return OV_EIMPL; hi->setup=new_template; hi->base_setting=new_base; vorbis_encode_setup_setting(vi,vi->channels,vi->rate); diff --git a/code/libvorbis-1.3.6/lib/vorbisfile.c b/code/libvorbis-1.3.7/lib/vorbisfile.c similarity index 99% rename from code/libvorbis-1.3.6/lib/vorbisfile.c rename to code/libvorbis-1.3.7/lib/vorbisfile.c index f879d31d..9219c2f2 100644 --- a/code/libvorbis-1.3.6/lib/vorbisfile.c +++ b/code/libvorbis-1.3.7/lib/vorbisfile.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2015 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** @@ -264,6 +264,10 @@ static ogg_int64_t _get_prev_page_serial(OggVorbis_File *vf, ogg_int64_t begin, } } } + /*We started from the beginning of the stream and found nothing. + This should be impossible unless the contents of the stream changed out + from under us after we read from it.*/ + if(!begin&&vf->offset<0)return OV_EBADLINK; } /* we're not interested in the page... just the serialno and granpos. */ @@ -1230,7 +1234,6 @@ double ov_time_total(OggVorbis_File *vf,int i){ int ov_raw_seek(OggVorbis_File *vf,ogg_int64_t pos){ ogg_stream_state work_os; - int ret; if(vf->ready_stateseekable) @@ -1253,8 +1256,12 @@ int ov_raw_seek(OggVorbis_File *vf,ogg_int64_t pos){ vf->current_serialno); /* must set serialno */ vorbis_synthesis_restart(&vf->vd); - ret=_seek_helper(vf,pos); - if(ret)goto seek_error; + if(_seek_helper(vf,pos)) { + /* dump the machine so we're in a known state */ + vf->pcm_offset=-1; + _decode_clear(vf); + return OV_EBADLINK; + } /* we need to make sure the pcm_offset is set, but we don't want to advance the raw cursor past good packets just to get to the first @@ -1388,13 +1395,6 @@ int ov_raw_seek(OggVorbis_File *vf,ogg_int64_t pos){ vf->bittrack=0.f; vf->samptrack=0.f; return(0); - - seek_error: - /* dump the machine so we're in a known state */ - vf->pcm_offset=-1; - ogg_stream_clear(&work_os); - _decode_clear(vf); - return OV_EBADLINK; } /* Page granularity seek (faster than sample granularity because we @@ -1910,7 +1910,7 @@ vorbis_comment *ov_comment(OggVorbis_File *vf,int link){ } } -static int host_is_big_endian(void) { +static int host_is_big_endian() { ogg_int32_t pattern = 0xfeedface; /* deadbeef */ unsigned char *bytewise = (unsigned char *)&pattern; if (bytewise[0] == 0xfe) return 1; @@ -1964,6 +1964,7 @@ long ov_read_filter(OggVorbis_File *vf,char *buffer,int length, long samples; if(vf->ready_stateready_state==INITSET){ @@ -1989,6 +1990,8 @@ long ov_read_filter(OggVorbis_File *vf,char *buffer,int length, long channels=ov_info(vf,-1)->channels; long bytespersample=word * channels; vorbis_fpu_control fpu; + + if(channels<1||channels>255)return(OV_EINVAL); if(samples>length/bytespersample)samples=length/bytespersample; if(samples <= 0) diff --git a/code/libvorbis-1.3.6/lib/window.c b/code/libvorbis-1.3.7/lib/window.c similarity index 99% rename from code/libvorbis-1.3.6/lib/window.c rename to code/libvorbis-1.3.7/lib/window.c index b3b7ce01..2151b278 100644 --- a/code/libvorbis-1.3.6/lib/window.c +++ b/code/libvorbis-1.3.7/lib/window.c @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2009 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/libvorbis-1.3.6/lib/window.h b/code/libvorbis-1.3.7/lib/window.h similarity index 94% rename from code/libvorbis-1.3.6/lib/window.h rename to code/libvorbis-1.3.7/lib/window.h index 6ac26074..33d83f85 100644 --- a/code/libvorbis-1.3.6/lib/window.h +++ b/code/libvorbis-1.3.7/lib/window.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2007 * - * by the Xiph.Org Foundation http://www.xiph.org/ * + * by the Xiph.Org Foundation https://xiph.org/ * * * ******************************************************************** diff --git a/code/opus-1.2.1/celt/arm/arm2gnu.pl b/code/opus-1.2.1/celt/arm/arm2gnu.pl deleted file mode 100755 index a2895f74..00000000 --- a/code/opus-1.2.1/celt/arm/arm2gnu.pl +++ /dev/null @@ -1,353 +0,0 @@ -#!/usr/bin/perl -# Copyright (C) 2002-2013 Xiph.org Foundation -# -# Redistribution and use in source and binary forms, with or without -# modification, are permitted provided that the following conditions -# are met: -# -# - Redistributions of source code must retain the above copyright -# notice, this list of conditions and the following disclaimer. -# -# - Redistributions in binary form must reproduce the above copyright -# notice, this list of conditions and the following disclaimer in the -# documentation and/or other materials provided with the distribution. -# -# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -# ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER -# OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, -# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, -# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR -# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF -# LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING -# NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -my $bigend; # little/big endian -my $nxstack; -my $apple = 0; -my $symprefix = ""; - -$nxstack = 0; - -eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}' - if $running_under_some_shell; - -while ($ARGV[0] =~ /^-/) { - $_ = shift; - last if /^--$/; - if (/^-n$/) { - $nflag++; - next; - } - if (/^--apple$/) { - $apple = 1; - $symprefix = "_"; - next; - } - die "I don't recognize this switch: $_\\n"; -} -$printit++ unless $nflag; - -$\ = "\n"; # automatically add newline on print -$n=0; - -$thumb = 0; # ARM mode by default, not Thumb. -@proc_stack = (); - -printf (" .syntax unified\n"); - -LINE: -while (<>) { - - # For ADRLs we need to add a new line after the substituted one. - $addPadding = 0; - - # First, we do not dare to touch *anything* inside double quotes, do we? - # Second, if you want a dollar character in the string, - # insert two of them -- that's how ARM C and assembler treat strings. - s/^([A-Za-z_]\w*)[ \t]+DCB[ \t]*\"/$1: .ascii \"/ && do { s/\$\$/\$/g; next }; - s/\bDCB\b[ \t]*\"/.ascii \"/ && do { s/\$\$/\$/g; next }; - s/^(\S+)\s+RN\s+(\S+)/$1 .req r$2/ && do { s/\$\$/\$/g; next }; - # If there's nothing on a line but a comment, don't try to apply any further - # substitutions (this is a cheap hack to avoid mucking up the license header) - s/^([ \t]*);/$1@/ && do { s/\$\$/\$/g; next }; - # If substituted -- leave immediately ! - - s/@/,:/; - s/;/@/; - while ( /@.*'/ ) { - s/(@.*)'/$1/g; - } - s/\{FALSE\}/0/g; - s/\{TRUE\}/1/g; - s/\{(\w\w\w\w+)\}/$1/g; - s/\bINCLUDE[ \t]*([^ \t\n]+)/.include \"$1\"/; - s/\bGET[ \t]*([^ \t\n]+)/.include \"${ my $x=$1; $x =~ s|\.s|-gnu.S|; \$x }\"/; - s/\bIMPORT\b/.extern/; - s/\bEXPORT\b\s*/.global $symprefix/; - s/^(\s+)\[/$1IF/; - s/^(\s+)\|/$1ELSE/; - s/^(\s+)\]/$1ENDIF/; - s/IF *:DEF:/ .ifdef/; - s/IF *:LNOT: *:DEF:/ .ifndef/; - s/ELSE/ .else/; - s/ENDIF/ .endif/; - - if( /\bIF\b/ ) { - s/\bIF\b/ .if/; - s/=/==/; - } - if ( $n == 2) { - s/\$/\\/g; - } - if ($n == 1) { - s/\$//g; - s/label//g; - $n = 2; - } - if ( /MACRO/ ) { - s/MACRO *\n/.macro/; - $n=1; - } - if ( /\bMEND\b/ ) { - s/\bMEND\b/.endm/; - $n=0; - } - - # ".rdata" doesn't work in 'as' version 2.13.2, as it is ".rodata" there. - # - if ( /\bAREA\b/ ) { - my $align; - $align = "2"; - if ( /ALIGN=(\d+)/ ) { - $align = $1; - } - if ( /CODE/ ) { - $nxstack = 1; - } - s/^(.+)CODE(.+)READONLY(.*)/ .text/; - s/^(.+)DATA(.+)READONLY(.*)/ .section .rdata/; - s/^(.+)\|\|\.data\|\|(.+)/ .data/; - s/^(.+)\|\|\.bss\|\|(.+)/ .bss/; - s/$/; .p2align $align/; - # Enable NEON instructions but don't produce a binary that requires - # ARMv7. RVCT does not have equivalent directives, so we just do this - # for all CODE areas. - if ( /.text/ ) { - # Separating .arch, .fpu, etc., by semicolons does not work (gas - # thinks the semicolon is part of the arch name, even when there's - # whitespace separating them). Sadly this means our line numbers - # won't match the original source file (we could use the .line - # directive, which is documented to be obsolete, but then gdb will - # show the wrong line in the translated source file). - s/$/; .arch armv7-a\n .fpu neon\n .object_arch armv4t/ unless ($apple); - } - } - - s/\|\|\.constdata\$(\d+)\|\|/.L_CONST$1/; # ||.constdata$3|| - s/\|\|\.bss\$(\d+)\|\|/.L_BSS$1/; # ||.bss$2|| - s/\|\|\.data\$(\d+)\|\|/.L_DATA$1/; # ||.data$2|| - s/\|\|([a-zA-Z0-9_]+)\@([a-zA-Z0-9_]+)\|\|/@ $&/; - s/^(\s+)\%(\s)/ .space $1/; - - s/\|(.+)\.(\d+)\|/\.$1_$2/; # |L80.123| -> .L80_123 - s/\bCODE32\b/.code 32/ && do {$thumb = 0}; - s/\bCODE16\b/.code 16/ && do {$thumb = 1}; - if (/\bPROC\b/) - { - my $prefix; - my $proc; - /^([A-Za-z_\.]\w+)\b/; - $proc = $1; - $prefix = ""; - if ($proc) - { - $prefix = $prefix.sprintf("\t.type\t%s, %%function", $proc) unless ($apple); - # Make sure we $prefix isn't empty here (for the $apple case). - # We handle mangling the label here, make sure it doesn't match - # the label handling below (if $prefix would be empty). - $prefix = $prefix."; "; - push(@proc_stack, $proc); - s/^[A-Za-z_\.]\w+/$symprefix$&:/; - } - $prefix = $prefix."\t.thumb_func; " if ($thumb); - s/\bPROC\b/@ $&/; - $_ = $prefix.$_; - } - s/^(\s*)(S|Q|SH|U|UQ|UH)ASX\b/$1$2ADDSUBX/; - s/^(\s*)(S|Q|SH|U|UQ|UH)SAX\b/$1$2SUBADDX/; - if (/\bENDP\b/) - { - my $proc; - s/\bENDP\b/@ $&/; - $proc = pop(@proc_stack); - $_ = "\t.size $proc, .-$proc".$_ if ($proc && !$apple); - } - s/\bSUBT\b/@ $&/; - s/\bDATA\b/@ $&/; # DATA directive is deprecated -- Asm guide, p.7-25 - s/\bKEEP\b/@ $&/; - s/\bEXPORTAS\b/@ $&/; - s/\|\|(.)+\bEQU\b/@ $&/; - s/\|\|([\w\$]+)\|\|/$1/; - s/\bENTRY\b/@ $&/; - s/\bASSERT\b/@ $&/; - s/\bGBLL\b/@ $&/; - s/\bGBLA\b/@ $&/; - s/^\W+OPT\b/@ $&/; - s/:OR:/|/g; - s/:SHL:/<>/g; - s/:AND:/&/g; - s/:LAND:/&&/g; - s/CPSR/cpsr/; - s/SPSR/spsr/; - s/ALIGN$/.balign 4/; - s/ALIGN\s+([0-9x]+)$/.balign $1/; - s/psr_cxsf/psr_all/; - s/LTORG/.ltorg/; - s/^([A-Za-z_]\w*)[ \t]+EQU/ .set $1,/; - s/^([A-Za-z_]\w*)[ \t]+SETL/ .set $1,/; - s/^([A-Za-z_]\w*)[ \t]+SETA/ .set $1,/; - s/^([A-Za-z_]\w*)[ \t]+\*/ .set $1,/; - - # {PC} + 0xdeadfeed --> . + 0xdeadfeed - s/\{PC\} \+/ \. +/; - - # Single hex constant on the line ! - # - # >>> NOTE <<< - # Double-precision floats in gcc are always mixed-endian, which means - # bytes in two words are little-endian, but words are big-endian. - # So, 0x0000deadfeed0000 would be stored as 0x0000dead at low address - # and 0xfeed0000 at high address. - # - s/\bDCFD\b[ \t]+0x([a-fA-F0-9]{8})([a-fA-F0-9]{8})/.long 0x$1, 0x$2/; - # Only decimal constants on the line, no hex ! - s/\bDCFD\b[ \t]+([0-9\.\-]+)/.double $1/; - - # Single hex constant on the line ! -# s/\bDCFS\b[ \t]+0x([a-f0-9]{8})([a-f0-9]{8})/.long 0x$1, 0x$2/; - # Only decimal constants on the line, no hex ! -# s/\bDCFS\b[ \t]+([0-9\.\-]+)/.double $1/; - s/\bDCFS[ \t]+0x/.word 0x/; - s/\bDCFS\b/.float/; - - s/^([A-Za-z_]\w*)[ \t]+DCD/$1 .word/; - s/\bDCD\b/.word/; - s/^([A-Za-z_]\w*)[ \t]+DCW/$1 .short/; - s/\bDCW\b/.short/; - s/^([A-Za-z_]\w*)[ \t]+DCB/$1 .byte/; - s/\bDCB\b/.byte/; - s/^([A-Za-z_]\w*)[ \t]+\%/.comm $1,/; - s/^[A-Za-z_\.]\w+/$&:/; - s/^(\d+)/$1:/; - s/\%(\d+)/$1b_or_f/; - s/\%[Bb](\d+)/$1b/; - s/\%[Ff](\d+)/$1f/; - s/\%[Ff][Tt](\d+)/$1f/; - s/&([\dA-Fa-f]+)/0x$1/; - if ( /\b2_[01]+\b/ ) { - s/\b2_([01]+)\b/conv$1&&&&/g; - while ( /[01][01][01][01]&&&&/ ) { - s/0000&&&&/&&&&0/g; - s/0001&&&&/&&&&1/g; - s/0010&&&&/&&&&2/g; - s/0011&&&&/&&&&3/g; - s/0100&&&&/&&&&4/g; - s/0101&&&&/&&&&5/g; - s/0110&&&&/&&&&6/g; - s/0111&&&&/&&&&7/g; - s/1000&&&&/&&&&8/g; - s/1001&&&&/&&&&9/g; - s/1010&&&&/&&&&A/g; - s/1011&&&&/&&&&B/g; - s/1100&&&&/&&&&C/g; - s/1101&&&&/&&&&D/g; - s/1110&&&&/&&&&E/g; - s/1111&&&&/&&&&F/g; - } - s/000&&&&/&&&&0/g; - s/001&&&&/&&&&1/g; - s/010&&&&/&&&&2/g; - s/011&&&&/&&&&3/g; - s/100&&&&/&&&&4/g; - s/101&&&&/&&&&5/g; - s/110&&&&/&&&&6/g; - s/111&&&&/&&&&7/g; - s/00&&&&/&&&&0/g; - s/01&&&&/&&&&1/g; - s/10&&&&/&&&&2/g; - s/11&&&&/&&&&3/g; - s/0&&&&/&&&&0/g; - s/1&&&&/&&&&1/g; - s/conv&&&&/0x/g; - } - - if ( /commandline/) - { - if( /-bigend/) - { - $bigend=1; - } - } - - if ( /\bDCDU\b/ ) - { - my $cmd=$_; - my $value; - my $prefix; - my $w1; - my $w2; - my $w3; - my $w4; - - s/\s+DCDU\b/@ $&/; - - $cmd =~ /\bDCDU\b\s+0x(\d+)/; - $value = $1; - $value =~ /(\w\w)(\w\w)(\w\w)(\w\w)/; - $w1 = $1; - $w2 = $2; - $w3 = $3; - $w4 = $4; - - if( $bigend ne "") - { - # big endian - $prefix = "\t.byte\t0x".$w1.";". - "\t.byte\t0x".$w2.";". - "\t.byte\t0x".$w3.";". - "\t.byte\t0x".$w4."; "; - } - else - { - # little endian - $prefix = "\t.byte\t0x".$w4.";". - "\t.byte\t0x".$w3.";". - "\t.byte\t0x".$w2.";". - "\t.byte\t0x".$w1."; "; - } - $_=$prefix.$_; - } - - if ( /\badrl\b/i ) - { - s/\badrl\s+(\w+)\s*,\s*(\w+)/ldr $1,=$2/i; - $addPadding = 1; - } - s/\bEND\b/@ END/; -} continue { - printf ("%s", $_) if $printit; - if ($addPadding != 0) - { - printf (" mov r0,r0\n"); - $addPadding = 0; - } -} -#If we had a code section, mark that this object doesn't need an executable -# stack. -if ($nxstack && !$apple) { - printf (" .section\t.note.GNU-stack,\"\",\%\%progbits\n"); -} diff --git a/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm-gnu.S b/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm-gnu.S deleted file mode 100644 index 10668e54..00000000 --- a/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm-gnu.S +++ /dev/null @@ -1,555 +0,0 @@ - .syntax unified -@ Copyright (c) 2007-2008 CSIRO -@ Copyright (c) 2007-2009 Xiph.Org Foundation -@ Copyright (c) 2013 Parrot -@ Written by Aurélien Zanelli -@ -@ Redistribution and use in source and binary forms, with or without -@ modification, are permitted provided that the following conditions -@ are met: -@ -@ - Redistributions of source code must retain the above copyright -@ notice, this list of conditions and the following disclaimer. -@ -@ - Redistributions in binary form must reproduce the above copyright -@ notice, this list of conditions and the following disclaimer in the -@ documentation and/or other materials provided with the distribution. -@ -@ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -@ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -@ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -@ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER -@ OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, -@ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, -@ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR -@ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF -@ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING -@ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - - .text; .p2align 2; .arch armv7-a - .fpu neon - .object_arch armv4t - - .include "celt/arm/armopts-gnu.S" - - .if OPUS_ARM_MAY_HAVE_EDSP - .global celt_pitch_xcorr_edsp - .endif - - .if OPUS_ARM_MAY_HAVE_NEON - .global celt_pitch_xcorr_neon - .endif - - .if OPUS_ARM_MAY_HAVE_NEON - -@ Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 - .type xcorr_kernel_neon, %function; xcorr_kernel_neon: @ PROC -xcorr_kernel_neon_start: - @ input: - @ r3 = int len - @ r4 = opus_val16 *x - @ r5 = opus_val16 *y - @ q0 = opus_val32 sum[4] - @ output: - @ q0 = opus_val32 sum[4] - @ preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 - @ internal usage: - @ r12 = int j - @ d3 = y_3|y_2|y_1|y_0 - @ q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 - @ q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 - @ q8 = scratch - @ - @ Load y[0...3] - @ This requires len>0 to always be valid (which we assert in the C code). - VLD1.16 {d5}, [r5]! - SUBS r12, r3, #8 - BLE xcorr_kernel_neon_process4 -@ Process 8 samples at a time. -@ This loop loads one y value more than we actually need. Therefore we have to -@ stop as soon as there are 8 or fewer samples left (instead of 7), to avoid -@ reading past the end of the array. -xcorr_kernel_neon_process8: - @ This loop has 19 total instructions (10 cycles to issue, minimum), with - @ - 2 cycles of ARM insrtuctions, - @ - 10 cycles of load/store/byte permute instructions, and - @ - 9 cycles of data processing instructions. - @ On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the - @ latter two categories, meaning the whole loop should run in 10 cycles per - @ iteration, barring cache misses. - @ - @ Load x[0...7] - VLD1.16 {d6, d7}, [r4]! - @ Unlike VMOV, VAND is a data processsing instruction (and doesn't get - @ assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. - VAND d3, d5, d5 - SUBS r12, r12, #8 - @ Load y[4...11] - VLD1.16 {d4, d5}, [r5]! - VMLAL.S16 q0, d3, d6[0] - VEXT.16 d16, d3, d4, #1 - VMLAL.S16 q0, d4, d7[0] - VEXT.16 d17, d4, d5, #1 - VMLAL.S16 q0, d16, d6[1] - VEXT.16 d16, d3, d4, #2 - VMLAL.S16 q0, d17, d7[1] - VEXT.16 d17, d4, d5, #2 - VMLAL.S16 q0, d16, d6[2] - VEXT.16 d16, d3, d4, #3 - VMLAL.S16 q0, d17, d7[2] - VEXT.16 d17, d4, d5, #3 - VMLAL.S16 q0, d16, d6[3] - VMLAL.S16 q0, d17, d7[3] - BGT xcorr_kernel_neon_process8 -@ Process 4 samples here if we have > 4 left (still reading one extra y value). -xcorr_kernel_neon_process4: - ADDS r12, r12, #4 - BLE xcorr_kernel_neon_process2 - @ Load x[0...3] - VLD1.16 d6, [r4]! - @ Use VAND since it's a data processing instruction again. - VAND d4, d5, d5 - SUB r12, r12, #4 - @ Load y[4...7] - VLD1.16 d5, [r5]! - VMLAL.S16 q0, d4, d6[0] - VEXT.16 d16, d4, d5, #1 - VMLAL.S16 q0, d16, d6[1] - VEXT.16 d16, d4, d5, #2 - VMLAL.S16 q0, d16, d6[2] - VEXT.16 d16, d4, d5, #3 - VMLAL.S16 q0, d16, d6[3] -@ Process 2 samples here if we have > 2 left (still reading one extra y value). -xcorr_kernel_neon_process2: - ADDS r12, r12, #2 - BLE xcorr_kernel_neon_process1 - @ Load x[0...1] - VLD2.16 {d6[],d7[]}, [r4]! - @ Use VAND since it's a data processing instruction again. - VAND d4, d5, d5 - SUB r12, r12, #2 - @ Load y[4...5] - VLD1.32 {d5[]}, [r5]! - VMLAL.S16 q0, d4, d6 - VEXT.16 d16, d4, d5, #1 - @ Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI - @ instead of VEXT, since it's a data-processing instruction. - VSRI.64 d5, d4, #32 - VMLAL.S16 q0, d16, d7 -@ Process 1 sample using the extra y value we loaded above. -xcorr_kernel_neon_process1: - @ Load next *x - VLD1.16 {d6[]}, [r4]! - ADDS r12, r12, #1 - @ y[0...3] are left in d5 from prior iteration(s) (if any) - VMLAL.S16 q0, d5, d6 - MOVLE pc, lr -@ Now process 1 last sample, not reading ahead. - @ Load last *y - VLD1.16 {d4[]}, [r5]! - VSRI.64 d4, d5, #16 - @ Load last *x - VLD1.16 {d6[]}, [r4]! - VMLAL.S16 q0, d4, d6 - MOV pc, lr - .size xcorr_kernel_neon, .-xcorr_kernel_neon @ ENDP - -@ opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, -@ opus_val32 *xcorr, int len, int max_pitch, int arch) - .type celt_pitch_xcorr_neon, %function; celt_pitch_xcorr_neon: @ PROC - @ input: - @ r0 = opus_val16 *_x - @ r1 = opus_val16 *_y - @ r2 = opus_val32 *xcorr - @ r3 = int len - @ output: - @ r0 = int maxcorr - @ internal usage: - @ r4 = opus_val16 *x (for xcorr_kernel_neon()) - @ r5 = opus_val16 *y (for xcorr_kernel_neon()) - @ r6 = int max_pitch - @ r12 = int j - @ q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) - @ ignored: - @ int arch - STMFD sp!, {r4-r6, lr} - LDR r6, [sp, #16] - VMOV.S32 q15, #1 - @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done - SUBS r6, r6, #4 - BLT celt_pitch_xcorr_neon_process4_done -celt_pitch_xcorr_neon_process4: - @ xcorr_kernel_neon parameters: - @ r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} - MOV r4, r0 - MOV r5, r1 - VEOR q0, q0, q0 - @ xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. - @ So we don't save/restore any other registers. - BL xcorr_kernel_neon_start - SUBS r6, r6, #4 - VST1.32 {q0}, [r2]! - @ _y += 4 - ADD r1, r1, #8 - VMAX.S32 q15, q15, q0 - @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done - BGE celt_pitch_xcorr_neon_process4 -@ We have less than 4 sums left to compute. -celt_pitch_xcorr_neon_process4_done: - ADDS r6, r6, #4 - @ Reduce maxcorr to a single value - VMAX.S32 d30, d30, d31 - VPMAX.S32 d30, d30, d30 - @ if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done - BLE celt_pitch_xcorr_neon_done -@ Now compute each remaining sum one at a time. -celt_pitch_xcorr_neon_process_remaining: - MOV r4, r0 - MOV r5, r1 - VMOV.I32 q0, #0 - SUBS r12, r3, #8 - BLT celt_pitch_xcorr_neon_process_remaining4 -@ Sum terms 8 at a time. -celt_pitch_xcorr_neon_process_remaining_loop8: - @ Load x[0...7] - VLD1.16 {q1}, [r4]! - @ Load y[0...7] - VLD1.16 {q2}, [r5]! - SUBS r12, r12, #8 - VMLAL.S16 q0, d4, d2 - VMLAL.S16 q0, d5, d3 - BGE celt_pitch_xcorr_neon_process_remaining_loop8 -@ Sum terms 4 at a time. -celt_pitch_xcorr_neon_process_remaining4: - ADDS r12, r12, #4 - BLT celt_pitch_xcorr_neon_process_remaining4_done - @ Load x[0...3] - VLD1.16 {d2}, [r4]! - @ Load y[0...3] - VLD1.16 {d3}, [r5]! - SUB r12, r12, #4 - VMLAL.S16 q0, d3, d2 -celt_pitch_xcorr_neon_process_remaining4_done: - @ Reduce the sum to a single value. - VADD.S32 d0, d0, d1 - VPADDL.S32 d0, d0 - ADDS r12, r12, #4 - BLE celt_pitch_xcorr_neon_process_remaining_loop_done -@ Sum terms 1 at a time. -celt_pitch_xcorr_neon_process_remaining_loop1: - VLD1.16 {d2[]}, [r4]! - VLD1.16 {d3[]}, [r5]! - SUBS r12, r12, #1 - VMLAL.S16 q0, d2, d3 - BGT celt_pitch_xcorr_neon_process_remaining_loop1 -celt_pitch_xcorr_neon_process_remaining_loop_done: - VST1.32 {d0[0]}, [r2]! - VMAX.S32 d30, d30, d0 - SUBS r6, r6, #1 - @ _y++ - ADD r1, r1, #2 - @ if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining - BGT celt_pitch_xcorr_neon_process_remaining -celt_pitch_xcorr_neon_done: - VMOV.32 r0, d30[0] - LDMFD sp!, {r4-r6, pc} - .size celt_pitch_xcorr_neon, .-celt_pitch_xcorr_neon @ ENDP - - .endif - - .if OPUS_ARM_MAY_HAVE_EDSP - -@ This will get used on ARMv7 devices without NEON, so it has been optimized -@ to take advantage of dual-issuing where possible. - .type xcorr_kernel_edsp, %function; xcorr_kernel_edsp: @ PROC -xcorr_kernel_edsp_start: - @ input: - @ r3 = int len - @ r4 = opus_val16 *_x (must be 32-bit aligned) - @ r5 = opus_val16 *_y (must be 32-bit aligned) - @ r6...r9 = opus_val32 sum[4] - @ output: - @ r6...r9 = opus_val32 sum[4] - @ preserved: r0-r5 - @ internal usage - @ r2 = int j - @ r12,r14 = opus_val16 x[4] - @ r10,r11 = opus_val16 y[4] - STMFD sp!, {r2,r4,r5,lr} - LDR r10, [r5], #4 @ Load y[0...1] - SUBS r2, r3, #4 @ j = len-4 - LDR r11, [r5], #4 @ Load y[2...3] - BLE xcorr_kernel_edsp_process4_done - LDR r12, [r4], #4 @ Load x[0...1] - @ Stall -xcorr_kernel_edsp_process4: - @ The multiplies must issue from pipeline 0, and can't dual-issue with each - @ other. Every other instruction here dual-issues with a multiply, and is - @ thus "free". There should be no stalls in the body of the loop. - SMLABB r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x_0,y_0) - LDR r14, [r4], #4 @ Load x[2...3] - SMLABT r7, r12, r10, r7 @ sum[1] = MAC16_16(sum[1],x_0,y_1) - SUBS r2, r2, #4 @ j-=4 - SMLABB r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x_0,y_2) - SMLABT r9, r12, r11, r9 @ sum[3] = MAC16_16(sum[3],x_0,y_3) - SMLATT r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x_1,y_1) - LDR r10, [r5], #4 @ Load y[4...5] - SMLATB r7, r12, r11, r7 @ sum[1] = MAC16_16(sum[1],x_1,y_2) - SMLATT r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x_1,y_3) - SMLATB r9, r12, r10, r9 @ sum[3] = MAC16_16(sum[3],x_1,y_4) - LDRGT r12, [r4], #4 @ Load x[0...1] - SMLABB r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],x_2,y_2) - SMLABT r7, r14, r11, r7 @ sum[1] = MAC16_16(sum[1],x_2,y_3) - SMLABB r8, r14, r10, r8 @ sum[2] = MAC16_16(sum[2],x_2,y_4) - SMLABT r9, r14, r10, r9 @ sum[3] = MAC16_16(sum[3],x_2,y_5) - SMLATT r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],x_3,y_3) - LDR r11, [r5], #4 @ Load y[6...7] - SMLATB r7, r14, r10, r7 @ sum[1] = MAC16_16(sum[1],x_3,y_4) - SMLATT r8, r14, r10, r8 @ sum[2] = MAC16_16(sum[2],x_3,y_5) - SMLATB r9, r14, r11, r9 @ sum[3] = MAC16_16(sum[3],x_3,y_6) - BGT xcorr_kernel_edsp_process4 -xcorr_kernel_edsp_process4_done: - ADDS r2, r2, #4 - BLE xcorr_kernel_edsp_done - LDRH r12, [r4], #2 @ r12 = *x++ - SUBS r2, r2, #1 @ j-- - @ Stall - SMLABB r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x,y_0) - LDRHGT r14, [r4], #2 @ r14 = *x++ - SMLABT r7, r12, r10, r7 @ sum[1] = MAC16_16(sum[1],x,y_1) - SMLABB r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x,y_2) - SMLABT r9, r12, r11, r9 @ sum[3] = MAC16_16(sum[3],x,y_3) - BLE xcorr_kernel_edsp_done - SMLABT r6, r14, r10, r6 @ sum[0] = MAC16_16(sum[0],x,y_1) - SUBS r2, r2, #1 @ j-- - SMLABB r7, r14, r11, r7 @ sum[1] = MAC16_16(sum[1],x,y_2) - LDRH r10, [r5], #2 @ r10 = y_4 = *y++ - SMLABT r8, r14, r11, r8 @ sum[2] = MAC16_16(sum[2],x,y_3) - LDRHGT r12, [r4], #2 @ r12 = *x++ - SMLABB r9, r14, r10, r9 @ sum[3] = MAC16_16(sum[3],x,y_4) - BLE xcorr_kernel_edsp_done - SMLABB r6, r12, r11, r6 @ sum[0] = MAC16_16(sum[0],tmp,y_2) - CMP r2, #1 @ j-- - SMLABT r7, r12, r11, r7 @ sum[1] = MAC16_16(sum[1],tmp,y_3) - LDRH r2, [r5], #2 @ r2 = y_5 = *y++ - SMLABB r8, r12, r10, r8 @ sum[2] = MAC16_16(sum[2],tmp,y_4) - LDRHGT r14, [r4] @ r14 = *x - SMLABB r9, r12, r2, r9 @ sum[3] = MAC16_16(sum[3],tmp,y_5) - BLE xcorr_kernel_edsp_done - SMLABT r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],tmp,y_3) - LDRH r11, [r5] @ r11 = y_6 = *y - SMLABB r7, r14, r10, r7 @ sum[1] = MAC16_16(sum[1],tmp,y_4) - SMLABB r8, r14, r2, r8 @ sum[2] = MAC16_16(sum[2],tmp,y_5) - SMLABB r9, r14, r11, r9 @ sum[3] = MAC16_16(sum[3],tmp,y_6) -xcorr_kernel_edsp_done: - LDMFD sp!, {r2,r4,r5,pc} - .size xcorr_kernel_edsp, .-xcorr_kernel_edsp @ ENDP - - .type celt_pitch_xcorr_edsp, %function; celt_pitch_xcorr_edsp: @ PROC - @ input: - @ r0 = opus_val16 *_x (must be 32-bit aligned) - @ r1 = opus_val16 *_y (only needs to be 16-bit aligned) - @ r2 = opus_val32 *xcorr - @ r3 = int len - @ output: - @ r0 = maxcorr - @ internal usage - @ r4 = opus_val16 *x - @ r5 = opus_val16 *y - @ r6 = opus_val32 sum0 - @ r7 = opus_val32 sum1 - @ r8 = opus_val32 sum2 - @ r9 = opus_val32 sum3 - @ r1 = int max_pitch - @ r12 = int j - @ ignored: - @ int arch - STMFD sp!, {r4-r11, lr} - MOV r5, r1 - LDR r1, [sp, #36] - MOV r4, r0 - TST r5, #3 - @ maxcorr = 1 - MOV r0, #1 - BEQ celt_pitch_xcorr_edsp_process1u_done -@ Compute one sum at the start to make y 32-bit aligned. - SUBS r12, r3, #4 - @ r14 = sum = 0 - MOV r14, #0 - LDRH r8, [r5], #2 - BLE celt_pitch_xcorr_edsp_process1u_loop4_done - LDR r6, [r4], #4 - MOV r8, r8, LSL #16 -celt_pitch_xcorr_edsp_process1u_loop4: - LDR r9, [r5], #4 - SMLABT r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) - LDR r7, [r4], #4 - SMLATB r14, r6, r9, r14 @ sum = MAC16_16(sum, x_1, y_1) - LDR r8, [r5], #4 - SMLABT r14, r7, r9, r14 @ sum = MAC16_16(sum, x_2, y_2) - SUBS r12, r12, #4 @ j-=4 - SMLATB r14, r7, r8, r14 @ sum = MAC16_16(sum, x_3, y_3) - LDRGT r6, [r4], #4 - BGT celt_pitch_xcorr_edsp_process1u_loop4 - MOV r8, r8, LSR #16 -celt_pitch_xcorr_edsp_process1u_loop4_done: - ADDS r12, r12, #4 -celt_pitch_xcorr_edsp_process1u_loop1: - LDRHGE r6, [r4], #2 - @ Stall - SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, *x, *y) - SUBSGE r12, r12, #1 - LDRHGT r8, [r5], #2 - BGT celt_pitch_xcorr_edsp_process1u_loop1 - @ Restore _x - SUB r4, r4, r3, LSL #1 - @ Restore and advance _y - SUB r5, r5, r3, LSL #1 - @ maxcorr = max(maxcorr, sum) - CMP r0, r14 - ADD r5, r5, #2 - MOVLT r0, r14 - SUBS r1, r1, #1 - @ xcorr[i] = sum - STR r14, [r2], #4 - BLE celt_pitch_xcorr_edsp_done -celt_pitch_xcorr_edsp_process1u_done: - @ if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 - SUBS r1, r1, #4 - BLT celt_pitch_xcorr_edsp_process2 -celt_pitch_xcorr_edsp_process4: - @ xcorr_kernel_edsp parameters: - @ r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} - MOV r6, #0 - MOV r7, #0 - MOV r8, #0 - MOV r9, #0 - BL xcorr_kernel_edsp_start @ xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) - @ maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) - CMP r0, r6 - @ _y+=4 - ADD r5, r5, #8 - MOVLT r0, r6 - CMP r0, r7 - MOVLT r0, r7 - CMP r0, r8 - MOVLT r0, r8 - CMP r0, r9 - MOVLT r0, r9 - STMIA r2!, {r6-r9} - SUBS r1, r1, #4 - BGE celt_pitch_xcorr_edsp_process4 -celt_pitch_xcorr_edsp_process2: - ADDS r1, r1, #2 - BLT celt_pitch_xcorr_edsp_process1a - SUBS r12, r3, #4 - @ {r10, r11} = {sum0, sum1} = {0, 0} - MOV r10, #0 - MOV r11, #0 - LDR r8, [r5], #4 - BLE celt_pitch_xcorr_edsp_process2_loop_done - LDR r6, [r4], #4 - LDR r9, [r5], #4 -celt_pitch_xcorr_edsp_process2_loop4: - SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) - LDR r7, [r4], #4 - SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) - SUBS r12, r12, #4 @ j-=4 - SMLATT r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_1, y_1) - LDR r8, [r5], #4 - SMLATB r11, r6, r9, r11 @ sum1 = MAC16_16(sum1, x_1, y_2) - LDRGT r6, [r4], #4 - SMLABB r10, r7, r9, r10 @ sum0 = MAC16_16(sum0, x_2, y_2) - SMLABT r11, r7, r9, r11 @ sum1 = MAC16_16(sum1, x_2, y_3) - SMLATT r10, r7, r9, r10 @ sum0 = MAC16_16(sum0, x_3, y_3) - LDRGT r9, [r5], #4 - SMLATB r11, r7, r8, r11 @ sum1 = MAC16_16(sum1, x_3, y_4) - BGT celt_pitch_xcorr_edsp_process2_loop4 -celt_pitch_xcorr_edsp_process2_loop_done: - ADDS r12, r12, #2 - BLE celt_pitch_xcorr_edsp_process2_1 - LDR r6, [r4], #4 - @ Stall - SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) - LDR r9, [r5], #4 - SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) - SUB r12, r12, #2 - SMLATT r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_1, y_1) - MOV r8, r9 - SMLATB r11, r6, r9, r11 @ sum1 = MAC16_16(sum1, x_1, y_2) -celt_pitch_xcorr_edsp_process2_1: - LDRH r6, [r4], #2 - ADDS r12, r12, #1 - @ Stall - SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) - LDRHGT r7, [r4], #2 - SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) - BLE celt_pitch_xcorr_edsp_process2_done - LDRH r9, [r5], #2 - SMLABT r10, r7, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_1) - SMLABB r11, r7, r9, r11 @ sum1 = MAC16_16(sum1, x_0, y_2) -celt_pitch_xcorr_edsp_process2_done: - @ Restore _x - SUB r4, r4, r3, LSL #1 - @ Restore and advance _y - SUB r5, r5, r3, LSL #1 - @ maxcorr = max(maxcorr, sum0) - CMP r0, r10 - ADD r5, r5, #2 - MOVLT r0, r10 - SUB r1, r1, #2 - @ maxcorr = max(maxcorr, sum1) - CMP r0, r11 - @ xcorr[i] = sum - STR r10, [r2], #4 - MOVLT r0, r11 - STR r11, [r2], #4 -celt_pitch_xcorr_edsp_process1a: - ADDS r1, r1, #1 - BLT celt_pitch_xcorr_edsp_done - SUBS r12, r3, #4 - @ r14 = sum = 0 - MOV r14, #0 - BLT celt_pitch_xcorr_edsp_process1a_loop_done - LDR r6, [r4], #4 - LDR r8, [r5], #4 - LDR r7, [r4], #4 - LDR r9, [r5], #4 -celt_pitch_xcorr_edsp_process1a_loop4: - SMLABB r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) - SUBS r12, r12, #4 @ j-=4 - SMLATT r14, r6, r8, r14 @ sum = MAC16_16(sum, x_1, y_1) - LDRGE r6, [r4], #4 - SMLABB r14, r7, r9, r14 @ sum = MAC16_16(sum, x_2, y_2) - LDRGE r8, [r5], #4 - SMLATT r14, r7, r9, r14 @ sum = MAC16_16(sum, x_3, y_3) - LDRGE r7, [r4], #4 - LDRGE r9, [r5], #4 - BGE celt_pitch_xcorr_edsp_process1a_loop4 -celt_pitch_xcorr_edsp_process1a_loop_done: - ADDS r12, r12, #2 - LDRGE r6, [r4], #4 - LDRGE r8, [r5], #4 - @ Stall - SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) - SUBGE r12, r12, #2 - SMLATTGE r14, r6, r8, r14 @ sum = MAC16_16(sum, x_1, y_1) - ADDS r12, r12, #1 - LDRHGE r6, [r4], #2 - LDRHGE r8, [r5], #2 - @ Stall - SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, *x, *y) - @ maxcorr = max(maxcorr, sum) - CMP r0, r14 - @ xcorr[i] = sum - STR r14, [r2], #4 - MOVLT r0, r14 -celt_pitch_xcorr_edsp_done: - LDMFD sp!, {r4-r11, pc} - .size celt_pitch_xcorr_edsp, .-celt_pitch_xcorr_edsp @ ENDP - - .endif - -@ END: - .section .note.GNU-stack,"",%progbits diff --git a/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm.s b/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm.s deleted file mode 100644 index 6e873afc..00000000 --- a/code/opus-1.2.1/celt/arm/celt_pitch_xcorr_arm.s +++ /dev/null @@ -1,551 +0,0 @@ -; Copyright (c) 2007-2008 CSIRO -; Copyright (c) 2007-2009 Xiph.Org Foundation -; Copyright (c) 2013 Parrot -; Written by Aurélien Zanelli -; -; Redistribution and use in source and binary forms, with or without -; modification, are permitted provided that the following conditions -; are met: -; -; - Redistributions of source code must retain the above copyright -; notice, this list of conditions and the following disclaimer. -; -; - Redistributions in binary form must reproduce the above copyright -; notice, this list of conditions and the following disclaimer in the -; documentation and/or other materials provided with the distribution. -; -; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -; ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER -; OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, -; EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, -; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR -; PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF -; LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING -; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS -; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - - AREA |.text|, CODE, READONLY - - GET celt/arm/armopts.s - -IF OPUS_ARM_MAY_HAVE_EDSP - EXPORT celt_pitch_xcorr_edsp -ENDIF - -IF OPUS_ARM_MAY_HAVE_NEON - EXPORT celt_pitch_xcorr_neon -ENDIF - -IF OPUS_ARM_MAY_HAVE_NEON - -; Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 -xcorr_kernel_neon PROC -xcorr_kernel_neon_start - ; input: - ; r3 = int len - ; r4 = opus_val16 *x - ; r5 = opus_val16 *y - ; q0 = opus_val32 sum[4] - ; output: - ; q0 = opus_val32 sum[4] - ; preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 - ; internal usage: - ; r12 = int j - ; d3 = y_3|y_2|y_1|y_0 - ; q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 - ; q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 - ; q8 = scratch - ; - ; Load y[0...3] - ; This requires len>0 to always be valid (which we assert in the C code). - VLD1.16 {d5}, [r5]! - SUBS r12, r3, #8 - BLE xcorr_kernel_neon_process4 -; Process 8 samples at a time. -; This loop loads one y value more than we actually need. Therefore we have to -; stop as soon as there are 8 or fewer samples left (instead of 7), to avoid -; reading past the end of the array. -xcorr_kernel_neon_process8 - ; This loop has 19 total instructions (10 cycles to issue, minimum), with - ; - 2 cycles of ARM insrtuctions, - ; - 10 cycles of load/store/byte permute instructions, and - ; - 9 cycles of data processing instructions. - ; On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the - ; latter two categories, meaning the whole loop should run in 10 cycles per - ; iteration, barring cache misses. - ; - ; Load x[0...7] - VLD1.16 {d6, d7}, [r4]! - ; Unlike VMOV, VAND is a data processsing instruction (and doesn't get - ; assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. - VAND d3, d5, d5 - SUBS r12, r12, #8 - ; Load y[4...11] - VLD1.16 {d4, d5}, [r5]! - VMLAL.S16 q0, d3, d6[0] - VEXT.16 d16, d3, d4, #1 - VMLAL.S16 q0, d4, d7[0] - VEXT.16 d17, d4, d5, #1 - VMLAL.S16 q0, d16, d6[1] - VEXT.16 d16, d3, d4, #2 - VMLAL.S16 q0, d17, d7[1] - VEXT.16 d17, d4, d5, #2 - VMLAL.S16 q0, d16, d6[2] - VEXT.16 d16, d3, d4, #3 - VMLAL.S16 q0, d17, d7[2] - VEXT.16 d17, d4, d5, #3 - VMLAL.S16 q0, d16, d6[3] - VMLAL.S16 q0, d17, d7[3] - BGT xcorr_kernel_neon_process8 -; Process 4 samples here if we have > 4 left (still reading one extra y value). -xcorr_kernel_neon_process4 - ADDS r12, r12, #4 - BLE xcorr_kernel_neon_process2 - ; Load x[0...3] - VLD1.16 d6, [r4]! - ; Use VAND since it's a data processing instruction again. - VAND d4, d5, d5 - SUB r12, r12, #4 - ; Load y[4...7] - VLD1.16 d5, [r5]! - VMLAL.S16 q0, d4, d6[0] - VEXT.16 d16, d4, d5, #1 - VMLAL.S16 q0, d16, d6[1] - VEXT.16 d16, d4, d5, #2 - VMLAL.S16 q0, d16, d6[2] - VEXT.16 d16, d4, d5, #3 - VMLAL.S16 q0, d16, d6[3] -; Process 2 samples here if we have > 2 left (still reading one extra y value). -xcorr_kernel_neon_process2 - ADDS r12, r12, #2 - BLE xcorr_kernel_neon_process1 - ; Load x[0...1] - VLD2.16 {d6[],d7[]}, [r4]! - ; Use VAND since it's a data processing instruction again. - VAND d4, d5, d5 - SUB r12, r12, #2 - ; Load y[4...5] - VLD1.32 {d5[]}, [r5]! - VMLAL.S16 q0, d4, d6 - VEXT.16 d16, d4, d5, #1 - ; Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI - ; instead of VEXT, since it's a data-processing instruction. - VSRI.64 d5, d4, #32 - VMLAL.S16 q0, d16, d7 -; Process 1 sample using the extra y value we loaded above. -xcorr_kernel_neon_process1 - ; Load next *x - VLD1.16 {d6[]}, [r4]! - ADDS r12, r12, #1 - ; y[0...3] are left in d5 from prior iteration(s) (if any) - VMLAL.S16 q0, d5, d6 - MOVLE pc, lr -; Now process 1 last sample, not reading ahead. - ; Load last *y - VLD1.16 {d4[]}, [r5]! - VSRI.64 d4, d5, #16 - ; Load last *x - VLD1.16 {d6[]}, [r4]! - VMLAL.S16 q0, d4, d6 - MOV pc, lr - ENDP - -; opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, -; opus_val32 *xcorr, int len, int max_pitch, int arch) -celt_pitch_xcorr_neon PROC - ; input: - ; r0 = opus_val16 *_x - ; r1 = opus_val16 *_y - ; r2 = opus_val32 *xcorr - ; r3 = int len - ; output: - ; r0 = int maxcorr - ; internal usage: - ; r4 = opus_val16 *x (for xcorr_kernel_neon()) - ; r5 = opus_val16 *y (for xcorr_kernel_neon()) - ; r6 = int max_pitch - ; r12 = int j - ; q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) - ; ignored: - ; int arch - STMFD sp!, {r4-r6, lr} - LDR r6, [sp, #16] - VMOV.S32 q15, #1 - ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done - SUBS r6, r6, #4 - BLT celt_pitch_xcorr_neon_process4_done -celt_pitch_xcorr_neon_process4 - ; xcorr_kernel_neon parameters: - ; r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} - MOV r4, r0 - MOV r5, r1 - VEOR q0, q0, q0 - ; xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. - ; So we don't save/restore any other registers. - BL xcorr_kernel_neon_start - SUBS r6, r6, #4 - VST1.32 {q0}, [r2]! - ; _y += 4 - ADD r1, r1, #8 - VMAX.S32 q15, q15, q0 - ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done - BGE celt_pitch_xcorr_neon_process4 -; We have less than 4 sums left to compute. -celt_pitch_xcorr_neon_process4_done - ADDS r6, r6, #4 - ; Reduce maxcorr to a single value - VMAX.S32 d30, d30, d31 - VPMAX.S32 d30, d30, d30 - ; if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done - BLE celt_pitch_xcorr_neon_done -; Now compute each remaining sum one at a time. -celt_pitch_xcorr_neon_process_remaining - MOV r4, r0 - MOV r5, r1 - VMOV.I32 q0, #0 - SUBS r12, r3, #8 - BLT celt_pitch_xcorr_neon_process_remaining4 -; Sum terms 8 at a time. -celt_pitch_xcorr_neon_process_remaining_loop8 - ; Load x[0...7] - VLD1.16 {q1}, [r4]! - ; Load y[0...7] - VLD1.16 {q2}, [r5]! - SUBS r12, r12, #8 - VMLAL.S16 q0, d4, d2 - VMLAL.S16 q0, d5, d3 - BGE celt_pitch_xcorr_neon_process_remaining_loop8 -; Sum terms 4 at a time. -celt_pitch_xcorr_neon_process_remaining4 - ADDS r12, r12, #4 - BLT celt_pitch_xcorr_neon_process_remaining4_done - ; Load x[0...3] - VLD1.16 {d2}, [r4]! - ; Load y[0...3] - VLD1.16 {d3}, [r5]! - SUB r12, r12, #4 - VMLAL.S16 q0, d3, d2 -celt_pitch_xcorr_neon_process_remaining4_done - ; Reduce the sum to a single value. - VADD.S32 d0, d0, d1 - VPADDL.S32 d0, d0 - ADDS r12, r12, #4 - BLE celt_pitch_xcorr_neon_process_remaining_loop_done -; Sum terms 1 at a time. -celt_pitch_xcorr_neon_process_remaining_loop1 - VLD1.16 {d2[]}, [r4]! - VLD1.16 {d3[]}, [r5]! - SUBS r12, r12, #1 - VMLAL.S16 q0, d2, d3 - BGT celt_pitch_xcorr_neon_process_remaining_loop1 -celt_pitch_xcorr_neon_process_remaining_loop_done - VST1.32 {d0[0]}, [r2]! - VMAX.S32 d30, d30, d0 - SUBS r6, r6, #1 - ; _y++ - ADD r1, r1, #2 - ; if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining - BGT celt_pitch_xcorr_neon_process_remaining -celt_pitch_xcorr_neon_done - VMOV.32 r0, d30[0] - LDMFD sp!, {r4-r6, pc} - ENDP - -ENDIF - -IF OPUS_ARM_MAY_HAVE_EDSP - -; This will get used on ARMv7 devices without NEON, so it has been optimized -; to take advantage of dual-issuing where possible. -xcorr_kernel_edsp PROC -xcorr_kernel_edsp_start - ; input: - ; r3 = int len - ; r4 = opus_val16 *_x (must be 32-bit aligned) - ; r5 = opus_val16 *_y (must be 32-bit aligned) - ; r6...r9 = opus_val32 sum[4] - ; output: - ; r6...r9 = opus_val32 sum[4] - ; preserved: r0-r5 - ; internal usage - ; r2 = int j - ; r12,r14 = opus_val16 x[4] - ; r10,r11 = opus_val16 y[4] - STMFD sp!, {r2,r4,r5,lr} - LDR r10, [r5], #4 ; Load y[0...1] - SUBS r2, r3, #4 ; j = len-4 - LDR r11, [r5], #4 ; Load y[2...3] - BLE xcorr_kernel_edsp_process4_done - LDR r12, [r4], #4 ; Load x[0...1] - ; Stall -xcorr_kernel_edsp_process4 - ; The multiplies must issue from pipeline 0, and can't dual-issue with each - ; other. Every other instruction here dual-issues with a multiply, and is - ; thus "free". There should be no stalls in the body of the loop. - SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_0,y_0) - LDR r14, [r4], #4 ; Load x[2...3] - SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x_0,y_1) - SUBS r2, r2, #4 ; j-=4 - SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_0,y_2) - SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x_0,y_3) - SMLATT r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_1,y_1) - LDR r10, [r5], #4 ; Load y[4...5] - SMLATB r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],x_1,y_2) - SMLATT r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_1,y_3) - SMLATB r9, r12, r10, r9 ; sum[3] = MAC16_16(sum[3],x_1,y_4) - LDRGT r12, [r4], #4 ; Load x[0...1] - SMLABB r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_2,y_2) - SMLABT r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x_2,y_3) - SMLABB r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_2,y_4) - SMLABT r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x_2,y_5) - SMLATT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_3,y_3) - LDR r11, [r5], #4 ; Load y[6...7] - SMLATB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],x_3,y_4) - SMLATT r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_3,y_5) - SMLATB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],x_3,y_6) - BGT xcorr_kernel_edsp_process4 -xcorr_kernel_edsp_process4_done - ADDS r2, r2, #4 - BLE xcorr_kernel_edsp_done - LDRH r12, [r4], #2 ; r12 = *x++ - SUBS r2, r2, #1 ; j-- - ; Stall - SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_0) - LDRHGT r14, [r4], #2 ; r14 = *x++ - SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x,y_1) - SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_2) - SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x,y_3) - BLE xcorr_kernel_edsp_done - SMLABT r6, r14, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_1) - SUBS r2, r2, #1 ; j-- - SMLABB r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x,y_2) - LDRH r10, [r5], #2 ; r10 = y_4 = *y++ - SMLABT r8, r14, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_3) - LDRHGT r12, [r4], #2 ; r12 = *x++ - SMLABB r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x,y_4) - BLE xcorr_kernel_edsp_done - SMLABB r6, r12, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_2) - CMP r2, #1 ; j-- - SMLABT r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_3) - LDRH r2, [r5], #2 ; r2 = y_5 = *y++ - SMLABB r8, r12, r10, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_4) - LDRHGT r14, [r4] ; r14 = *x - SMLABB r9, r12, r2, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_5) - BLE xcorr_kernel_edsp_done - SMLABT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_3) - LDRH r11, [r5] ; r11 = y_6 = *y - SMLABB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_4) - SMLABB r8, r14, r2, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_5) - SMLABB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_6) -xcorr_kernel_edsp_done - LDMFD sp!, {r2,r4,r5,pc} - ENDP - -celt_pitch_xcorr_edsp PROC - ; input: - ; r0 = opus_val16 *_x (must be 32-bit aligned) - ; r1 = opus_val16 *_y (only needs to be 16-bit aligned) - ; r2 = opus_val32 *xcorr - ; r3 = int len - ; output: - ; r0 = maxcorr - ; internal usage - ; r4 = opus_val16 *x - ; r5 = opus_val16 *y - ; r6 = opus_val32 sum0 - ; r7 = opus_val32 sum1 - ; r8 = opus_val32 sum2 - ; r9 = opus_val32 sum3 - ; r1 = int max_pitch - ; r12 = int j - ; ignored: - ; int arch - STMFD sp!, {r4-r11, lr} - MOV r5, r1 - LDR r1, [sp, #36] - MOV r4, r0 - TST r5, #3 - ; maxcorr = 1 - MOV r0, #1 - BEQ celt_pitch_xcorr_edsp_process1u_done -; Compute one sum at the start to make y 32-bit aligned. - SUBS r12, r3, #4 - ; r14 = sum = 0 - MOV r14, #0 - LDRH r8, [r5], #2 - BLE celt_pitch_xcorr_edsp_process1u_loop4_done - LDR r6, [r4], #4 - MOV r8, r8, LSL #16 -celt_pitch_xcorr_edsp_process1u_loop4 - LDR r9, [r5], #4 - SMLABT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) - LDR r7, [r4], #4 - SMLATB r14, r6, r9, r14 ; sum = MAC16_16(sum, x_1, y_1) - LDR r8, [r5], #4 - SMLABT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) - SUBS r12, r12, #4 ; j-=4 - SMLATB r14, r7, r8, r14 ; sum = MAC16_16(sum, x_3, y_3) - LDRGT r6, [r4], #4 - BGT celt_pitch_xcorr_edsp_process1u_loop4 - MOV r8, r8, LSR #16 -celt_pitch_xcorr_edsp_process1u_loop4_done - ADDS r12, r12, #4 -celt_pitch_xcorr_edsp_process1u_loop1 - LDRHGE r6, [r4], #2 - ; Stall - SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) - SUBSGE r12, r12, #1 - LDRHGT r8, [r5], #2 - BGT celt_pitch_xcorr_edsp_process1u_loop1 - ; Restore _x - SUB r4, r4, r3, LSL #1 - ; Restore and advance _y - SUB r5, r5, r3, LSL #1 - ; maxcorr = max(maxcorr, sum) - CMP r0, r14 - ADD r5, r5, #2 - MOVLT r0, r14 - SUBS r1, r1, #1 - ; xcorr[i] = sum - STR r14, [r2], #4 - BLE celt_pitch_xcorr_edsp_done -celt_pitch_xcorr_edsp_process1u_done - ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 - SUBS r1, r1, #4 - BLT celt_pitch_xcorr_edsp_process2 -celt_pitch_xcorr_edsp_process4 - ; xcorr_kernel_edsp parameters: - ; r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} - MOV r6, #0 - MOV r7, #0 - MOV r8, #0 - MOV r9, #0 - BL xcorr_kernel_edsp_start ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) - ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) - CMP r0, r6 - ; _y+=4 - ADD r5, r5, #8 - MOVLT r0, r6 - CMP r0, r7 - MOVLT r0, r7 - CMP r0, r8 - MOVLT r0, r8 - CMP r0, r9 - MOVLT r0, r9 - STMIA r2!, {r6-r9} - SUBS r1, r1, #4 - BGE celt_pitch_xcorr_edsp_process4 -celt_pitch_xcorr_edsp_process2 - ADDS r1, r1, #2 - BLT celt_pitch_xcorr_edsp_process1a - SUBS r12, r3, #4 - ; {r10, r11} = {sum0, sum1} = {0, 0} - MOV r10, #0 - MOV r11, #0 - LDR r8, [r5], #4 - BLE celt_pitch_xcorr_edsp_process2_loop_done - LDR r6, [r4], #4 - LDR r9, [r5], #4 -celt_pitch_xcorr_edsp_process2_loop4 - SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) - LDR r7, [r4], #4 - SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) - SUBS r12, r12, #4 ; j-=4 - SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) - LDR r8, [r5], #4 - SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) - LDRGT r6, [r4], #4 - SMLABB r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_2, y_2) - SMLABT r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_2, y_3) - SMLATT r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_3, y_3) - LDRGT r9, [r5], #4 - SMLATB r11, r7, r8, r11 ; sum1 = MAC16_16(sum1, x_3, y_4) - BGT celt_pitch_xcorr_edsp_process2_loop4 -celt_pitch_xcorr_edsp_process2_loop_done - ADDS r12, r12, #2 - BLE celt_pitch_xcorr_edsp_process2_1 - LDR r6, [r4], #4 - ; Stall - SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) - LDR r9, [r5], #4 - SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) - SUB r12, r12, #2 - SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) - MOV r8, r9 - SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) -celt_pitch_xcorr_edsp_process2_1 - LDRH r6, [r4], #2 - ADDS r12, r12, #1 - ; Stall - SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) - LDRHGT r7, [r4], #2 - SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) - BLE celt_pitch_xcorr_edsp_process2_done - LDRH r9, [r5], #2 - SMLABT r10, r7, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_1) - SMLABB r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_0, y_2) -celt_pitch_xcorr_edsp_process2_done - ; Restore _x - SUB r4, r4, r3, LSL #1 - ; Restore and advance _y - SUB r5, r5, r3, LSL #1 - ; maxcorr = max(maxcorr, sum0) - CMP r0, r10 - ADD r5, r5, #2 - MOVLT r0, r10 - SUB r1, r1, #2 - ; maxcorr = max(maxcorr, sum1) - CMP r0, r11 - ; xcorr[i] = sum - STR r10, [r2], #4 - MOVLT r0, r11 - STR r11, [r2], #4 -celt_pitch_xcorr_edsp_process1a - ADDS r1, r1, #1 - BLT celt_pitch_xcorr_edsp_done - SUBS r12, r3, #4 - ; r14 = sum = 0 - MOV r14, #0 - BLT celt_pitch_xcorr_edsp_process1a_loop_done - LDR r6, [r4], #4 - LDR r8, [r5], #4 - LDR r7, [r4], #4 - LDR r9, [r5], #4 -celt_pitch_xcorr_edsp_process1a_loop4 - SMLABB r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) - SUBS r12, r12, #4 ; j-=4 - SMLATT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) - LDRGE r6, [r4], #4 - SMLABB r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) - LDRGE r8, [r5], #4 - SMLATT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_3, y_3) - LDRGE r7, [r4], #4 - LDRGE r9, [r5], #4 - BGE celt_pitch_xcorr_edsp_process1a_loop4 -celt_pitch_xcorr_edsp_process1a_loop_done - ADDS r12, r12, #2 - LDRGE r6, [r4], #4 - LDRGE r8, [r5], #4 - ; Stall - SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) - SUBGE r12, r12, #2 - SMLATTGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) - ADDS r12, r12, #1 - LDRHGE r6, [r4], #2 - LDRHGE r8, [r5], #2 - ; Stall - SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) - ; maxcorr = max(maxcorr, sum) - CMP r0, r14 - ; xcorr[i] = sum - STR r14, [r2], #4 - MOVLT r0, r14 -celt_pitch_xcorr_edsp_done - LDMFD sp!, {r4-r11, pc} - ENDP - -ENDIF - -END diff --git a/code/opus-1.2.1/celt/x86/x86cpu.h b/code/opus-1.2.1/celt/x86/x86cpu.h deleted file mode 100644 index 04fd48aa..00000000 --- a/code/opus-1.2.1/celt/x86/x86cpu.h +++ /dev/null @@ -1,93 +0,0 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang - - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#if !defined(X86CPU_H) -# define X86CPU_H - -# if defined(OPUS_X86_MAY_HAVE_SSE) -# define MAY_HAVE_SSE(name) name ## _sse -# else -# define MAY_HAVE_SSE(name) name ## _c -# endif - -# if defined(OPUS_X86_MAY_HAVE_SSE2) -# define MAY_HAVE_SSE2(name) name ## _sse2 -# else -# define MAY_HAVE_SSE2(name) name ## _c -# endif - -# if defined(OPUS_X86_MAY_HAVE_SSE4_1) -# define MAY_HAVE_SSE4_1(name) name ## _sse4_1 -# else -# define MAY_HAVE_SSE4_1(name) name ## _c -# endif - -# if defined(OPUS_X86_MAY_HAVE_AVX) -# define MAY_HAVE_AVX(name) name ## _avx -# else -# define MAY_HAVE_AVX(name) name ## _c -# endif - -# if defined(OPUS_HAVE_RTCD) -int opus_select_arch(void); -# endif - -/*gcc appears to emit MOVDQA's to load the argument of an _mm_cvtepi8_epi32() - or _mm_cvtepi16_epi32() when optimizations are disabled, even though the - actual PMOVSXWD instruction takes an m32 or m64. Unlike a normal memory - reference, these require 16-byte alignment and load a full 16 bytes (instead - of 4 or 8), possibly reading out of bounds. - - We can insert an explicit MOVD or MOVQ using _mm_cvtsi32_si128() or - _mm_loadl_epi64(), which should have the same semantics as an m32 or m64 - reference in the PMOVSXWD instruction itself, but gcc is not smart enough to - optimize this out when optimizations ARE enabled. - - Clang, in contrast, requires us to do this always for _mm_cvtepi8_epi32 - (which is fair, since technically the compiler is always allowed to do the - dereference before invoking the function implementing the intrinsic). - However, it is smart enough to eliminate the extra MOVD instruction. - For _mm_cvtepi16_epi32, it does the right thing, though does *not* optimize out - the extra MOVQ if it's specified explicitly */ - -# if defined(__clang__) || !defined(__OPTIMIZE__) -# define OP_CVTEPI8_EPI32_M32(x) \ - (_mm_cvtepi8_epi32(_mm_cvtsi32_si128(*(int *)(x)))) -# else -# define OP_CVTEPI8_EPI32_M32(x) \ - (_mm_cvtepi8_epi32(*(__m128i *)(x))) -#endif - -# if !defined(__OPTIMIZE__) -# define OP_CVTEPI16_EPI32_M64(x) \ - (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x)))) -# else -# define OP_CVTEPI16_EPI32_M64(x) \ - (_mm_cvtepi16_epi32(*(__m128i *)(x))) -# endif - -#endif diff --git a/code/opus-1.2.1/silk/x86/VQ_WMat_EC_sse.c b/code/opus-1.2.1/silk/x86/VQ_WMat_EC_sse.c deleted file mode 100644 index 74d6c6d0..00000000 --- a/code/opus-1.2.1/silk/x86/VQ_WMat_EC_sse.c +++ /dev/null @@ -1,142 +0,0 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang - - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include -#include -#include -#include "main.h" -#include "celt/x86/x86cpu.h" - -/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */ -void silk_VQ_WMat_EC_sse4_1( - opus_int8 *ind, /* O index of best codebook vector */ - opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ - opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ - const opus_int16 *in_Q14, /* I input vector to be quantized */ - const opus_int32 *W_Q18, /* I weighting matrix */ - const opus_int8 *cb_Q7, /* I codebook */ - const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ - const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ - const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ - const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ - opus_int L /* I number of vectors in codebook */ -) -{ - opus_int k, gain_tmp_Q7; - const opus_int8 *cb_row_Q7; - opus_int16 diff_Q14[ 5 ]; - opus_int32 sum1_Q14, sum2_Q16; - - __m128i C_tmp1, C_tmp2, C_tmp3, C_tmp4, C_tmp5; - /* Loop over codebook */ - *rate_dist_Q14 = silk_int32_MAX; - cb_row_Q7 = cb_Q7; - for( k = 0; k < L; k++ ) { - gain_tmp_Q7 = cb_gain_Q7[k]; - - diff_Q14[ 0 ] = in_Q14[ 0 ] - silk_LSHIFT( cb_row_Q7[ 0 ], 7 ); - - C_tmp1 = OP_CVTEPI16_EPI32_M64( &in_Q14[ 1 ] ); - C_tmp2 = OP_CVTEPI8_EPI32_M32( &cb_row_Q7[ 1 ] ); - C_tmp2 = _mm_slli_epi32( C_tmp2, 7 ); - C_tmp1 = _mm_sub_epi32( C_tmp1, C_tmp2 ); - - diff_Q14[ 1 ] = _mm_extract_epi16( C_tmp1, 0 ); - diff_Q14[ 2 ] = _mm_extract_epi16( C_tmp1, 2 ); - diff_Q14[ 3 ] = _mm_extract_epi16( C_tmp1, 4 ); - diff_Q14[ 4 ] = _mm_extract_epi16( C_tmp1, 6 ); - - /* Weighted rate */ - sum1_Q14 = silk_SMULBB( mu_Q9, cl_Q5[ k ] ); - - /* Penalty for too large gain */ - sum1_Q14 = silk_ADD_LSHIFT32( sum1_Q14, silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 10 ); - - silk_assert( sum1_Q14 >= 0 ); - - /* first row of W_Q18 */ - C_tmp3 = _mm_loadu_si128( (__m128i *)(&W_Q18[ 1 ] ) ); - C_tmp4 = _mm_mul_epi32( C_tmp3, C_tmp1 ); - C_tmp4 = _mm_srli_si128( C_tmp4, 2 ); - - C_tmp1 = _mm_shuffle_epi32( C_tmp1, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* shift right 4 bytes */ - C_tmp3 = _mm_shuffle_epi32( C_tmp3, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* shift right 4 bytes */ - - C_tmp5 = _mm_mul_epi32( C_tmp3, C_tmp1 ); - C_tmp5 = _mm_srli_si128( C_tmp5, 2 ); - - C_tmp5 = _mm_add_epi32( C_tmp4, C_tmp5 ); - C_tmp5 = _mm_slli_epi32( C_tmp5, 1 ); - - C_tmp5 = _mm_add_epi32( C_tmp5, _mm_shuffle_epi32( C_tmp5, _MM_SHUFFLE( 0, 0, 0, 2 ) ) ); - sum2_Q16 = _mm_cvtsi128_si32( C_tmp5 ); - - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 0 ], diff_Q14[ 0 ] ); - sum1_Q14 = silk_SMLAWB( sum1_Q14, sum2_Q16, diff_Q14[ 0 ] ); - - /* second row of W_Q18 */ - sum2_Q16 = silk_SMULWB( W_Q18[ 7 ], diff_Q14[ 2 ] ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 8 ], diff_Q14[ 3 ] ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 9 ], diff_Q14[ 4 ] ); - sum2_Q16 = silk_LSHIFT( sum2_Q16, 1 ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 6 ], diff_Q14[ 1 ] ); - sum1_Q14 = silk_SMLAWB( sum1_Q14, sum2_Q16, diff_Q14[ 1 ] ); - - /* third row of W_Q18 */ - sum2_Q16 = silk_SMULWB( W_Q18[ 13 ], diff_Q14[ 3 ] ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 14 ], diff_Q14[ 4 ] ); - sum2_Q16 = silk_LSHIFT( sum2_Q16, 1 ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 12 ], diff_Q14[ 2 ] ); - sum1_Q14 = silk_SMLAWB( sum1_Q14, sum2_Q16, diff_Q14[ 2 ] ); - - /* fourth row of W_Q18 */ - sum2_Q16 = silk_SMULWB( W_Q18[ 19 ], diff_Q14[ 4 ] ); - sum2_Q16 = silk_LSHIFT( sum2_Q16, 1 ); - sum2_Q16 = silk_SMLAWB( sum2_Q16, W_Q18[ 18 ], diff_Q14[ 3 ] ); - sum1_Q14 = silk_SMLAWB( sum1_Q14, sum2_Q16, diff_Q14[ 3 ] ); - - /* last row of W_Q18 */ - sum2_Q16 = silk_SMULWB( W_Q18[ 24 ], diff_Q14[ 4 ] ); - sum1_Q14 = silk_SMLAWB( sum1_Q14, sum2_Q16, diff_Q14[ 4 ] ); - - silk_assert( sum1_Q14 >= 0 ); - - /* find best */ - if( sum1_Q14 < *rate_dist_Q14 ) { - *rate_dist_Q14 = sum1_Q14; - *ind = (opus_int8)k; - *gain_Q7 = gain_tmp_Q7; - } - - /* Go to next cbk vector */ - cb_row_Q7 += LTP_ORDER; - } -} diff --git a/code/opus-1.2.1/silk/x86/main_sse.h b/code/opus-1.2.1/silk/x86/main_sse.h deleted file mode 100644 index 2f15d448..00000000 --- a/code/opus-1.2.1/silk/x86/main_sse.h +++ /dev/null @@ -1,248 +0,0 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang - - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifndef MAIN_SSE_H -#define MAIN_SSE_H - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -# if defined(OPUS_X86_MAY_HAVE_SSE4_1) - -#if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ -# define OVERRIDE_silk_VQ_WMat_EC - -void silk_VQ_WMat_EC_sse4_1( - opus_int8 *ind, /* O index of best codebook vector */ - opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ - opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ - const opus_int16 *in_Q14, /* I input vector to be quantized */ - const opus_int32 *W_Q18, /* I weighting matrix */ - const opus_int8 *cb_Q7, /* I codebook */ - const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ - const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ - const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ - const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ - opus_int L /* I number of vectors in codebook */ -); - -#if defined OPUS_X86_PRESUME_SSE4_1 - -#define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ - mu_Q9, max_gain_Q7, L, arch) \ - ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ - mu_Q9, max_gain_Q7, L)) - -#else - -extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( - opus_int8 *ind, /* O index of best codebook vector */ - opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ - opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ - const opus_int16 *in_Q14, /* I input vector to be quantized */ - const opus_int32 *W_Q18, /* I weighting matrix */ - const opus_int8 *cb_Q7, /* I codebook */ - const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ - const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ - const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ - const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ - opus_int L /* I number of vectors in codebook */ -); - -# define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ - mu_Q9, max_gain_Q7, L, arch) \ - ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ - mu_Q9, max_gain_Q7, L)) - -#endif -#endif - -#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ -# define OVERRIDE_silk_NSQ - -void silk_NSQ_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ -); - -#if defined OPUS_X86_PRESUME_SSE4_1 - -#define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ - ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) - -#else - -extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ -); - -# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ - ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) - -#endif - -# define OVERRIDE_silk_NSQ_del_dec - -void silk_NSQ_del_dec_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ -); - -#if defined OPUS_X86_PRESUME_SSE4_1 - -#define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ - ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) - -#else - -extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ -); - -# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ - ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ - HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) - -#endif -#endif - -void silk_noise_shape_quantizer( - silk_nsq_state *NSQ, /* I/O NSQ state */ - opus_int signalType, /* I Signal type */ - const opus_int32 x_sc_Q10[], /* I */ - opus_int8 pulses[], /* O */ - opus_int16 xq[], /* O */ - opus_int32 sLTP_Q15[], /* I/O LTP state */ - const opus_int16 a_Q12[], /* I Short term prediction coefs */ - const opus_int16 b_Q14[], /* I Long term prediction coefs */ - const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ - opus_int lag, /* I Pitch lag */ - opus_int32 HarmShapeFIRPacked_Q14, /* I */ - opus_int Tilt_Q14, /* I Spectral tilt */ - opus_int32 LF_shp_Q14, /* I */ - opus_int32 Gain_Q16, /* I */ - opus_int Lambda_Q10, /* I */ - opus_int offset_Q10, /* I */ - opus_int length, /* I Input length */ - opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ - opus_int predictLPCOrder, /* I Prediction filter order */ - int arch /* I Architecture */ -); - -/**************************/ -/* Noise level estimation */ -/**************************/ -void silk_VAD_GetNoiseLevels( - const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ - silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ -); - -# define OVERRIDE_silk_VAD_GetSA_Q8 - -opus_int silk_VAD_GetSA_Q8_sse4_1( - silk_encoder_state *psEnC, - const opus_int16 pIn[] -); - -#if defined(OPUS_X86_PRESUME_SSE4_1) -#define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) - -#else - -# define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ - ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) - -extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( - silk_encoder_state *psEnC, - const opus_int16 pIn[]); - -#endif - -# endif -#endif diff --git a/code/opus-1.2.1/src/mlp.c b/code/opus-1.2.1/src/mlp.c deleted file mode 100644 index ff9e50df..00000000 --- a/code/opus-1.2.1/src/mlp.c +++ /dev/null @@ -1,145 +0,0 @@ -/* Copyright (c) 2008-2011 Octasic Inc. - Written by Jean-Marc Valin */ -/* - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR - CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include "opus_types.h" -#include "opus_defines.h" - -#include -#include "mlp.h" -#include "arch.h" -#include "tansig_table.h" -#define MAX_NEURONS 100 - -#if 0 -static OPUS_INLINE opus_val16 tansig_approx(opus_val32 _x) /* Q19 */ -{ - int i; - opus_val16 xx; /* Q11 */ - /*double x, y;*/ - opus_val16 dy, yy; /* Q14 */ - /*x = 1.9073e-06*_x;*/ - if (_x>=QCONST32(8,19)) - return QCONST32(1.,14); - if (_x<=-QCONST32(8,19)) - return -QCONST32(1.,14); - xx = EXTRACT16(SHR32(_x, 8)); - /*i = lrint(25*x);*/ - i = SHR32(ADD32(1024,MULT16_16(25, xx)),11); - /*x -= .04*i;*/ - xx -= EXTRACT16(SHR32(MULT16_16(20972,i),8)); - /*x = xx*(1./2048);*/ - /*y = tansig_table[250+i];*/ - yy = tansig_table[250+i]; - /*y = yy*(1./16384);*/ - dy = 16384-MULT16_16_Q14(yy,yy); - yy = yy + MULT16_16_Q14(MULT16_16_Q11(xx,dy),(16384 - MULT16_16_Q11(yy,xx))); - return yy; -} -#else -/*extern const float tansig_table[501];*/ -static OPUS_INLINE float tansig_approx(float x) -{ - int i; - float y, dy; - float sign=1; - /* Tests are reversed to catch NaNs */ - if (!(x<8)) - return 1; - if (!(x>-8)) - return -1; -#ifndef FIXED_POINT - /* Another check in case of -ffast-math */ - if (celt_isnan(x)) - return 0; -#endif - if (x<0) - { - x=-x; - sign=-1; - } - i = (int)floor(.5f+25*x); - x -= .04f*i; - y = tansig_table[i]; - dy = 1-y*y; - y = y + x*dy*(1 - y*x); - return sign*y; -} -#endif - -#if 0 -void mlp_process(const MLP *m, const opus_val16 *in, opus_val16 *out) -{ - int j; - opus_val16 hidden[MAX_NEURONS]; - const opus_val16 *W = m->weights; - /* Copy to tmp_in */ - for (j=0;jtopo[1];j++) - { - int k; - opus_val32 sum = SHL32(EXTEND32(*W++),8); - for (k=0;ktopo[0];k++) - sum = MAC16_16(sum, in[k],*W++); - hidden[j] = tansig_approx(sum); - } - for (j=0;jtopo[2];j++) - { - int k; - opus_val32 sum = SHL32(EXTEND32(*W++),14); - for (k=0;ktopo[1];k++) - sum = MAC16_16(sum, hidden[k], *W++); - out[j] = tansig_approx(EXTRACT16(PSHR32(sum,17))); - } -} -#else -void mlp_process(const MLP *m, const float *in, float *out) -{ - int j; - float hidden[MAX_NEURONS]; - const float *W = m->weights; - /* Copy to tmp_in */ - for (j=0;jtopo[1];j++) - { - int k; - float sum = *W++; - for (k=0;ktopo[0];k++) - sum = sum + in[k]**W++; - hidden[j] = tansig_approx(sum); - } - for (j=0;jtopo[2];j++) - { - int k; - float sum = *W++; - for (k=0;ktopo[1];k++) - sum = sum + hidden[k]**W++; - out[j] = tansig_approx(sum); - } -} -#endif diff --git a/code/opus-1.2.1/src/mlp_data.c b/code/opus-1.2.1/src/mlp_data.c deleted file mode 100644 index a819880b..00000000 --- a/code/opus-1.2.1/src/mlp_data.c +++ /dev/null @@ -1,112 +0,0 @@ -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include "mlp.h" - -/* RMS error was 0.280492, seed was 1480478173 */ -/* 0.005976 0.031821 (0.280494 0.280492) done */ - -static const float weights[450] = { - -/* hidden layer */ --0.514624f, 0.0234227f, -0.14329f, -0.0878216f, -0.00187827f, --0.0257443f, 0.108524f, 0.00333881f, 0.00585017f, -0.0246132f, -0.142723f, -0.00436494f, 0.0101354f, -0.11124f, -0.0809367f, --0.0750772f, 0.0295524f, 0.00823944f, 0.150392f, 0.0320876f, --0.0710564f, -1.43818f, 0.652076f, 0.0650744f, -1.54821f, -0.168949f, -1.92724f, 0.0517976f, -0.0670737f, -0.0690121f, -0.00247528f, -0.0522024f, 0.0631368f, 0.0532776f, 0.047751f, --0.011715f, 0.142374f, -0.0290885f, -0.279263f, -0.433499f, --0.0795174f, -0.380458f, -0.051263f, 0.218537f, -0.322478f, -1.06667f, -0.104607f, -4.70108f, 0.312037f, 0.277397f, --2.71859f, 1.70037f, -0.141845f, 0.0115618f, 0.0629883f, -0.0403871f, 0.0139428f, -0.00430733f, -0.0429038f, -0.0590318f, --0.0501526f, -0.0284802f, -0.0415686f, -0.0438999f, 0.0822666f, -0.197194f, 0.0363275f, -0.0584307f, 0.0752364f, -0.0799796f, --0.146275f, 0.161661f, -0.184585f, 0.145568f, 0.442823f, -1.61221f, 1.11162f, 2.62177f, -2.482f, -0.112599f, --0.110366f, -0.140794f, -0.181694f, 0.0648674f, 0.0842248f, -0.0933993f, 0.150122f, 0.129171f, 0.176848f, 0.141758f, --0.271822f, 0.235113f, 0.0668579f, -0.433957f, 0.113633f, --0.169348f, -1.40091f, 0.62861f, -0.134236f, 0.402173f, -1.86373f, 1.53998f, -4.32084f, 0.735343f, 0.800214f, --0.00968415f, 0.0425904f, 0.0196811f, -0.018426f, -0.000343953f, --0.00416389f, 0.00111558f, 0.0173069f, -0.00998596f, -0.025898f, -0.00123764f, -0.00520373f, -0.0565033f, 0.0637394f, 0.0051213f, -0.0221361f, 0.00819962f, -0.0467061f, -0.0548258f, -0.00314063f, --1.18332f, 1.88091f, -0.41148f, -2.95727f, -0.521449f, --0.271641f, 0.124946f, -0.0532936f, 0.101515f, 0.000208564f, --0.0488748f, 0.0642388f, -0.0383848f, 0.0135046f, -0.0413592f, --0.0326402f, -0.0137421f, -0.0225219f, -0.0917294f, -0.277759f, --0.185418f, 0.0471128f, -0.125879f, 0.262467f, -0.212794f, --0.112931f, -1.99885f, -0.404787f, 0.224402f, 0.637962f, --0.27808f, -0.0723953f, -0.0537655f, -0.0336359f, -0.0906601f, --0.0641309f, -0.0713542f, 0.0524317f, 0.00608819f, 0.0754101f, --0.0488401f, -0.00671865f, 0.0418239f, 0.0536284f, -0.132639f, -0.0267648f, -0.248432f, -0.0104153f, 0.035544f, -0.212753f, --0.302895f, -0.0357854f, 0.376838f, 0.597025f, -0.664647f, -0.268422f, -0.376772f, -1.05472f, 0.0144178f, 0.179122f, -0.0360155f, 0.220262f, -0.0056381f, 0.0317197f, 0.0621066f, --0.00779298f, 0.00789378f, 0.00350605f, 0.0104809f, 0.0362871f, --0.157708f, -0.0659779f, -0.0926278f, 0.00770791f, 0.0631621f, -0.0817343f, -0.424295f, -0.0437727f, -0.24251f, 0.711217f, --0.736455f, -2.194f, -0.107612f, -0.175156f, -0.0366573f, --0.0123156f, -0.0628516f, -0.0218977f, -0.00693699f, 0.00695185f, -0.00507362f, 0.00359334f, 0.0052661f, 0.035561f, 0.0382701f, -0.0342179f, -0.00790271f, -0.0170925f, 0.047029f, 0.0197362f, --0.0153435f, 0.0644152f, -0.36862f, -0.0674876f, -2.82672f, -1.34122f, -0.0788029f, -3.47792f, 0.507246f, -0.816378f, --0.0142383f, -0.127349f, -0.106926f, -0.0359524f, 0.105045f, -0.291554f, 0.195413f, 0.0866214f, -0.066577f, -0.102188f, -0.0979466f, -0.12982f, 0.400181f, -0.409336f, -0.0593326f, --0.0656203f, -0.204474f, 0.179802f, 0.000509084f, 0.0995954f, --2.377f, -0.686359f, 0.934861f, 1.10261f, 1.3901f, --4.33616f, -0.00264017f, 0.00713045f, 0.106264f, 0.143726f, --0.0685305f, -0.054656f, -0.0176725f, -0.0772669f, -0.0264526f, --0.0103824f, -0.0269872f, -0.00687f, 0.225804f, 0.407751f, --0.0612611f, -0.0576863f, -0.180131f, -0.222772f, -0.461742f, -0.335236f, 1.03399f, 4.24112f, -0.345796f, -0.594549f, --76.1407f, -0.265276f, 0.0507719f, 0.0643044f, 0.0384832f, -0.0424459f, -0.0387817f, -0.0235996f, -0.0740556f, -0.0270029f, -0.00882177f, -0.0552371f, -0.00485851f, 0.314295f, 0.360431f, --0.0787085f, 0.110355f, -0.415958f, -0.385088f, -0.272224f, --1.55108f, -0.141848f, 0.448877f, -0.563447f, -2.31403f, --0.120077f, -1.49918f, -0.817726f, -0.0495854f, -0.0230782f, --0.0224014f, 0.117076f, 0.0393216f, 0.051997f, 0.0330763f, --0.110796f, 0.0211117f, -0.0197258f, 0.0187461f, 0.0125183f, -0.14876f, 0.0920565f, -0.342475f, 0.135272f, -0.168155f, --0.033423f, -0.0604611f, -0.128835f, 0.664947f, -0.144997f, -2.27649f, 1.28663f, 0.841217f, -2.42807f, 0.0230471f, -0.226709f, -0.0374803f, 0.155436f, 0.0400342f, -0.184686f, -0.128488f, -0.0939518f, -0.0578559f, 0.0265967f, -0.0999322f, --0.0322768f, -0.322994f, -0.189371f, -0.738069f, -0.0754914f, -0.214717f, -0.093728f, -0.695741f, 0.0899298f, -2.06188f, --0.273719f, -0.896977f, 0.130553f, 0.134638f, 1.29355f, -0.00520749f, -0.0324224f, 0.00530451f, 0.0192385f, 0.00328708f, -0.0250838f, 0.0053365f, -0.0177321f, 0.00618789f, 0.00525364f, -0.00104596f, -0.0360459f, 0.0402403f, -0.0406351f, 0.0136883f, -0.0880722f, -0.0197449f, 0.089938f, 0.0100456f, -0.0475638f, --0.73267f, 0.037433f, -0.146551f, -0.230221f, -3.06489f, --1.40194f, 0.0198483f, 0.0397953f, -0.0190239f, 0.0470715f, --0.131363f, -0.191721f, -0.0176224f, -0.0480352f, -0.221799f, --0.26794f, -0.0292615f, 0.0612127f, -0.129877f, 0.00628332f, --0.085918f, 0.0175379f, 0.0541011f, -0.0810874f, -0.380809f, --0.222056f, -0.508859f, -0.473369f, 0.484958f, -2.28411f, -0.0139516f, -/* output layer */ -3.90017f, 1.71789f, -1.43372f, -2.70839f, 1.77107f, -5.48006f, 1.44661f, 2.01134f, -1.88383f, -3.64958f, --1.26351f, 0.779421f, 2.11357f, 3.10409f, 1.68846f, --4.46197f, -1.61455f, 3.59832f, 2.43531f, -1.26458f, -0.417941f, 1.47437f, 2.16635f, -1.909f, -0.828869f, -1.38805f, -2.67975f, -0.110044f, 1.95596f, 0.697931f, --0.313226f, -0.889315f, 0.283236f, 0.946102f, }; - -static const int topo[3] = {25, 16, 2}; - -const MLP net = { - 3, - topo, - weights -}; diff --git a/code/opus-1.2.1/src/tansig_table.h b/code/opus-1.2.1/src/tansig_table.h deleted file mode 100644 index c76f844a..00000000 --- a/code/opus-1.2.1/src/tansig_table.h +++ /dev/null @@ -1,45 +0,0 @@ -/* This file is auto-generated by gen_tables */ - -static const float tansig_table[201] = { -0.000000f, 0.039979f, 0.079830f, 0.119427f, 0.158649f, -0.197375f, 0.235496f, 0.272905f, 0.309507f, 0.345214f, -0.379949f, 0.413644f, 0.446244f, 0.477700f, 0.507977f, -0.537050f, 0.564900f, 0.591519f, 0.616909f, 0.641077f, -0.664037f, 0.685809f, 0.706419f, 0.725897f, 0.744277f, -0.761594f, 0.777888f, 0.793199f, 0.807569f, 0.821040f, -0.833655f, 0.845456f, 0.856485f, 0.866784f, 0.876393f, -0.885352f, 0.893698f, 0.901468f, 0.908698f, 0.915420f, -0.921669f, 0.927473f, 0.932862f, 0.937863f, 0.942503f, -0.946806f, 0.950795f, 0.954492f, 0.957917f, 0.961090f, -0.964028f, 0.966747f, 0.969265f, 0.971594f, 0.973749f, -0.975743f, 0.977587f, 0.979293f, 0.980869f, 0.982327f, -0.983675f, 0.984921f, 0.986072f, 0.987136f, 0.988119f, -0.989027f, 0.989867f, 0.990642f, 0.991359f, 0.992020f, -0.992631f, 0.993196f, 0.993718f, 0.994199f, 0.994644f, -0.995055f, 0.995434f, 0.995784f, 0.996108f, 0.996407f, -0.996682f, 0.996937f, 0.997172f, 0.997389f, 0.997590f, -0.997775f, 0.997946f, 0.998104f, 0.998249f, 0.998384f, -0.998508f, 0.998623f, 0.998728f, 0.998826f, 0.998916f, -0.999000f, 0.999076f, 0.999147f, 0.999213f, 0.999273f, -0.999329f, 0.999381f, 0.999428f, 0.999472f, 0.999513f, -0.999550f, 0.999585f, 0.999617f, 0.999646f, 0.999673f, -0.999699f, 0.999722f, 0.999743f, 0.999763f, 0.999781f, -0.999798f, 0.999813f, 0.999828f, 0.999841f, 0.999853f, -0.999865f, 0.999875f, 0.999885f, 0.999893f, 0.999902f, -0.999909f, 0.999916f, 0.999923f, 0.999929f, 0.999934f, -0.999939f, 0.999944f, 0.999948f, 0.999952f, 0.999956f, -0.999959f, 0.999962f, 0.999965f, 0.999968f, 0.999970f, -0.999973f, 0.999975f, 0.999977f, 0.999978f, 0.999980f, -0.999982f, 0.999983f, 0.999984f, 0.999986f, 0.999987f, -0.999988f, 0.999989f, 0.999990f, 0.999990f, 0.999991f, -0.999992f, 0.999992f, 0.999993f, 0.999994f, 0.999994f, -0.999994f, 0.999995f, 0.999995f, 0.999996f, 0.999996f, -0.999996f, 0.999997f, 0.999997f, 0.999997f, 0.999997f, -0.999997f, 0.999998f, 0.999998f, 0.999998f, 0.999998f, -0.999998f, 0.999998f, 0.999999f, 0.999999f, 0.999999f, -0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, -0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, -1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, -1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, -1.000000f, -}; diff --git a/code/opus-1.2.1/celt/_kiss_fft_guts.h b/code/opus-1.5.2/celt/_kiss_fft_guts.h similarity index 100% rename from code/opus-1.2.1/celt/_kiss_fft_guts.h rename to code/opus-1.5.2/celt/_kiss_fft_guts.h diff --git a/code/opus-1.2.1/celt/arch.h b/code/opus-1.5.2/celt/arch.h similarity index 88% rename from code/opus-1.2.1/celt/arch.h rename to code/opus-1.5.2/celt/arch.h index d1e6457c..3845c3a0 100644 --- a/code/opus-1.2.1/celt/arch.h +++ b/code/opus-1.5.2/celt/arch.h @@ -56,23 +56,43 @@ #define CELT_SIG_SCALE 32768.f -#define celt_fatal(str) _celt_fatal(str, __FILE__, __LINE__); -#ifdef ENABLE_ASSERTIONS +#define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__); + +#if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING) +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +void celt_fatal(const char *str, const char *file, int line); + +#if defined(CELT_C) && !defined(OVERRIDE_celt_fatal) #include #include #ifdef __GNUC__ __attribute__((noreturn)) #endif -static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line) +void celt_fatal(const char *str, const char *file, int line) { fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); +#if defined(_MSC_VER) + _set_abort_behavior( 0, _WRITE_ABORT_MSG); +#endif abort(); } -#define celt_assert(cond) {if (!(cond)) {celt_fatal("assertion failed: " #cond);}} -#define celt_assert2(cond, message) {if (!(cond)) {celt_fatal("assertion failed: " #cond "\n" message);}} +#endif + +#define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}} +#define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}} +#define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK) #else #define celt_assert(cond) #define celt_assert2(cond, message) +#define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0) +#endif + +#if defined(ENABLE_ASSERTIONS) +#define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}} +#else +#define celt_sig_assert(cond) #endif #define IMUL32(a,b) ((a)*(b)) @@ -107,6 +127,8 @@ typedef opus_val32 celt_sig; typedef opus_val16 celt_norm; typedef opus_val32 celt_ener; +#define celt_isnan(x) 0 + #define Q15ONE 32767 #define SIG_SHIFT 12 @@ -141,7 +163,7 @@ static OPUS_INLINE opus_int16 SAT16(opus_int32 x) { #ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR #include "arm/fixed_arm64.h" -#elif OPUS_ARM_INLINE_EDSP +#elif defined (OPUS_ARM_INLINE_EDSP) #include "arm/fixed_armv5e.h" #elif defined (OPUS_ARM_INLINE_ASM) #include "arm/fixed_armv4.h" diff --git a/code/opus-1.2.1/celt/arm/arm_celt_map.c b/code/opus-1.5.2/celt/arm/arm_celt_map.c similarity index 85% rename from code/opus-1.2.1/celt/arm/arm_celt_map.c rename to code/opus-1.5.2/celt/arm/arm_celt_map.c index ca988b66..cbaea495 100644 --- a/code/opus-1.2.1/celt/arm/arm_celt_map.c +++ b/code/opus-1.5.2/celt/arm/arm_celt_map.c @@ -40,7 +40,8 @@ opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, c celt_inner_prod_c, /* ARMv4 */ celt_inner_prod_c, /* EDSP */ celt_inner_prod_c, /* Media */ - celt_inner_prod_neon /* NEON */ + celt_inner_prod_neon,/* NEON */ + celt_inner_prod_neon /* DOTPROD */ }; void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, @@ -48,7 +49,8 @@ void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const o dual_inner_prod_c, /* ARMv4 */ dual_inner_prod_c, /* EDSP */ dual_inner_prod_c, /* Media */ - dual_inner_prod_neon /* NEON */ + dual_inner_prod_neon,/* NEON */ + dual_inner_prod_neon /* DOTPROD */ }; # endif @@ -61,7 +63,8 @@ opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, celt_pitch_xcorr_c, /* ARMv4 */ MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */ MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */ - MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */ + MAY_HAVE_NEON(celt_pitch_xcorr), /* NEON */ + MAY_HAVE_NEON(celt_pitch_xcorr) /* DOTPROD */ }; # endif @@ -72,7 +75,8 @@ void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, celt_pitch_xcorr_c, /* ARMv4 */ celt_pitch_xcorr_c, /* EDSP */ celt_pitch_xcorr_c, /* Media */ - celt_pitch_xcorr_float_neon /* Neon */ + celt_pitch_xcorr_float_neon, /* Neon */ + celt_pitch_xcorr_float_neon /* DOTPROD */ }; # endif # endif /* FIXED_POINT */ @@ -90,6 +94,7 @@ void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( xcorr_kernel_c, /* EDSP */ xcorr_kernel_c, /* Media */ xcorr_kernel_neon_fixed, /* Neon */ + xcorr_kernel_neon_fixed /* DOTPROD */ }; #endif @@ -101,14 +106,16 @@ int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { opus_fft_alloc_arch_c, /* ARMv4 */ opus_fft_alloc_arch_c, /* EDSP */ opus_fft_alloc_arch_c, /* Media */ - opus_fft_alloc_arm_neon /* Neon with NE10 library support */ + opus_fft_alloc_arm_neon, /* Neon with NE10 library support */ + opus_fft_alloc_arm_neon /* DOTPROD with NE10 library support */ }; void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { opus_fft_free_arch_c, /* ARMv4 */ opus_fft_free_arch_c, /* EDSP */ opus_fft_free_arch_c, /* Media */ - opus_fft_free_arm_neon /* Neon with NE10 */ + opus_fft_free_arm_neon, /* Neon with NE10 */ + opus_fft_free_arm_neon /* DOTPROD with NE10 */ }; # endif /* CUSTOM_MODES */ @@ -118,7 +125,8 @@ void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, opus_fft_c, /* ARMv4 */ opus_fft_c, /* EDSP */ opus_fft_c, /* Media */ - opus_fft_neon /* Neon with NE10 */ + opus_fft_neon, /* Neon with NE10 */ + opus_fft_neon /* DOTPROD with NE10 */ }; void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, @@ -127,7 +135,8 @@ void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, opus_ifft_c, /* ARMv4 */ opus_ifft_c, /* EDSP */ opus_ifft_c, /* Media */ - opus_ifft_neon /* Neon with NE10 */ + opus_ifft_neon, /* Neon with NE10 */ + opus_ifft_neon /* DOTPROD with NE10 */ }; void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, @@ -139,7 +148,8 @@ void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, clt_mdct_forward_c, /* ARMv4 */ clt_mdct_forward_c, /* EDSP */ clt_mdct_forward_c, /* Media */ - clt_mdct_forward_neon /* Neon with NE10 */ + clt_mdct_forward_neon, /* Neon with NE10 */ + clt_mdct_forward_neon /* DOTPROD with NE10 */ }; void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, @@ -151,7 +161,8 @@ void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, clt_mdct_backward_c, /* ARMv4 */ clt_mdct_backward_c, /* EDSP */ clt_mdct_backward_c, /* Media */ - clt_mdct_backward_neon /* Neon with NE10 */ + clt_mdct_backward_neon, /* Neon with NE10 */ + clt_mdct_backward_neon /* DOTPROD with NE10 */ }; # endif /* HAVE_ARM_NE10 */ diff --git a/code/opus-1.2.1/celt/arm/armcpu.c b/code/opus-1.5.2/celt/arm/armcpu.c similarity index 64% rename from code/opus-1.2.1/celt/arm/armcpu.c rename to code/opus-1.5.2/celt/arm/armcpu.c index 694a63b7..6785121a 100644 --- a/code/opus-1.2.1/celt/arm/armcpu.c +++ b/code/opus-1.5.2/celt/arm/armcpu.c @@ -43,6 +43,7 @@ #define OPUS_CPU_ARM_EDSP_FLAG (1< + +static opus_uint32 opus_cpu_capabilities(void) { opus_uint32 flags = 0; FILE *cpuinfo; @@ -124,6 +127,14 @@ opus_uint32 opus_cpu_capabilities(void) p = strstr(buf, " neon"); if(p != NULL && (p[5] == ' ' || p[5] == '\n')) flags |= OPUS_CPU_ARM_NEON_FLAG; + p = strstr(buf, " asimd"); + if(p != NULL && (p[6] == ' ' || p[6] == '\n')) + flags |= OPUS_CPU_ARM_NEON_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_EDSP_FLAG; +# endif +# if defined(OPUS_ARM_MAY_HAVE_DOTPROD) + p = strstr(buf, " asimddp"); + if(p != NULL && (p[8] == ' ' || p[8] == '\n')) + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; # endif } # endif @@ -142,10 +153,92 @@ opus_uint32 opus_cpu_capabilities(void) # endif } +#if defined(OPUS_ARM_PRESUME_AARCH64_NEON_INTR) + flags |= OPUS_CPU_ARM_EDSP_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_NEON_FLAG; +# if defined(OPUS_ARM_PRESUME_DOTPROD) + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; +# endif +#endif + fclose(cpuinfo); } return flags; } + +#elif defined(__APPLE__) +#include +#include + +static opus_uint32 opus_cpu_capabilities(void) +{ + opus_uint32 flags = 0; + +#if defined(OPUS_ARM_MAY_HAVE_DOTPROD) + size_t size = sizeof(uint32_t); + uint32_t value = 0; + if (!sysctlbyname("hw.optional.arm.FEAT_DotProd", &value, &size, NULL, 0) && value) + { + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; + } +#endif + +#if defined(OPUS_ARM_PRESUME_AARCH64_NEON_INTR) + flags |= OPUS_CPU_ARM_EDSP_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_NEON_FLAG; +# if defined(OPUS_ARM_PRESUME_DOTPROD) + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; +# endif +#endif + return flags; +} + +#elif defined(__FreeBSD__) +#include + +static opus_uint32 opus_cpu_capabilities(void) +{ + long hwcap = 0; + opus_uint32 flags = 0; + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + /* FreeBSD requires armv6+, which always supports media instructions */ + flags |= OPUS_CPU_ARM_MEDIA_FLAG; +# endif + + elf_aux_info(AT_HWCAP, &hwcap, sizeof hwcap); + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +# ifdef HWCAP_EDSP + if (hwcap & HWCAP_EDSP) + flags |= OPUS_CPU_ARM_EDSP_FLAG; +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +# ifdef HWCAP_NEON + if (hwcap & HWCAP_NEON) + flags |= OPUS_CPU_ARM_NEON_FLAG; +# elif defined(HWCAP_ASIMD) + if (hwcap & HWCAP_ASIMD) + flags |= OPUS_CPU_ARM_NEON_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_EDSP_FLAG; +# endif +# endif +# if defined(OPUS_ARM_MAY_HAVE_DOTPROD) && defined(HWCAP_ASIMDDP) + if (hwcap & HWCAP_ASIMDDP) + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; +# endif +# endif + +#if defined(OPUS_ARM_PRESUME_AARCH64_NEON_INTR) + flags |= OPUS_CPU_ARM_EDSP_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_NEON_FLAG; +# if defined(OPUS_ARM_PRESUME_DOTPROD) + flags |= OPUS_CPU_ARM_DOTPROD_FLAG; +# endif +#endif + + return (flags); +} + #else /* The feature registers which can tell us what the processor supports are * accessible in priveleged modes only, so we can't have a general user-space @@ -154,7 +247,7 @@ opus_uint32 opus_cpu_capabilities(void) "your platform. Reconfigure with --disable-rtcd (or send patches)." #endif -int opus_select_arch(void) +static int opus_select_arch_impl(void) { opus_uint32 flags = opus_cpu_capabilities(); int arch = 0; @@ -178,8 +271,21 @@ int opus_select_arch(void) } arch++; - celt_assert(arch == OPUS_ARCH_ARM_NEON); + if(!(flags & OPUS_CPU_ARM_DOTPROD_FLAG)) { + celt_assert(arch == OPUS_ARCH_ARM_NEON); + return arch; + } + arch++; + + celt_assert(arch == OPUS_ARCH_ARM_DOTPROD); return arch; } +int opus_select_arch(void) { + int arch = opus_select_arch_impl(); +#ifdef FUZZING + arch = rand()%(arch+1); +#endif + return arch; +} #endif diff --git a/code/opus-1.2.1/celt/arm/armcpu.h b/code/opus-1.5.2/celt/arm/armcpu.h similarity index 87% rename from code/opus-1.2.1/celt/arm/armcpu.h rename to code/opus-1.5.2/celt/arm/armcpu.h index 820262ff..6d5803d8 100644 --- a/code/opus-1.2.1/celt/arm/armcpu.h +++ b/code/opus-1.5.2/celt/arm/armcpu.h @@ -46,6 +46,12 @@ # define MAY_HAVE_NEON(name) MAY_HAVE_MEDIA(name) # endif +# if defined(OPUS_ARM_MAY_HAVE_DOTPROD) +# define MAY_HAVE_DOTPROD(name) name ## _dotprod +# else +# define MAY_HAVE_DOTPROD(name) MAY_HAVE_NEON(name) +# endif + # if defined(OPUS_ARM_PRESUME_EDSP) # define PRESUME_EDSP(name) name ## _edsp # else @@ -64,6 +70,12 @@ # define PRESUME_NEON(name) PRESUME_MEDIA(name) # endif +# if defined(OPUS_ARM_PRESUME_DOTPROD) +# define PRESUME_DOTPROD(name) name ## _dotprod +# else +# define PRESUME_DOTPROD(name) PRESUME_NEON(name) +# endif + # if defined(OPUS_HAVE_RTCD) int opus_select_arch(void); @@ -71,6 +83,7 @@ int opus_select_arch(void); #define OPUS_ARCH_ARM_EDSP (1) #define OPUS_ARCH_ARM_MEDIA (2) #define OPUS_ARCH_ARM_NEON (3) +#define OPUS_ARCH_ARM_DOTPROD (4) # endif diff --git a/code/opus-1.2.1/celt/arm/celt_ne10_fft.c b/code/opus-1.5.2/celt/arm/celt_fft_ne10.c similarity index 99% rename from code/opus-1.2.1/celt/arm/celt_ne10_fft.c rename to code/opus-1.5.2/celt/arm/celt_fft_ne10.c index b8c60abe..ea5fd780 100644 --- a/code/opus-1.2.1/celt/arm/celt_ne10_fft.c +++ b/code/opus-1.5.2/celt/arm/celt_fft_ne10.c @@ -1,7 +1,7 @@ /* Copyright (c) 2015 Xiph.Org Foundation Written by Viswanath Puttagunta */ /** - @file celt_ne10_fft.c + @file celt_fft_ne10.c @brief ARM Neon optimizations for fft using NE10 library */ diff --git a/code/opus-1.2.1/celt/arm/celt_ne10_mdct.c b/code/opus-1.5.2/celt/arm/celt_mdct_ne10.c similarity index 99% rename from code/opus-1.2.1/celt/arm/celt_ne10_mdct.c rename to code/opus-1.5.2/celt/arm/celt_mdct_ne10.c index 293c3efd..3531d02d 100644 --- a/code/opus-1.2.1/celt/arm/celt_ne10_mdct.c +++ b/code/opus-1.5.2/celt/arm/celt_mdct_ne10.c @@ -1,7 +1,7 @@ /* Copyright (c) 2015 Xiph.Org Foundation Written by Viswanath Puttagunta */ /** - @file celt_ne10_mdct.c + @file celt_mdct_ne10.c @brief ARM Neon optimizations for mdct using NE10 library */ diff --git a/code/opus-1.2.1/celt/arm/celt_neon_intr.c b/code/opus-1.5.2/celt/arm/celt_neon_intr.c similarity index 71% rename from code/opus-1.2.1/celt/arm/celt_neon_intr.c rename to code/opus-1.5.2/celt/arm/celt_neon_intr.c index cf443988..250f8362 100644 --- a/code/opus-1.2.1/celt/arm/celt_neon_intr.c +++ b/code/opus-1.5.2/celt/arm/celt_neon_intr.c @@ -38,6 +38,8 @@ #include "../pitch.h" #if defined(FIXED_POINT) +#include + void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) { int j; @@ -47,7 +49,10 @@ void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_va int16x4_t y0 = vld1_s16(y); y += 4; - for (j = 0; j + 8 <= len; j += 8) + /* This loop loads one y value more than we actually need. + Therefore we have to stop as soon as there are 8 or fewer samples left + (instead of 7), to avoid reading past the end of the array. */ + for (j = 0; j + 8 < len; j += 8) { /* Load x[0...7] */ int16x8_t xx = vld1q_s16(x); @@ -80,23 +85,79 @@ void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_va x += 8; y += 8; } - - for (; j < len; j++) - { - int16x4_t x0 = vld1_dup_s16(x); /* load next x */ + if (j + 4 < len) { + /* Load x[0...3] */ + int16x4_t x0 = vld1_s16(x); + /* Load y[4...7] */ + int16x4_t y4 = vld1_s16(y); + int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0); + int16x4_t y1 = vext_s16(y0, y4, 1); + int32x4_t a1 = vmlal_lane_s16(a0, y1, x0, 1); + int16x4_t y2 = vext_s16(y0, y4, 2); + int32x4_t a2 = vmlal_lane_s16(a1, y2, x0, 2); + int16x4_t y3 = vext_s16(y0, y4, 3); + int32x4_t a3 = vmlal_lane_s16(a2, y3, x0, 3); + y0 = y4; + a = a3; + x += 4; + y += 4; + j += 4; + } + if (j + 2 < len) { + /* Load x[0...1] */ + int16x4x2_t xx = vld2_dup_s16(x); + int16x4_t x0 = xx.val[0]; + int16x4_t x1 = xx.val[1]; + /* Load y[4...5]. + We would like to use vld1_dup_s32(), but casting the pointer would + break strict aliasing rules and potentially have alignment issues. + Fortunately the compiler seems capable of translating this memcpy() + and vdup_n_s32() into the equivalent vld1_dup_s32().*/ + int32_t yy; + memcpy(&yy, y, sizeof(yy)); + int16x4_t y4 = vreinterpret_s16_s32(vdup_n_s32(yy)); int32x4_t a0 = vmlal_s16(a, y0, x0); - - int16x4_t y4 = vld1_dup_s16(y); /* load next y */ - y0 = vext_s16(y0, y4, 1); + int16x4_t y1 = vext_s16(y0, y4, 1); + /* Replace bottom copy of {y[5], y[4]} in y4 with {y[3], y[2]} from y0, + using VSRI instead of VEXT, since it's a data-processing + instruction. */ + y0 = vreinterpret_s16_s64(vsri_n_s64(vreinterpret_s64_s16(y4), + vreinterpret_s64_s16(y0), 32)); + int32x4_t a1 = vmlal_s16(a0, y1, x1); + a = a1; + x += 2; + y += 2; + j += 2; + } + if (j + 1 < len) { + /* Load next x. */ + int16x4_t x0 = vld1_dup_s16(x); + int32x4_t a0 = vmlal_s16(a, y0, x0); + /* Load last y. */ + int16x4_t y4 = vld1_dup_s16(y); + y0 = vreinterpret_s16_s64(vsri_n_s64(vreinterpret_s64_s16(y4), + vreinterpret_s64_s16(y0), 16)); a = a0; x++; - y++; } - - vst1q_s32(sum, a); + /* Load last x. */ + int16x4_t x0 = vld1_dup_s16(x); + int32x4_t a0 = vmlal_s16(a, y0, x0); + vst1q_s32(sum, a0); } #else + +#if defined(__ARM_FEATURE_FMA) && defined(__ARM_ARCH_ISA_A64) +/* If we can, force the compiler to use an FMA instruction rather than break + * vmlaq_f32() into fmul/fadd. */ +#ifdef vmlaq_lane_f32 +#undef vmlaq_lane_f32 +#endif +#define vmlaq_lane_f32(a,b,c,lane) vfmaq_lane_f32(a,b,c,lane) +#endif + + /* * Function: xcorr_kernel_neon_float * --------------------------------- @@ -196,7 +257,7 @@ void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y, int i; (void)arch; celt_assert(max_pitch > 0); - celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); + celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); for (i = 0; i < (max_pitch-3); i += 4) { xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i, diff --git a/code/opus-1.2.1/celt/arm/fft_arm.h b/code/opus-1.5.2/celt/arm/fft_arm.h similarity index 100% rename from code/opus-1.2.1/celt/arm/fft_arm.h rename to code/opus-1.5.2/celt/arm/fft_arm.h diff --git a/code/opus-1.2.1/celt/arm/fixed_arm64.h b/code/opus-1.5.2/celt/arm/fixed_arm64.h similarity index 100% rename from code/opus-1.2.1/celt/arm/fixed_arm64.h rename to code/opus-1.5.2/celt/arm/fixed_arm64.h diff --git a/code/opus-1.2.1/celt/arm/fixed_armv4.h b/code/opus-1.5.2/celt/arm/fixed_armv4.h similarity index 100% rename from code/opus-1.2.1/celt/arm/fixed_armv4.h rename to code/opus-1.5.2/celt/arm/fixed_armv4.h diff --git a/code/opus-1.2.1/celt/arm/fixed_armv5e.h b/code/opus-1.5.2/celt/arm/fixed_armv5e.h similarity index 100% rename from code/opus-1.2.1/celt/arm/fixed_armv5e.h rename to code/opus-1.5.2/celt/arm/fixed_armv5e.h diff --git a/code/opus-1.2.1/celt/arm/kiss_fft_armv4.h b/code/opus-1.5.2/celt/arm/kiss_fft_armv4.h similarity index 100% rename from code/opus-1.2.1/celt/arm/kiss_fft_armv4.h rename to code/opus-1.5.2/celt/arm/kiss_fft_armv4.h diff --git a/code/opus-1.2.1/celt/arm/kiss_fft_armv5e.h b/code/opus-1.5.2/celt/arm/kiss_fft_armv5e.h similarity index 100% rename from code/opus-1.2.1/celt/arm/kiss_fft_armv5e.h rename to code/opus-1.5.2/celt/arm/kiss_fft_armv5e.h diff --git a/code/opus-1.2.1/celt/arm/mdct_arm.h b/code/opus-1.5.2/celt/arm/mdct_arm.h similarity index 100% rename from code/opus-1.2.1/celt/arm/mdct_arm.h rename to code/opus-1.5.2/celt/arm/mdct_arm.h diff --git a/code/opus-1.2.1/celt/arm/pitch_arm.h b/code/opus-1.5.2/celt/arm/pitch_arm.h similarity index 95% rename from code/opus-1.2.1/celt/arm/pitch_arm.h rename to code/opus-1.5.2/celt/arm/pitch_arm.h index 4ee13bd0..bed8b04e 100644 --- a/code/opus-1.2.1/celt/arm/pitch_arm.h +++ b/code/opus-1.5.2/celt/arm/pitch_arm.h @@ -90,7 +90,9 @@ extern opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, const opus_val16 *, opus_val32 *, int, int, int); # define OVERRIDE_PITCH_XCORR (1) -# define celt_pitch_xcorr (*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK]) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) # elif defined(OPUS_ARM_PRESUME_EDSP) || \ defined(OPUS_ARM_PRESUME_MEDIA) || \ @@ -142,7 +144,9 @@ extern void const opus_val16 *, opus_val32 *, int, int, int); # define OVERRIDE_PITCH_XCORR (1) -# define celt_pitch_xcorr (*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK]) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) # elif defined(OPUS_ARM_PRESUME_NEON_INTR) diff --git a/code/opus-1.2.1/celt/arm/pitch_neon_intr.c b/code/opus-1.5.2/celt/arm/pitch_neon_intr.c similarity index 88% rename from code/opus-1.2.1/celt/arm/pitch_neon_intr.c rename to code/opus-1.5.2/celt/arm/pitch_neon_intr.c index 1ac38c43..43885f52 100644 --- a/code/opus-1.2.1/celt/arm/pitch_neon_intr.c +++ b/code/opus-1.5.2/celt/arm/pitch_neon_intr.c @@ -130,6 +130,13 @@ void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus /* ========================================================================== */ +#ifdef __ARM_FEATURE_FMA +/* If we can, force the compiler to use an FMA instruction rather than break + vmlaq_f32() into fmul/fadd. */ +#define vmlaq_f32(a,b,c) vfmaq_f32(a,b,c) +#endif + + #ifdef OPUS_CHECK_ASM /* This part of code simulates floating-point NEON operations. */ @@ -137,22 +144,27 @@ void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus /* celt_inner_prod_neon_float_c_simulation() simulates the floating-point */ /* operations of celt_inner_prod_neon(), and both functions should have bit */ /* exact output. */ -static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y, int N) +static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y, float *err, int N) { int i; + *err = 0; opus_val32 xy, xy0 = 0, xy1 = 0, xy2 = 0, xy3 = 0; for (i = 0; i < N - 3; i += 4) { xy0 = MAC16_16(xy0, x[i + 0], y[i + 0]); xy1 = MAC16_16(xy1, x[i + 1], y[i + 1]); xy2 = MAC16_16(xy2, x[i + 2], y[i + 2]); xy3 = MAC16_16(xy3, x[i + 3], y[i + 3]); + *err += ABS32(xy0)+ABS32(xy1)+ABS32(xy2)+ABS32(xy3); } xy0 += xy2; xy1 += xy3; xy = xy0 + xy1; + *err += ABS32(xy1)+ABS32(xy0)+ABS32(xy); for (; i < N; i++) { xy = MAC16_16(xy, x[i], y[i]); + *err += ABS32(xy); } + *err = *err*2e-7 + N*1e-37; return xy; } @@ -160,32 +172,10 @@ static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, c /* operations of dual_inner_prod_neon(), and both functions should have bit */ /* exact output. */ static void dual_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, - int N, opus_val32 *xy1, opus_val32 *xy2) + int N, opus_val32 *xy1, opus_val32 *xy2, float *err) { - int i; - opus_val32 xy01, xy02, xy01_0 = 0, xy01_1 = 0, xy01_2 = 0, xy01_3 = 0, xy02_0 = 0, xy02_1 = 0, xy02_2 = 0, xy02_3 = 0; - for (i = 0; i < N - 3; i += 4) { - xy01_0 = MAC16_16(xy01_0, x[i + 0], y01[i + 0]); - xy01_1 = MAC16_16(xy01_1, x[i + 1], y01[i + 1]); - xy01_2 = MAC16_16(xy01_2, x[i + 2], y01[i + 2]); - xy01_3 = MAC16_16(xy01_3, x[i + 3], y01[i + 3]); - xy02_0 = MAC16_16(xy02_0, x[i + 0], y02[i + 0]); - xy02_1 = MAC16_16(xy02_1, x[i + 1], y02[i + 1]); - xy02_2 = MAC16_16(xy02_2, x[i + 2], y02[i + 2]); - xy02_3 = MAC16_16(xy02_3, x[i + 3], y02[i + 3]); - } - xy01_0 += xy01_2; - xy02_0 += xy02_2; - xy01_1 += xy01_3; - xy02_1 += xy02_3; - xy01 = xy01_0 + xy01_1; - xy02 = xy02_0 + xy02_1; - for (; i < N; i++) { - xy01 = MAC16_16(xy01, x[i], y01[i]); - xy02 = MAC16_16(xy02, x[i], y02[i]); - } - *xy1 = xy01; - *xy2 = xy02; + *xy1 = celt_inner_prod_neon_float_c_simulation(x, y01, &err[0], N); + *xy2 = celt_inner_prod_neon_float_c_simulation(x, y02, &err[1], N); } #endif /* OPUS_CHECK_ASM */ @@ -225,7 +215,12 @@ opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N) } #ifdef OPUS_CHECK_ASM - celt_assert(ABS32(celt_inner_prod_neon_float_c_simulation(x, y, N) - xy) <= VERY_SMALL); + { + float err, res; + res = celt_inner_prod_neon_float_c_simulation(x, y, &err, N); + /*if (ABS32(res - xy) > err) fprintf(stderr, "%g %g %g\n", res, xy, err);*/ + celt_assert(ABS32(res - xy) <= err); + } #endif return xy; @@ -280,9 +275,12 @@ void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus #ifdef OPUS_CHECK_ASM { opus_val32 xy1_c, xy2_c; - dual_inner_prod_neon_float_c_simulation(x, y01, y02, N, &xy1_c, &xy2_c); - celt_assert(ABS32(xy1_c - *xy1) <= VERY_SMALL); - celt_assert(ABS32(xy2_c - *xy2) <= VERY_SMALL); + float err[2]; + dual_inner_prod_neon_float_c_simulation(x, y01, y02, N, &xy1_c, &xy2_c, err); + /*if (ABS32(xy1_c - *xy1) > err[0]) fprintf(stderr, "dual1 fail: %g %g %g\n", xy1_c, *xy1, err[0]); + if (ABS32(xy2_c - *xy2) > err[1]) fprintf(stderr, "dual2 fail: %g %g %g\n", xy2_c, *xy2, err[1]);*/ + celt_assert(ABS32(xy1_c - *xy1) <= err[0]); + celt_assert(ABS32(xy2_c - *xy2) <= err[1]); } #endif } diff --git a/code/opus-1.2.1/celt/bands.c b/code/opus-1.5.2/celt/bands.c similarity index 98% rename from code/opus-1.2.1/celt/bands.c rename to code/opus-1.5.2/celt/bands.c index 3b1f5cfb..6785e08e 100644 --- a/code/opus-1.2.1/celt/bands.c +++ b/code/opus-1.5.2/celt/bands.c @@ -70,10 +70,10 @@ opus_int16 bitexact_cos(opus_int16 x) opus_int32 tmp; opus_int16 x2; tmp = (4096+((opus_int32)(x)*(x)))>>13; - celt_assert(tmp<=32767); + celt_sig_assert(tmp<=32767); x2 = tmp; x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2))))); - celt_assert(x2<=32766); + celt_sig_assert(x2<=32766); return 1+x2; } @@ -282,7 +282,7 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas N0 = m->eBands[i+1]-m->eBands[i]; /* depth in 1/8 bits */ - celt_assert(pulses[i]>=0); + celt_sig_assert(pulses[i]>=0); depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM; #ifdef FIXED_POINT @@ -371,14 +371,14 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas static void compute_channel_weights(celt_ener Ex, celt_ener Ey, opus_val16 w[2]) { celt_ener minE; -#if FIXED_POINT +#ifdef FIXED_POINT int shift; #endif minE = MIN32(Ex, Ey); /* Adjustment to make the weights a bit more conservative. */ Ex = ADD32(Ex, minE/3); Ey = ADD32(Ey, minE/3); -#if FIXED_POINT +#ifdef FIXED_POINT shift = celt_ilog2(EPSILON+MAX32(Ex, Ey))-14; #endif w[0] = VSHR32(Ex, shift); @@ -478,7 +478,7 @@ static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT /* Decide whether we should spread the pulses in the current frame */ int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, int last_decision, int *hf_average, int *tapset_decision, int update_hf, - int end, int C, int M) + int end, int C, int M, const int *spread_weight) { int i, c, N0; int sum = 0, nbBands=0; @@ -519,8 +519,8 @@ int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, if (i>m->nbEBands-4) hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N); tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N); - sum += tmp*256; - nbBands++; + sum += tmp*spread_weight[i]; + nbBands+=spread_weight[i]; } } while (++c0); /* end has to be non-zero */ celt_assert(sum>=0); - sum = celt_udiv(sum, nbBands); + sum = celt_udiv((opus_int32)sum<<8, nbBands); /* Recursive averaging */ sum = (sum+*average)>>1; *average = sum; @@ -901,7 +901,7 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, sctx->itheta = itheta; sctx->qalloc = qalloc; } -static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, int b, +static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, celt_norm *lowband_out) { int c; @@ -926,7 +926,6 @@ static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, sign = ec_dec_bits(ec, 1); } ctx->remaining_bits -= 1<resynth) x[0] = sign ? -NORM_SCALING : NORM_SCALING; @@ -1134,7 +1133,7 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, /* Special case for one sample */ if (N==1) { - return quant_band_n1(ctx, X, NULL, b, lowband_out); + return quant_band_n1(ctx, X, NULL, lowband_out); } if (tf_change>0) @@ -1256,7 +1255,7 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm /* Special case for one sample */ if (N==1) { - return quant_band_n1(ctx, X, Y, b, lowband_out); + return quant_band_n1(ctx, X, Y, lowband_out); } orig_fill = fill; @@ -1381,6 +1380,7 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm return cm; } +#ifndef DISABLE_UPDATE_DRAFT static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm *norm2, int start, int M, int dual_stereo) { int n1, n2; @@ -1393,6 +1393,7 @@ static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm if (dual_stereo) OPUS_COPY(&norm2[n1], &norm2[2*n1 - n2], n2-n1); } +#endif void quant_all_bands(int encode, const CELTMode *m, int start, int end, celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, @@ -1449,7 +1450,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, if (encode && resynth) lowband_scratch = _lowband_scratch; else - lowband_scratch = X_+M*eBands[m->nbEBands-1]; + lowband_scratch = X_+M*eBands[m->effEBands-1]; ALLOC(X_save, resynth_alloc, celt_norm); ALLOC(Y_save, resynth_alloc, celt_norm); ALLOC(X_save2, resynth_alloc, celt_norm); @@ -1492,6 +1493,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, else Y = NULL; N = M*eBands[i+1]-M*eBands[i]; + celt_assert(N > 0); tell = ec_tell_frac(ec); /* Compute how many bits we want to allocate to this band */ @@ -1507,7 +1509,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, b = 0; } -#ifdef ENABLE_UPDATE_DRAFT +#ifndef DISABLE_UPDATE_DRAFT if (resynth && (M*eBands[i]-N >= M*eBands[start] || i==start+1) && (update_lowband || lowband_offset==0)) lowband_offset = i; if (i == start+1) @@ -1541,7 +1543,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, fold_start = lowband_offset; while(M*eBands[--fold_start] > effective_lowband+norm_offset); fold_end = lowband_offset-1; -#ifdef ENABLE_UPDATE_DRAFT +#ifndef DISABLE_UPDATE_DRAFT while(++fold_end < i && M*eBands[fold_end] < effective_lowband+norm_offset+N); #else while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); @@ -1621,8 +1623,10 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, ctx = ctx_save; OPUS_COPY(X, X_save, N); OPUS_COPY(Y, Y_save, N); +#ifndef DISABLE_UPDATE_DRAFT if (i == start+1) special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); +#endif /* Encode and round up. */ ctx.theta_round = 1; x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, diff --git a/code/opus-1.2.1/celt/bands.h b/code/opus-1.5.2/celt/bands.h similarity index 98% rename from code/opus-1.2.1/celt/bands.h rename to code/opus-1.5.2/celt/bands.h index 2488c185..422b32cf 100644 --- a/code/opus-1.2.1/celt/bands.h +++ b/code/opus-1.5.2/celt/bands.h @@ -72,7 +72,7 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, int last_decision, int *hf_average, int *tapset_decision, int update_hf, - int end, int C, int M); + int end, int C, int M, const int *spread_weight); #ifdef MEASURE_NORM_MSE void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C); diff --git a/code/opus-1.2.1/celt/celt.c b/code/opus-1.5.2/celt/celt.c similarity index 100% rename from code/opus-1.2.1/celt/celt.c rename to code/opus-1.5.2/celt/celt.c diff --git a/code/opus-1.2.1/celt/celt.h b/code/opus-1.5.2/celt/celt.h similarity index 91% rename from code/opus-1.2.1/celt/celt.h rename to code/opus-1.5.2/celt/celt.h index 70175301..2f501951 100644 --- a/code/opus-1.2.1/celt/celt.h +++ b/code/opus-1.5.2/celt/celt.h @@ -42,6 +42,10 @@ #include "entdec.h" #include "arch.h" +#ifdef ENABLE_DEEP_PLC +#include "lpcnet.h" +#endif + #ifdef __cplusplus extern "C" { #endif @@ -59,9 +63,11 @@ typedef struct { float noisiness; float activity; float music_prob; - float vad_prob; + float music_prob_min; + float music_prob_max; int bandwidth; float activity_probability; + float max_pitch_ratio; /* Store as Q6 char to save space. */ unsigned char leak_boost[LEAK_BANDS]; } AnalysisInfo; @@ -147,6 +153,13 @@ int celt_decoder_get_size(int channels); int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels); +int celt_decode_with_ec_dred(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum +#ifdef ENABLE_DEEP_PLC + ,LPCNetPLCState *lpcnet +#endif + ); + int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum); @@ -207,6 +220,13 @@ static OPUS_INLINE int fromOpus(unsigned char c) extern const signed char tf_select_table[4][8]; +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +void validate_celt_decoder(CELTDecoder *st); +#define VALIDATE_CELT_DECODER(st) validate_celt_decoder(st) +#else +#define VALIDATE_CELT_DECODER(st) +#endif + int resampling_factor(opus_int32 rate); void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, @@ -216,23 +236,13 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, const opus_val16 *window, int overlap, int arch); -#ifdef NON_STATIC_COMB_FILTER_CONST_C -void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, - opus_val16 g10, opus_val16 g11, opus_val16 g12); -#endif - -#ifndef OVERRIDE_COMB_FILTER_CONST -# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ - ((void)(arch),comb_filter_const_c(y, x, T, N, g10, g11, g12)) -#endif - void init_caps(const CELTMode *m,int *cap,int LM,int C); #ifdef RESYNTH -void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem); +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, int accum); void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[], opus_val16 *oldBandE, int start, int effEnd, int C, int CC, int isTransient, - int LM, int downsample, int silence); + int LM, int downsample, int silence, int arch); #endif #ifdef __cplusplus diff --git a/code/opus-1.2.1/celt/celt_decoder.c b/code/opus-1.5.2/celt/celt_decoder.c similarity index 72% rename from code/opus-1.2.1/celt/celt_decoder.c rename to code/opus-1.5.2/celt/celt_decoder.c index 567d7456..743c2031 100644 --- a/code/opus-1.2.1/celt/celt_decoder.c +++ b/code/opus-1.5.2/celt/celt_decoder.c @@ -51,9 +51,19 @@ #include "celt_lpc.h" #include "vq.h" -#if defined(SMALL_FOOTPRINT) && defined(FIXED_POINT) -#define NORM_ALIASING_HACK +#ifdef ENABLE_DEEP_PLC +#include "lpcnet.h" +#include "lpcnet_private.h" #endif + +/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save + CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The + current value corresponds to a pitch of 66.67 Hz. */ +#define PLC_PITCH_LAG_MAX (720) +/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a + pitch of 480 Hz. */ +#define PLC_PITCH_LAG_MIN (100) + /**********************************************************************/ /* */ /* DECODER */ @@ -61,6 +71,9 @@ /**********************************************************************/ #define DECODE_BUFFER_SIZE 2048 +#define PLC_UPDATE_FRAMES 4 +#define PLC_UPDATE_SAMPLES (PLC_UPDATE_FRAMES*FRAME_SIZE) + /** Decoder state @brief Decoder state */ @@ -74,6 +87,7 @@ struct OpusCustomDecoder { int start, end; int signalling; int disable_inv; + int complexity; int arch; /* Everything beyond this point gets cleared on a reset */ @@ -82,7 +96,7 @@ struct OpusCustomDecoder { opus_uint32 rng; int error; int last_pitch_index; - int loss_count; + int loss_duration; int skip_plc; int postfilter_period; int postfilter_period_old; @@ -90,17 +104,62 @@ struct OpusCustomDecoder { opus_val16 postfilter_gain_old; int postfilter_tapset; int postfilter_tapset_old; + int prefilter_and_fold; celt_sig preemph_memD[2]; +#ifdef ENABLE_DEEP_PLC + opus_int16 plc_pcm[PLC_UPDATE_SAMPLES]; + int plc_fill; + float plc_preemphasis_mem; +#endif + celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ - /* opus_val16 lpc[], Size = channels*LPC_ORDER */ + /* opus_val16 lpc[], Size = channels*CELT_LPC_ORDER */ /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ }; +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +/* Make basic checks on the CELT state to ensure we don't end + up writing all over memory. */ +void validate_celt_decoder(CELTDecoder *st) +{ +#ifndef CUSTOM_MODES + celt_assert(st->mode == opus_custom_mode_create(48000, 960, NULL)); + celt_assert(st->overlap == 120); + celt_assert(st->end <= 21); +#else +/* From Section 4.3 in the spec: "The normal CELT layer uses 21 of those bands, + though Opus Custom (see Section 6.2) may use a different number of bands" + + Check if it's within the maximum number of Bark frequency bands instead */ + celt_assert(st->end <= 25); +#endif + celt_assert(st->channels == 1 || st->channels == 2); + celt_assert(st->stream_channels == 1 || st->stream_channels == 2); + celt_assert(st->downsample > 0); + celt_assert(st->start == 0 || st->start == 17); + celt_assert(st->start < st->end); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif + celt_assert(st->last_pitch_index <= PLC_PITCH_LAG_MAX); + celt_assert(st->last_pitch_index >= PLC_PITCH_LAG_MIN || st->last_pitch_index == 0); + celt_assert(st->postfilter_period < MAX_PERIOD); + celt_assert(st->postfilter_period >= COMBFILTER_MINPERIOD || st->postfilter_period == 0); + celt_assert(st->postfilter_period_old < MAX_PERIOD); + celt_assert(st->postfilter_period_old >= COMBFILTER_MINPERIOD || st->postfilter_period_old == 0); + celt_assert(st->postfilter_tapset <= 2); + celt_assert(st->postfilter_tapset >= 0); + celt_assert(st->postfilter_tapset_old <= 2); + celt_assert(st->postfilter_tapset_old >= 0); +} +#endif + int celt_decoder_get_size(int channels) { const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); @@ -111,7 +170,7 @@ OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_get_size(const CELTMode *mode, int { int size = sizeof(struct CELTDecoder) + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) - + channels*LPC_ORDER*sizeof(opus_val16) + + channels*CELT_LPC_ORDER*sizeof(opus_val16) + 4*2*mode->nbEBands*sizeof(opus_val16); return size; } @@ -164,7 +223,7 @@ OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMod st->start = 0; st->end = st->mode->effEBands; st->signalling = 1; -#ifdef ENABLE_UPDATE_DRAFT +#ifndef DISABLE_UPDATE_DRAFT st->disable_inv = channels == 1; #else st->disable_inv = 0; @@ -437,14 +496,6 @@ static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, } } -/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save - CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The - current value corresponds to a pitch of 66.67 Hz. */ -#define PLC_PITCH_LAG_MAX (720) -/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a - pitch of 480 Hz. */ -#define PLC_PITCH_LAG_MIN (100) - static int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch) { int pitch_index; @@ -461,7 +512,100 @@ static int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch) return pitch_index; } -static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) +static void prefilter_and_fold(CELTDecoder * OPUS_RESTRICT st, int N) +{ + int c; + int CC; + int i; + int overlap; + celt_sig *decode_mem[2]; + const OpusCustomMode *mode; + VARDECL(opus_val32, etmp); + mode = st->mode; + overlap = st->overlap; + CC = st->channels; + ALLOC(etmp, overlap, opus_val32); + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + } while (++cpostfilter_period_old, st->postfilter_period, overlap, + -st->postfilter_gain_old, -st->postfilter_gain, + st->postfilter_tapset_old, st->postfilter_tapset, NULL, 0, st->arch); + + /* Simulate TDAC on the concealed audio so that it blends with the + MDCT of the next frame. */ + for (i=0;iwindow[i], etmp[overlap-1-i]) + + MULT16_32_Q15(mode->window[overlap-i-1], etmp[i]); + } + } while (++cfec_read_pos; + tmp_fec_skip = lpcnet->fec_skip; + for (i=0;ifec_read_pos = tmp_read_post; + lpcnet->fec_skip = tmp_fec_skip; +} +#endif + +static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM +#ifdef ENABLE_DEEP_PLC + ,LPCNetPLCState *lpcnet +#endif + ) { int c; int i; @@ -474,7 +618,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) int nbEBands; int overlap; int start; - int loss_count; + int loss_duration; int noise_based; const opus_int16 *eBands; SAVE_STACK; @@ -489,22 +633,22 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; } while (++c_decode_mem+(DECODE_BUFFER_SIZE+overlap)*C); - oldBandE = lpc+C*LPC_ORDER; + oldBandE = lpc+C*CELT_LPC_ORDER; oldLogE = oldBandE + 2*nbEBands; oldLogE2 = oldLogE + 2*nbEBands; backgroundLogE = oldLogE2 + 2*nbEBands; - loss_count = st->loss_count; + loss_duration = st->loss_duration; start = st->start; - noise_based = loss_count >= 5 || start != 0 || st->skip_plc; +#ifdef ENABLE_DEEP_PLC + noise_based = start != 0 || (lpcnet->fec_fill_pos == 0 && (st->skip_plc || loss_duration >= 80)); +#else + noise_based = loss_duration >= 40 || start != 0 || st->skip_plc; +#endif if (noise_based) { /* Noise-based PLC/CNG */ -#ifdef NORM_ALIASING_HACK - celt_norm *X; -#else VARDECL(celt_norm, X); -#endif opus_uint32 seed; int end; int effEnd; @@ -512,16 +656,18 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) end = st->end; effEnd = IMAX(start, IMIN(end, mode->effEBands)); -#ifdef NORM_ALIASING_HACK - /* This is an ugly hack that breaks aliasing rules and would be easily broken, - but it saves almost 4kB of stack. */ - X = (celt_norm*)(out_syn[C-1]+overlap/2); -#else ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ -#endif + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, + DECODE_BUFFER_SIZE-N+overlap); + } while (++cprefilter_and_fold) { + prefilter_and_fold(st, N); + } /* Energy decay */ - decay = loss_count==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); + decay = loss_duration==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); c=0; do { for (i=start;irng = seed; - c=0; do { - OPUS_MOVE(decode_mem[c], decode_mem[c]+N, - DECODE_BUFFER_SIZE-N+(overlap>>1)); - } while (++cdownsample, 0, st->arch); + st->prefilter_and_fold = 0; + /* Skip regular PLC until we get two consecutive packets. */ + st->skip_plc = 1; } else { + int exc_length; /* Pitch-based PLC */ const opus_val16 *window; opus_val16 *exc; opus_val16 fade = Q15ONE; int pitch_index; - VARDECL(opus_val32, etmp); VARDECL(opus_val16, _exc); + VARDECL(opus_val16, fir_tmp); - if (loss_count == 0) + if (loss_duration == 0) { +#ifdef ENABLE_DEEP_PLC + if (lpcnet->loaded) update_plc_state(lpcnet, decode_mem, &st->plc_preemphasis_mem, C); +#endif st->last_pitch_index = pitch_index = celt_plc_pitch_search(decode_mem, C, st->arch); } else { pitch_index = st->last_pitch_index; fade = QCONST16(.8f,15); } - ALLOC(etmp, overlap, opus_val32); - ALLOC(_exc, MAX_PERIOD+LPC_ORDER, opus_val16); - exc = _exc+LPC_ORDER; + /* We want the excitation for 2 pitch periods in order to look for a + decaying signal, but we can't get more than MAX_PERIOD. */ + exc_length = IMIN(2*pitch_index, MAX_PERIOD); + + ALLOC(_exc, MAX_PERIOD+CELT_LPC_ORDER, opus_val16); + ALLOC(fir_tmp, exc_length, opus_val16); + exc = _exc+CELT_LPC_ORDER; window = mode->window; c=0; do { opus_val16 decay; @@ -581,21 +733,19 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) celt_sig *buf; int extrapolation_offset; int extrapolation_len; - int exc_length; int j; buf = decode_mem[c]; - for (i=0;iarch); + CELT_LPC_ORDER, MAX_PERIOD, st->arch); /* Add a noise floor of -40 dB. */ #ifdef FIXED_POINT ac[0] += SHR32(ac[0],13); @@ -603,7 +753,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) ac[0] *= 1.0001f; #endif /* Use lag windowing to stabilize the Levinson-Durbin recursion. */ - for (i=1;i<=LPC_ORDER;i++) + for (i=1;i<=CELT_LPC_ORDER;i++) { /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ #ifdef FIXED_POINT @@ -612,7 +762,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) ac[i] -= ac[i]*(0.008f*0.008f)*i*i; #endif } - _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); + _celt_lpc(lpc+c*CELT_LPC_ORDER, ac, CELT_LPC_ORDER); #ifdef FIXED_POINT /* For fixed-point, apply bandwidth expansion until we can guarantee that no overflow can happen in the IIR filter. This means: @@ -620,31 +770,25 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) while (1) { opus_val16 tmp=Q15ONE; opus_val32 sum=QCONST16(1., SIG_SHIFT); - for (i=0;iarch); + /* Compute the excitation for exc_length samples before the loss. We need the copy + because celt_fir() cannot filter in-place. */ + celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*CELT_LPC_ORDER, + fir_tmp, exc_length, CELT_LPC_ORDER, st->arch); + OPUS_COPY(exc+MAX_PERIOD-exc_length, fir_tmp, exc_length); } /* Check if the waveform is decaying, and if so how fast. @@ -695,21 +839,21 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) exc[extrapolation_offset+j])), SIG_SHIFT); /* Compute the energy of the previously decoded signal whose excitation we're copying. */ - tmp = ROUND16( + tmp = SROUND16( buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j], SIG_SHIFT); S1 += SHR32(MULT16_16(tmp, tmp), 10); } { - opus_val16 lpc_mem[LPC_ORDER]; + opus_val16 lpc_mem[CELT_LPC_ORDER]; /* Copy the last decoded samples (prior to the overlap region) to synthesis filter memory so we can have a continuous signal. */ - for (i=0;iarch); #ifdef FIXED_POINT for (i=0; i < extrapolation_len; i++) @@ -724,7 +868,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) opus_val32 S2=0; for (i=0;ipostfilter_period, st->postfilter_period, overlap, - -st->postfilter_gain, -st->postfilter_gain, - st->postfilter_tapset, st->postfilter_tapset, NULL, 0, st->arch); - - /* Simulate TDAC on the concealed audio so that it blends with the - MDCT of the next frame. */ - for (i=0;iloaded && (st->complexity >= 5 || lpcnet->fec_fill_pos > 0)) { + float overlap_mem; + int samples_needed16k; + celt_sig *buf; + VARDECL(float, buf_copy); + buf = decode_mem[0]; + ALLOC(buf_copy, C*overlap, float); + c=0; do { + OPUS_COPY(buf_copy+c*overlap, &decode_mem[c][DECODE_BUFFER_SIZE-N], overlap); + } while (++cplc_fill = 0; + } + while (st->plc_fill < samples_needed16k) { + lpcnet_plc_conceal(lpcnet, &st->plc_pcm[st->plc_fill]); + st->plc_fill += FRAME_SIZE; + } + /* Resample to 48 kHz. */ + for (i=0;i<(N+overlap)/3;i++) { + int j; + float sum; + for (sum=0, j=0;j<17;j++) sum += 3*st->plc_pcm[i+j]*sinc_filter[3*j]; + buf[DECODE_BUFFER_SIZE-N+3*i] = sum; + for (sum=0, j=0;j<16;j++) sum += 3*st->plc_pcm[i+j+1]*sinc_filter[3*j+2]; + buf[DECODE_BUFFER_SIZE-N+3*i+1] = sum; + for (sum=0, j=0;j<16;j++) sum += 3*st->plc_pcm[i+j+1]*sinc_filter[3*j+1]; + buf[DECODE_BUFFER_SIZE-N+3*i+2] = sum; + } + OPUS_MOVE(st->plc_pcm, &st->plc_pcm[N/3], st->plc_fill-N/3); + st->plc_fill -= N/3; + for (i=0;iplc_preemphasis_mem; + st->plc_preemphasis_mem = tmp; + } + overlap_mem = st->plc_preemphasis_mem; + for (i=0;iprefilter_and_fold = 1; } - st->loss_count = loss_count+1; + /* Saturate to soemthing large to avoid wrap-around. */ + st->loss_duration = IMIN(10000, loss_duration+(1<mode; nbEBands = mode->nbEBands; overlap = mode->overlap; @@ -842,7 +1031,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat frame_size *= st->downsample; lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC); - oldBandE = lpc+CC*LPC_ORDER; + oldBandE = lpc+CC*CELT_LPC_ORDER; oldLogE = oldBandE + 2*nbEBands; oldLogE2 = oldLogE + 2*nbEBands; backgroundLogE = oldLogE2 + 2*nbEBands; @@ -896,15 +1085,25 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat if (data == NULL || len<=1) { - celt_decode_lost(st, N, LM); + celt_decode_lost(st, N, LM +#ifdef ENABLE_DEEP_PLC + , lpcnet +#endif + ); deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); RESTORE_STACK; return frame_size/st->downsample; } +#ifdef ENABLE_DEEP_PLC + else { + /* FIXME: This is a bit of a hack just to make sure opus_decode_native() knows we're no longer in PLC. */ + if (lpcnet) lpcnet->blend = 0; + } +#endif /* Check if there are at least two packets received consecutively before * turning on the pitch-based PLC */ - st->skip_plc = st->loss_count != 0; + if (st->loss_duration == 0) st->skip_plc = 0; if (dec == NULL) { @@ -967,6 +1166,36 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat /* Decode the global flags (first symbols in the stream) */ intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; + /* If recovering from packet loss, make sure we make the energy prediction safe to reduce the + risk of getting loud artifacts. */ + if (!intra_ener && st->loss_duration != 0) { + c=0; do + { + opus_val16 safety = 0; + int missing = IMIN(10, st->loss_duration>>LM); + if (LM==0) safety = QCONST16(1.5f,DB_SHIFT); + else if (LM==1) safety = QCONST16(.5f,DB_SHIFT); + for (i=start;iprefilter_and_fold) { + prefilter_and_fold(st, N); + } celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, CC, isTransient, LM, st->downsample, silence, st->arch); @@ -1102,25 +1327,21 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat if (C==1) OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); - /* In case start or end were to change */ if (!isTransient) { - opus_val16 max_background_increase; OPUS_COPY(oldLogE2, oldLogE, 2*nbEBands); OPUS_COPY(oldLogE, oldBandE, 2*nbEBands); - /* In normal circumstances, we only allow the noise floor to increase by - up to 2.4 dB/second, but when we're in DTX, we allow up to 6 dB - increase for each update.*/ - if (st->loss_count < 10) - max_background_increase = M*QCONST16(0.001f,DB_SHIFT); - else - max_background_increase = QCONST16(1.f,DB_SHIFT); - for (i=0;i<2*nbEBands;i++) - backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]); } else { for (i=0;i<2*nbEBands;i++) oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); } + /* In normal circumstances, we only allow the noise floor to increase by + up to 2.4 dB/second, but when we're in DTX we give the weight of + all missing packets to the update packet. */ + max_background_increase = IMIN(160, st->loss_duration+M)*QCONST16(0.001f,DB_SHIFT); + for (i=0;i<2*nbEBands;i++) + backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]); + /* In case start or end were to change */ c=0; do { for (i=0;irng = dec->rng; deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); - st->loss_count = 0; + st->loss_duration = 0; + st->prefilter_and_fold = 0; RESTORE_STACK; if (ec_tell(dec) > 8*len) return OPUS_INTERNAL_ERROR; @@ -1146,6 +1368,15 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat return frame_size/st->downsample; } +int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum) +{ + return celt_decode_with_ec_dred(st, data, len, pcm, frame_size, dec, accum +#ifdef ENABLE_DEEP_PLC + , NULL +#endif + ); +} #ifdef CUSTOM_MODES @@ -1219,6 +1450,26 @@ int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) va_start(ap, request); switch (request) { + case OPUS_SET_COMPLEXITY_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>10) + { + goto bad_arg; + } + st->complexity = value; + } + break; + case OPUS_GET_COMPLEXITY_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->complexity; + } + break; case CELT_SET_START_BAND_REQUEST: { opus_int32 value = va_arg(ap, opus_int32); @@ -1265,7 +1516,7 @@ int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) int i; opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); - oldBandE = lpc+st->channels*LPC_ORDER; + oldBandE = lpc+st->channels*CELT_LPC_ORDER; oldLogE = oldBandE + 2*st->mode->nbEBands; oldLogE2 = oldLogE + 2*st->mode->nbEBands; OPUS_CLEAR((char*)&st->DECODER_RESET_START, diff --git a/code/opus-1.2.1/celt/celt_encoder.c b/code/opus-1.5.2/celt/celt_encoder.c similarity index 92% rename from code/opus-1.2.1/celt/celt_encoder.c rename to code/opus-1.5.2/celt/celt_encoder.c index 053e5a3b..7f32a801 100644 --- a/code/opus-1.2.1/celt/celt_encoder.c +++ b/code/opus-1.5.2/celt/celt_encoder.c @@ -281,6 +281,9 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ for (i=0;i 50 && LM>=1 && !lfe) + { + /* Compute a really simple masking model to avoid taking into account completely masked + bands when computing the spreading decision. */ + VARDECL(opus_val16, mask); + VARDECL(opus_val16, sig); + ALLOC(mask, nbEBands, opus_val16); + ALLOC(sig, nbEBands, opus_val16); + for (i=0;i=0;i--) + mask[i] = MAX16(mask[i], mask[i+1] - QCONST16(3.f, DB_SHIFT)); + for (i=0;i> shift; + } + /*for (i=0;i= (30 + 5*LM) && !lfe) { int last=0; c=0;do @@ -999,30 +1057,38 @@ static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 opus_val16 offset; opus_val16 tmp; opus_val16 *f; + OPUS_COPY(bandLogE3, &bandLogE2[c*nbEBands], end); + if (LM==0) { + /* For 2.5 ms frames, the first 8 bands have just one bin, so the + energy is highly unreliable (high variance). For that reason, + we take the max with the previous energy so that at least 2 bins + are getting used. */ + for (i=0;i bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT)) + if (bandLogE3[i] > bandLogE3[i-1]+QCONST16(.5f,DB_SHIFT)) last=i; - f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]); + f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE3[i]); } for (i=last-1;i>=0;i--) - f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i])); + f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE3[i])); /* Combine with a median filter to avoid dynalloc triggering unnecessarily. The "offset" value controls how conservative we are -- a higher offset reduces the impact of the median filter and makes dynalloc use more bits. */ offset = QCONST16(1.f, DB_SHIFT); for (i=2;ivalid) + gain1 = (opus_val16)(gain1 * analysis->max_pitch_ratio); +#else + (void)analysis; +#endif /* Gain threshold for enabling the prefilter/postfilter */ pf_threshold = QCONST16(.2f,15); @@ -1362,6 +1444,8 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, VARDECL(int, pulses); VARDECL(int, cap); VARDECL(int, offsets); + VARDECL(int, importance); + VARDECL(int, spread_weight); VARDECL(int, fine_priority); VARDECL(int, tf_res); VARDECL(unsigned char, collapse_masks); @@ -1414,6 +1498,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, opus_int32 equiv_rate; int hybrid; int weak_transient = 0; + int enable_tf_analysis; VARDECL(opus_val16, surround_dynalloc); ALLOC_STACK; @@ -1454,7 +1539,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, tell0_frac=tell=1; nbFilledBytes=0; } else { - tell0_frac=tell=ec_tell_frac(enc); + tell0_frac=ec_tell_frac(enc); tell=ec_tell(enc); nbFilledBytes=(tell+4)>>3; } @@ -1503,13 +1588,16 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, vbr_rate = 0; tmp = st->bitrate*frame_size; if (tell>1) - tmp += tell; + tmp += tell*mode->Fs; if (st->bitrate!=OPUS_BITRATE_MAX) + { nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + ec_enc_shrink(enc, nbCompressedBytes); + } effectiveBytes = nbCompressedBytes - nbFilledBytes; } - equiv_rate = ((opus_int32)nbCompressedBytes*8*50 >> (3-LM)) - (40*C+20)*((400>>LM) - 50); + equiv_rate = ((opus_int32)nbCompressedBytes*8*50 << (3-LM)) - (40*C+20)*((400>>LM) - 50); if (st->bitrate != OPUS_BITRATE_MAX) equiv_rate = IMIN(equiv_rate, st->bitrate - (40*C+20)*((400>>LM) - 50)); @@ -1603,7 +1691,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, && st->complexity >= 5; prefilter_tapset = st->tapset_decision; - pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes); + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes, &st->analysis); if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) pitch_change = 1; @@ -1633,7 +1721,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, /* Reduces the likelihood of energy instability on fricatives at low bitrate in hybrid mode. It seems like we still want to have real transients on vowels though (small SILK quantization offset value). */ - int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.offset >= 100; + int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.signalType != 2; isTransient = transient_analysis(in, N+overlap, CC, &tf_estimate, &tf_chan, allow_weak_transients, &weak_transient); } @@ -1657,11 +1745,17 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch); compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); amp2Log2(mode, effEnd, end, bandE, bandLogE2, C); - for (i=0;iupsample, st->arch); + /* This should catch any NaN in the CELT input. Since we're not supposed to see any (they're filtered + at the Opus layer), just abort. */ + celt_assert(!celt_isnan(freq[0]) && (C==1 || !celt_isnan(freq[N]))); if (CC==2&&C==1) tf_chan = 0; compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); @@ -1791,8 +1885,11 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); amp2Log2(mode, effEnd, end, bandE, bandLogE, C); /* Compensate for the scaling of short vs long mdcts */ - for (i=0;i=15*C && !hybrid && st->complexity>=2 && !st->lfe; + + ALLOC(offsets, nbEBands, int); + ALLOC(importance, nbEBands, int); + ALLOC(spread_weight, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, oldBandE, nbEBands, start, end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis, importance, spread_weight); + ALLOC(tf_res, nbEBands, int); /* Disable variable tf resolution for hybrid and at very low bitrate */ - if (effectiveBytes>=15*C && !hybrid && st->complexity>=2 && !st->lfe) + if (enable_tf_analysis) { int lambda; - lambda = IMAX(5, 1280/effectiveBytes + 2); - tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan); + lambda = IMAX(80, 20480/effectiveBytes + 2); + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan, importance); for (i=effEnd;isilk_info.signalType != 2) { /* For low bitrate hybrid, we force temporal resolution to 5 ms rather than 2.5 ms. */ for (i=0;ispread_decision = spreading_decision(mode, X, &st->tonal_average, st->spread_decision, &st->hf_average, - &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M); + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M, spread_weight); } /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ @@ -1901,11 +2008,6 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); } - ALLOC(offsets, nbEBands, int); - - maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, start, end, C, offsets, - st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, - eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis); /* For LFE, everything interesting is in the first band */ if (st->lfe) offsets[0] = IMIN(8, effectiveBytes/3); @@ -2121,7 +2223,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, #endif if (st->lfe) signalBandwidth = 1; - codedBands = compute_allocation(mode, start, end, offsets, cap, + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); if (st->lastCodedBands) @@ -2170,7 +2272,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, if (anti_collapse_on) { anti_collapse(mode, X, collapse_masks, LM, C, N, - start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng, st->arch); } c=0; do { @@ -2189,15 +2291,15 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, - mode->window, overlap); + mode->window, overlap, st->arch); if (LM!=0) comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, - mode->window, overlap); + mode->window, overlap, st->arch); } while (++cupsample, mode->preemph, st->preemph_memD); + deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD, 0); st->prefilter_period_old = st->prefilter_period; st->prefilter_gain_old = st->prefilter_gain; st->prefilter_tapset_old = st->prefilter_tapset; diff --git a/code/opus-1.2.1/celt/celt_lpc.c b/code/opus-1.5.2/celt/celt_lpc.c similarity index 71% rename from code/opus-1.2.1/celt/celt_lpc.c rename to code/opus-1.5.2/celt/celt_lpc.c index 57bc5f3d..fabca65c 100644 --- a/code/opus-1.2.1/celt/celt_lpc.c +++ b/code/opus-1.5.2/celt/celt_lpc.c @@ -44,23 +44,27 @@ int p opus_val32 r; opus_val32 error = ac[0]; #ifdef FIXED_POINT - opus_val32 lpc[LPC_ORDER]; + opus_val32 lpc[CELT_LPC_ORDER]; #else float *lpc = _lpc; #endif OPUS_CLEAR(lpc, p); +#ifdef FIXED_POINT if (ac[0] != 0) +#else + if (ac[0] > 1e-10f) +#endif { for (i = 0; i < p; i++) { /* Sum up this iteration's reflection coefficient */ opus_val32 rr = 0; for (j = 0; j < i; j++) rr += MULT32_32_Q31(lpc[j],ac[i - j]); - rr += SHR32(ac[i + 1],3); - r = -frac_div32(SHL32(rr,3), error); + rr += SHR32(ac[i + 1],6); + r = -frac_div32(SHL32(rr,6), error); /* Update LPC coefficients and total error */ - lpc[i] = SHR32(r,3); + lpc[i] = SHR32(r,6); for (j = 0; j < (i+1)>>1; j++) { opus_val32 tmp1, tmp2; @@ -73,17 +77,61 @@ int p error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); /* Bail out once we get 30 dB gain */ #ifdef FIXED_POINT - if (error maxabs) { + maxabs = absval; + idx = i; + } + } + maxabs = PSHR32(maxabs, 13); /* Q25->Q12 */ + + if (maxabs > 32767) { + maxabs = MIN32(maxabs, 163838); + chirp_Q16 = QCONST32(0.999, 16) - DIV32(SHL32(maxabs - 32767, 14), + SHR32(MULT32_32_32(maxabs, idx + 1), 2)); + chirp_minus_one_Q16 = chirp_Q16 - 65536; + + /* Apply bandwidth expansion. */ + for (i = 0; i < p - 1; i++) { + lpc[i] = MULT32_32_Q16(chirp_Q16, lpc[i]); + chirp_Q16 += PSHR32(MULT32_32_32(chirp_Q16, chirp_minus_one_Q16), 16); + } + lpc[p - 1] = MULT32_32_Q16(chirp_Q16, lpc[p - 1]); + } else { + break; + } + } + + if (iter == 10) { + /* If the coeffs still do not fit into the 16 bit range after 10 iterations, + fall back to the A(z)=1 filter. */ + OPUS_CLEAR(lpc, p); + _lpc[0] = 4096; /* Q12 */ + } else { + for (i = 0; i < p; i++) { + _lpc[i] = EXTRACT16(PSHR32(lpc[i], 13)); /* Q25->Q12 */ + } + } + } #endif } @@ -99,7 +147,7 @@ void celt_fir_c( int i,j; VARDECL(opus_val16, rnum); SAVE_STACK; - + celt_assert(x != y); ALLOC(rnum, ord, opus_val16); for(i=0;i ARMv4 * arch[1] -> ARMv5E * arch[2] -> ARMv6 * arch[3] -> NEON + * arch[4] -> NEON+DOTPROD */ -#define OPUS_ARCHMASK 3 +#define OPUS_ARCHMASK 7 -#elif (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ +#elif defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ - (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX)) + (defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_PRESUME_AVX2))) #include "x86/x86cpu.h" /* We currently support 5 x86 variants: diff --git a/code/opus-1.2.1/celt/cwrs.c b/code/opus-1.5.2/celt/cwrs.c similarity index 99% rename from code/opus-1.2.1/celt/cwrs.c rename to code/opus-1.5.2/celt/cwrs.c index 9722f0ac..a552e4f0 100644 --- a/code/opus-1.2.1/celt/cwrs.c +++ b/code/opus-1.5.2/celt/cwrs.c @@ -482,7 +482,7 @@ static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y){ k0=_k; q=row[_n]; if(q>_i){ - celt_assert(p>q); + celt_sig_assert(p>q); _k=_n; do p=CELT_PVQ_U_ROW[--_k][_n]; while(p>_i); diff --git a/code/opus-1.2.1/celt/cwrs.h b/code/opus-1.5.2/celt/cwrs.h similarity index 100% rename from code/opus-1.2.1/celt/cwrs.h rename to code/opus-1.5.2/celt/cwrs.h diff --git a/code/opus-1.2.1/celt/ecintrin.h b/code/opus-1.5.2/celt/ecintrin.h similarity index 96% rename from code/opus-1.2.1/celt/ecintrin.h rename to code/opus-1.5.2/celt/ecintrin.h index 2263cff6..66a4c36e 100644 --- a/code/opus-1.2.1/celt/ecintrin.h +++ b/code/opus-1.5.2/celt/ecintrin.h @@ -49,7 +49,11 @@ This macro should only be used for implementing ec_ilog(), if it is defined. All other code should use EC_ILOG() instead.*/ #if defined(_MSC_VER) && (_MSC_VER >= 1400) +#if defined(_MSC_VER) && (_MSC_VER >= 1910) +# include /* Improve compiler throughput. */ +#else # include +#endif /*In _DEBUG mode this is not an intrinsic by default.*/ # pragma intrinsic(_BitScanReverse) diff --git a/code/opus-1.2.1/celt/entcode.c b/code/opus-1.5.2/celt/entcode.c similarity index 100% rename from code/opus-1.2.1/celt/entcode.c rename to code/opus-1.5.2/celt/entcode.c diff --git a/code/opus-1.2.1/celt/entcode.h b/code/opus-1.5.2/celt/entcode.h similarity index 99% rename from code/opus-1.2.1/celt/entcode.h rename to code/opus-1.5.2/celt/entcode.h index 13d6c84e..3763e3f2 100644 --- a/code/opus-1.2.1/celt/entcode.h +++ b/code/opus-1.5.2/celt/entcode.h @@ -122,7 +122,7 @@ opus_uint32 ec_tell_frac(ec_ctx *_this); /* Tested exhaustively for all n and for 1<=d<=256 */ static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) { - celt_assert(d>0); + celt_sig_assert(d>0); #ifdef USE_SMALL_DIV_TABLE if (d>256) return n/d; @@ -138,7 +138,7 @@ static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) { } static OPUS_INLINE opus_int32 celt_sudiv(opus_int32 n, opus_int32 d) { - celt_assert(d>0); + celt_sig_assert(d>0); #ifdef USE_SMALL_DIV_TABLE if (n<0) return -(opus_int32)celt_udiv(-n, d); diff --git a/code/opus-1.2.1/celt/entdec.c b/code/opus-1.5.2/celt/entdec.c similarity index 95% rename from code/opus-1.2.1/celt/entdec.c rename to code/opus-1.5.2/celt/entdec.c index 0b3433ed..027aa24b 100644 --- a/code/opus-1.2.1/celt/entdec.c +++ b/code/opus-1.5.2/celt/entdec.c @@ -195,6 +195,27 @@ int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb){ return ret; } +int ec_dec_icdf16(ec_dec *_this,const opus_uint16 *_icdf,unsigned _ftb){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + opus_uint32 t; + int ret; + s=_this->rng; + d=_this->val; + r=s>>_ftb; + ret=-1; + do{ + t=s; + s=IMUL32(r,_icdf[++ret]); + } + while(dval=d-s; + _this->rng=t-s; + ec_dec_normalize(_this); + return ret; +} + opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft){ unsigned ft; unsigned s; diff --git a/code/opus-1.2.1/celt/entdec.h b/code/opus-1.5.2/celt/entdec.h similarity index 89% rename from code/opus-1.2.1/celt/entdec.h rename to code/opus-1.5.2/celt/entdec.h index d8ab3187..c81f26fd 100644 --- a/code/opus-1.2.1/celt/entdec.h +++ b/code/opus-1.5.2/celt/entdec.h @@ -81,11 +81,21 @@ int ec_dec_bit_logp(ec_dec *_this,unsigned _logp); Return: The decoded symbol s.*/ int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb); +/*Decodes a symbol given an "inverse" CDF table. + No call to ec_dec_update() is necessary after this call. + _icdf: The "inverse" CDF, such that symbol s falls in the range + [s>0?ft-_icdf[s-1]:0,ft-_icdf[s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution. + Return: The decoded symbol s.*/ +int ec_dec_icdf16(ec_dec *_this,const opus_uint16 *_icdf,unsigned _ftb); + /*Extracts a raw unsigned integer with a non-power-of-2 range from the stream. The bits must have been encoded with ec_enc_uint(). No call to ec_dec_update() is necessary after this call. _ft: The number of integers that can be decoded (one more than the max). - This must be at least one, and no more than 2**32-1. + This must be at least 2, and no more than 2**32-1. Return: The decoded bits.*/ opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft); diff --git a/code/opus-1.2.1/celt/entenc.c b/code/opus-1.5.2/celt/entenc.c similarity index 96% rename from code/opus-1.2.1/celt/entenc.c rename to code/opus-1.5.2/celt/entenc.c index f1750d25..69c6f835 100644 --- a/code/opus-1.2.1/celt/entenc.c +++ b/code/opus-1.5.2/celt/entenc.c @@ -172,6 +172,17 @@ void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb){ ec_enc_normalize(_this); } +void ec_enc_icdf16(ec_enc *_this,int _s,const opus_uint16 *_icdf,unsigned _ftb){ + opus_uint32 r; + r=_this->rng>>_ftb; + if(_s>0){ + _this->val+=_this->rng-IMUL32(r,_icdf[_s-1]); + _this->rng=IMUL32(r,_icdf[_s-1]-_icdf[_s]); + } + else _this->rng-=IMUL32(r,_icdf[_s]); + ec_enc_normalize(_this); +} + void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft){ unsigned ft; unsigned fl; diff --git a/code/opus-1.2.1/celt/entenc.h b/code/opus-1.5.2/celt/entenc.h similarity index 90% rename from code/opus-1.2.1/celt/entenc.h rename to code/opus-1.5.2/celt/entenc.h index 796bc4d5..010874bb 100644 --- a/code/opus-1.2.1/celt/entenc.h +++ b/code/opus-1.5.2/celt/entenc.h @@ -64,10 +64,19 @@ void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp); _ftb: The number of bits of precision in the cumulative distribution.*/ void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb); +/*Encodes a symbol given an "inverse" CDF table. + _s: The index of the symbol to encode. + _icdf: The "inverse" CDF, such that symbol _s falls in the range + [_s>0?ft-_icdf[_s-1]:0,ft-_icdf[_s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution.*/ +void ec_enc_icdf16(ec_enc *_this,int _s,const opus_uint16 *_icdf,unsigned _ftb); + /*Encodes a raw unsigned integer in the stream. _fl: The integer to encode. _ft: The number of integers that can be encoded (one more than the max). - This must be at least one, and no more than 2**32-1.*/ + This must be at least 2, and no more than 2**32-1.*/ void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft); /*Encodes a sequence of raw bits in the stream. diff --git a/code/opus-1.2.1/celt/fixed_debug.h b/code/opus-1.5.2/celt/fixed_debug.h similarity index 91% rename from code/opus-1.2.1/celt/fixed_debug.h rename to code/opus-1.5.2/celt/fixed_debug.h index f4352952..ef2e5d02 100644 --- a/code/opus-1.2.1/celt/fixed_debug.h +++ b/code/opus-1.5.2/celt/fixed_debug.h @@ -167,7 +167,7 @@ static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line) #define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__) static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) { - int res; + opus_int32 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) { fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line); @@ -175,7 +175,7 @@ static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) celt_assert(0); #endif } - res = a<> 16; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT32_32_Q16: output is not int: %lld*%lld=%lld\n", (long long)a, (long long)b, (long long)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=5; + return res; +} + #define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __LINE__) static OPUS_INLINE int MULT16_16_(int a, int b, char *file, int line) { @@ -446,7 +491,7 @@ static OPUS_INLINE int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int celt_assert(0); #endif } - if (ABS32(b)>=((opus_val32)(1)<<(15+Q))) + if (ABS32(b)>=((opus_int64)(1)<<(16+Q))) { fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); #ifdef FIXED_DEBUG_ASSERT @@ -479,7 +524,7 @@ static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int celt_assert(0); #endif } - if (ABS32(b)>=((opus_int64)(1)<<(15+Q))) + if (ABS32(b)>=((opus_int64)(1)<<(16+Q))) { fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n\n", Q, (int)a, (int)b,file, line); #ifdef FIXED_DEBUG_ASSERT @@ -786,6 +831,6 @@ static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x) #undef PRINT_MIPS -#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", celt_mips);} while (0); +#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", (unsigned long long)celt_mips);} while (0); #endif diff --git a/code/opus-1.2.1/celt/fixed_generic.h b/code/opus-1.5.2/celt/fixed_generic.h similarity index 91% rename from code/opus-1.2.1/celt/fixed_generic.h rename to code/opus-1.5.2/celt/fixed_generic.h index 5f4abda7..8f29d46b 100644 --- a/code/opus-1.2.1/celt/fixed_generic.h +++ b/code/opus-1.5.2/celt/fixed_generic.h @@ -57,6 +57,13 @@ #define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15)) #endif +/** 32x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT32_32_Q16(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),16)) +#else +#define MULT32_32_Q16(a,b) (ADD32(ADD32(ADD32((opus_val32)(SHR32(((opus_uint32)((a)&0x0000ffff)*(opus_uint32)((b)&0x0000ffff)),16)), MULT16_16SU(SHR32(a,16),((b)&0x0000ffff))), MULT16_16SU(SHR32(b,16),((a)&0x0000ffff))), SHL32(MULT16_16(SHR32(a,16),SHR32(b,16)),16))) +#endif + /** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ #if OPUS_FAST_INT64 #define MULT32_32_Q31(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),31)) @@ -102,9 +109,9 @@ #define SATURATE16(x) (EXTRACT16((x)>32767 ? 32767 : (x)<-32768 ? -32768 : (x))) -/** Shift by a and round-to-neareast 32-bit value. Result is a 16-bit value */ +/** Shift by a and round-to-nearest 32-bit value. Result is a 16-bit value */ #define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a)))) -/** Shift by a and round-to-neareast 32-bit value. Result is a saturated 16-bit value */ +/** Shift by a and round-to-nearest 32-bit value. Result is a saturated 16-bit value */ #define SROUND16(x,a) EXTRACT16(SATURATE(PSHR32(x,a), 32767)); /** Divide by two */ @@ -131,6 +138,9 @@ /** 16x16 multiplication where the result fits in 16 bits */ #define MULT16_16_16(a,b) ((((opus_val16)(a))*((opus_val16)(b)))) +/** 32x32 multiplication where the result fits in 32 bits */ +#define MULT32_32_32(a,b) ((((opus_val32)(a))*((opus_val32)(b)))) + /* (opus_val32)(opus_val16) gives TI compiler a hint that it's 16x16->32 multiply */ /** 16x16 multiplication where the result fits in 32 bits */ #define MULT16_16(a,b) (((opus_val32)(opus_val16)(a))*((opus_val32)(opus_val16)(b))) diff --git a/code/opus-1.2.1/celt/float_cast.h b/code/opus-1.5.2/celt/float_cast.h similarity index 89% rename from code/opus-1.2.1/celt/float_cast.h rename to code/opus-1.5.2/celt/float_cast.h index 98b40abc..8915a5fd 100644 --- a/code/opus-1.2.1/celt/float_cast.h +++ b/code/opus-1.5.2/celt/float_cast.h @@ -67,7 +67,39 @@ #include static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_ss(x));} -#elif defined(HAVE_LRINTF) +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1)) + + #include + static OPUS_INLINE opus_int32 float2int(float value) + { + /* _mm_load_ss will generate same code as _mm_set_ss + ** in _MSC_VER >= 1914 /02 so keep __mm_load__ss + ** for backward compatibility. + */ + return _mm_cvtss_si32(_mm_load_ss(&value)); + } + +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86) + + #include + + /* Win32 doesn't seem to have these functions. + ** Therefore implement OPUS_INLINE versions of these functions here. + */ + + static OPUS_INLINE opus_int32 + float2int (float flt) + { int intgr; + + _asm + { fld flt + fistp intgr + } ; + + return intgr ; + } + +#elif defined(HAVE_LRINTF) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* These defines enable functionality introduced with the 1999 ISO C ** standard. They must be defined before the inclusion of math.h to @@ -85,7 +117,7 @@ static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_s #include #define float2int(x) lrintf(x) -#elif (defined(HAVE_LRINT)) +#elif defined(HAVE_LRINT) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L #define _ISOC9X_SOURCE 1 #define _ISOC99_SOURCE 1 @@ -96,32 +128,6 @@ static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_s #include #define float2int(x) lrint(x) -#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_X64) - #include - - __inline long int float2int(float value) - { - return _mm_cvtss_si32(_mm_load_ss(&value)); - } -#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86) - #include - - /* Win32 doesn't seem to have these functions. - ** Therefore implement OPUS_INLINE versions of these functions here. - */ - - __inline long int - float2int (float flt) - { int intgr; - - _asm - { fld flt - fistp intgr - } ; - - return intgr ; - } - #else #if (defined(__GNUC__) && defined(__STDC__) && __STDC__ && __STDC_VERSION__ >= 199901L) diff --git a/code/opus-1.2.1/celt/kiss_fft.c b/code/opus-1.5.2/celt/kiss_fft.c similarity index 100% rename from code/opus-1.2.1/celt/kiss_fft.c rename to code/opus-1.5.2/celt/kiss_fft.c diff --git a/code/opus-1.2.1/celt/kiss_fft.h b/code/opus-1.5.2/celt/kiss_fft.h similarity index 93% rename from code/opus-1.2.1/celt/kiss_fft.h rename to code/opus-1.5.2/celt/kiss_fft.h index bffa2bfa..267f72f9 100644 --- a/code/opus-1.2.1/celt/kiss_fft.h +++ b/code/opus-1.5.2/celt/kiss_fft.h @@ -52,6 +52,10 @@ extern "C" { # define kiss_fft_scalar opus_int32 # define kiss_twiddle_scalar opus_int16 +/* Some 32-bit CPUs would load/store a kiss_twiddle_cpx with a single memory + * access, and could benefit from additional alignment. + */ +# define KISS_TWIDDLE_CPX_ALIGNMENT (sizeof(opus_int32)) #else # ifndef kiss_fft_scalar @@ -62,6 +66,12 @@ extern "C" { # endif #endif +#if defined(__GNUC__) && defined(KISS_TWIDDLE_CPX_ALIGNMENT) +#define KISS_TWIDDLE_CPX_ALIGNED __attribute__((aligned(KISS_TWIDDLE_CPX_ALIGNMENT))) +#else +#define KISS_TWIDDLE_CPX_ALIGNED +#endif + typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; @@ -70,7 +80,7 @@ typedef struct { typedef struct { kiss_twiddle_scalar r; kiss_twiddle_scalar i; -}kiss_twiddle_cpx; +} KISS_TWIDDLE_CPX_ALIGNED kiss_twiddle_cpx; #define MAXFACTORS 8 /* e.g. an fft of length 128 has 4 factors diff --git a/code/opus-1.2.1/celt/laplace.c b/code/opus-1.5.2/celt/laplace.c similarity index 64% rename from code/opus-1.2.1/celt/laplace.c rename to code/opus-1.5.2/celt/laplace.c index a7bca874..21809666 100644 --- a/code/opus-1.2.1/celt/laplace.c +++ b/code/opus-1.5.2/celt/laplace.c @@ -132,3 +132,104 @@ int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay) ec_dec_update(dec, fl, IMIN(fl+fs,32768), 32768); return val; } + +void ec_laplace_encode_p0(ec_enc *enc, int value, opus_uint16 p0, opus_uint16 decay) +{ + int s; + opus_uint16 sign_icdf[3]; + sign_icdf[0] = 32768-p0; + sign_icdf[1] = sign_icdf[0]/2; + sign_icdf[2] = 0; + s = value == 0 ? 0 : (value > 0 ? 1 : 2); + ec_enc_icdf16(enc, s, sign_icdf, 15); + value = abs(value); + if (value) + { + int i; + opus_uint16 icdf[8]; + icdf[0] = IMAX(7, decay); + for (i=1;i<7;i++) + { + icdf[i] = IMAX(7-i, (icdf[i-1] * (opus_int32)decay) >> 15); + } + icdf[7] = 0; + value--; + do { + ec_enc_icdf16(enc, IMIN(value, 7), icdf, 15); + value -= 7; + } while (value >= 0); + } +} + +int ec_laplace_decode_p0(ec_dec *dec, opus_uint16 p0, opus_uint16 decay) +{ + int s; + int value; + opus_uint16 sign_icdf[3]; + sign_icdf[0] = 32768-p0; + sign_icdf[1] = sign_icdf[0]/2; + sign_icdf[2] = 0; + s = ec_dec_icdf16(dec, sign_icdf, 15); + if (s==2) s = -1; + if (s != 0) + { + int i; + int v; + opus_uint16 icdf[8]; + icdf[0] = IMAX(7, decay); + for (i=1;i<7;i++) + { + icdf[i] = IMAX(7-i, (icdf[i-1] * (opus_int32)decay) >> 15); + } + icdf[7] = 0; + value = 1; + do { + v = ec_dec_icdf16(dec, icdf, 15); + value += v; + } while (v == 7); + return s*value; + } else return 0; +} + +#if 0 + +#include +#define NB_VALS 10 +#define DATA_SIZE 10000 +int main() { + ec_enc enc; + ec_dec dec; + unsigned char *ptr; + int i; + int decay, p0; + int val[NB_VALS] = {6, 7, 8, 9, 10, 11, 12, 13, 14, 15}; + /*for (i=0;i0, "celt_rcp() only defined for positive values"); + celt_sig_assert(x>0); i = celt_ilog2(x); /* n is Q15 with range [0,1). */ n = VSHR32(x,i-15)-32768; diff --git a/code/opus-1.2.1/celt/mathops.h b/code/opus-1.5.2/celt/mathops.h similarity index 97% rename from code/opus-1.2.1/celt/mathops.h rename to code/opus-1.5.2/celt/mathops.h index 1f8a20cb..e2eece29 100644 --- a/code/opus-1.2.1/celt/mathops.h +++ b/code/opus-1.5.2/celt/mathops.h @@ -72,7 +72,7 @@ static OPUS_INLINE float fast_atan2f(float y, float x) { #undef cA #undef cB #undef cC -#undef cD +#undef cE #endif @@ -137,7 +137,7 @@ static OPUS_INLINE float celt_log2(float x) } in; in.f = x; integer = (in.i>>23)-127; - in.i -= integer<<23; + in.i -= (opus_uint32)integer<<23; frac = in.f - 1.5f; frac = -0.41445418f + frac*(0.95909232f + frac*(-0.33951290f + frac*0.16541097f)); @@ -153,14 +153,14 @@ static OPUS_INLINE float celt_exp2(float x) float f; opus_uint32 i; } res; - integer = floor(x); + integer = (int)floor(x); if (integer < -50) return 0; frac = x-integer; /* K0 = 1, K1 = log(2), K2 = 3-4*log(2), K3 = 3*log(2) - 2 */ res.f = 0.99992522f + frac * (0.69583354f + frac * (0.22606716f + 0.078024523f*frac)); - res.i = (res.i + (integer<<23)) & 0x7fffffff; + res.i = (res.i + ((opus_uint32)integer<<23)) & 0x7fffffff; return res.f; } @@ -179,7 +179,7 @@ static OPUS_INLINE float celt_exp2(float x) /** Integer log in base2. Undefined for zero and negative numbers */ static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x) { - celt_assert2(x>0, "celt_ilog2() only defined for strictly positive numbers"); + celt_sig_assert(x>0); return EC_ILOG(x)-1; } #endif @@ -230,6 +230,12 @@ static OPUS_INLINE opus_val32 celt_exp2_frac(opus_val16 x) frac = SHL16(x, 4); return ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); } + +#undef D0 +#undef D1 +#undef D2 +#undef D3 + /** Base-2 exponential approximation (2^x). (Q10 input, Q16 output) */ static OPUS_INLINE opus_val32 celt_exp2(opus_val16 x) { diff --git a/code/opus-1.2.1/celt/mdct.c b/code/opus-1.5.2/celt/mdct.c similarity index 100% rename from code/opus-1.2.1/celt/mdct.c rename to code/opus-1.5.2/celt/mdct.c diff --git a/code/opus-1.2.1/celt/mdct.h b/code/opus-1.5.2/celt/mdct.h similarity index 100% rename from code/opus-1.2.1/celt/mdct.h rename to code/opus-1.5.2/celt/mdct.h diff --git a/code/opus-1.2.1/celt/mfrngcod.h b/code/opus-1.5.2/celt/mfrngcod.h similarity index 100% rename from code/opus-1.2.1/celt/mfrngcod.h rename to code/opus-1.5.2/celt/mfrngcod.h diff --git a/code/opus-1.2.1/celt/mips/celt_mipsr1.h b/code/opus-1.5.2/celt/mips/celt_mipsr1.h similarity index 97% rename from code/opus-1.2.1/celt/mips/celt_mipsr1.h rename to code/opus-1.5.2/celt/mips/celt_mipsr1.h index e85661a6..d1b25c20 100644 --- a/code/opus-1.2.1/celt/mips/celt_mipsr1.h +++ b/code/opus-1.5.2/celt/mips/celt_mipsr1.h @@ -27,8 +27,8 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef __CELT_MIPSR1_H__ -#define __CELT_MIPSR1_H__ +#ifndef CELT_MIPSR1_H__ +#define CELT_MIPSR1_H__ #ifdef HAVE_CONFIG_H #include "config.h" @@ -53,6 +53,7 @@ #include "celt_lpc.h" #include "vq.h" +#define OVERRIDE_COMB_FILTER_CONST #define OVERRIDE_comb_filter void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, @@ -148,4 +149,4 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, } } -#endif /* __CELT_MIPSR1_H__ */ +#endif /* CELT_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/celt/mips/fixed_generic_mipsr1.h b/code/opus-1.5.2/celt/mips/fixed_generic_mipsr1.h similarity index 100% rename from code/opus-1.2.1/celt/mips/fixed_generic_mipsr1.h rename to code/opus-1.5.2/celt/mips/fixed_generic_mipsr1.h diff --git a/code/opus-1.2.1/celt/mips/kiss_fft_mipsr1.h b/code/opus-1.5.2/celt/mips/kiss_fft_mipsr1.h similarity index 100% rename from code/opus-1.2.1/celt/mips/kiss_fft_mipsr1.h rename to code/opus-1.5.2/celt/mips/kiss_fft_mipsr1.h diff --git a/code/opus-1.2.1/celt/mips/mdct_mipsr1.h b/code/opus-1.5.2/celt/mips/mdct_mipsr1.h similarity index 99% rename from code/opus-1.2.1/celt/mips/mdct_mipsr1.h rename to code/opus-1.5.2/celt/mips/mdct_mipsr1.h index 2934dab7..7456c181 100644 --- a/code/opus-1.2.1/celt/mips/mdct_mipsr1.h +++ b/code/opus-1.5.2/celt/mips/mdct_mipsr1.h @@ -38,8 +38,8 @@ MDCT implementation in FFMPEG, but has differences in signs, ordering and scaling in many places. */ -#ifndef __MDCT_MIPSR1_H__ -#define __MDCT_MIPSR1_H__ +#ifndef MDCT_MIPSR1_H__ +#define MDCT_MIPSR1_H__ #ifndef SKIP_CONFIG_H #ifdef HAVE_CONFIG_H @@ -285,4 +285,4 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala } } } -#endif /* __MDCT_MIPSR1_H__ */ +#endif /* MDCT_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/celt/mips/pitch_mipsr1.h b/code/opus-1.5.2/celt/mips/pitch_mipsr1.h similarity index 100% rename from code/opus-1.2.1/celt/mips/pitch_mipsr1.h rename to code/opus-1.5.2/celt/mips/pitch_mipsr1.h diff --git a/code/opus-1.2.1/celt/mips/vq_mipsr1.h b/code/opus-1.5.2/celt/mips/vq_mipsr1.h similarity index 90% rename from code/opus-1.2.1/celt/mips/vq_mipsr1.h rename to code/opus-1.5.2/celt/mips/vq_mipsr1.h index fd18eab7..1621c562 100644 --- a/code/opus-1.2.1/celt/mips/vq_mipsr1.h +++ b/code/opus-1.5.2/celt/mips/vq_mipsr1.h @@ -26,8 +26,8 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef __VQ_MIPSR1_H__ -#define __VQ_MIPSR1_H__ +#ifndef VQ_MIPSR1_H__ +#define VQ_MIPSR1_H__ #ifdef HAVE_CONFIG_H #include "config.h" @@ -36,8 +36,6 @@ #include "mathops.h" #include "arch.h" -static void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch); - #define OVERRIDE_vq_exp_rotation1 static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) { @@ -66,11 +64,7 @@ static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_ } #define OVERRIDE_renormalise_vector - -#define renormalise_vector(X, N, gain, arch) \ - (renormalise_vector_mips(X, N, gain, arch)) - -void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch) +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch) { int i; #ifdef FIXED_POINT @@ -119,4 +113,4 @@ void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch) /*return celt_sqrt(E);*/ } -#endif /* __VQ_MIPSR1_H__ */ +#endif /* VQ_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/celt/modes.c b/code/opus-1.5.2/celt/modes.c similarity index 99% rename from code/opus-1.2.1/celt/modes.c rename to code/opus-1.5.2/celt/modes.c index 390c5e8a..23f7cde6 100644 --- a/code/opus-1.2.1/celt/modes.c +++ b/code/opus-1.5.2/celt/modes.c @@ -173,7 +173,10 @@ static void compute_allocation_table(CELTMode *mode) mode->nbAllocVectors = BITALLOC_SIZE; allocVectors = opus_alloc(sizeof(unsigned char)*(BITALLOC_SIZE*mode->nbEBands)); if (allocVectors==NULL) + { + mode->allocVectors = NULL; return; + } /* Check for standard mode */ if (mode->Fs == 400*(opus_int32)mode->shortMdctSize) diff --git a/code/opus-1.2.1/celt/modes.h b/code/opus-1.5.2/celt/modes.h similarity index 100% rename from code/opus-1.2.1/celt/modes.h rename to code/opus-1.5.2/celt/modes.h diff --git a/code/opus-1.2.1/celt/opus_custom_demo.c b/code/opus-1.5.2/celt/opus_custom_demo.c similarity index 100% rename from code/opus-1.2.1/celt/opus_custom_demo.c rename to code/opus-1.5.2/celt/opus_custom_demo.c diff --git a/code/opus-1.2.1/celt/os_support.h b/code/opus-1.5.2/celt/os_support.h similarity index 85% rename from code/opus-1.2.1/celt/os_support.h rename to code/opus-1.5.2/celt/os_support.h index a2171971..7d2d3781 100644 --- a/code/opus-1.2.1/celt/os_support.h +++ b/code/opus-1.5.2/celt/os_support.h @@ -39,10 +39,9 @@ #include "opus_defines.h" #include -#include #include -/** Opus wrapper for malloc(). To do your own dynamic allocation, all you need to do is replace this function and opus_free */ +/** Opus wrapper for malloc(). To do your own dynamic allocation replace this function, opus_realloc, and opus_free */ #ifndef OVERRIDE_OPUS_ALLOC static OPUS_INLINE void *opus_alloc (size_t size) { @@ -50,7 +49,15 @@ static OPUS_INLINE void *opus_alloc (size_t size) } #endif -/** Same as celt_alloc(), except that the area is only needed inside a CELT call (might cause problem with wideband though) */ +#ifndef OVERRIDE_OPUS_REALLOC +static OPUS_INLINE void *opus_realloc (void *ptr, size_t size) +{ + return realloc(ptr, size); +} +#endif + +/** Used only for non-threadsafe pseudostack. + If desired, this can always return the same area of memory rather than allocating a new one every time. */ #ifndef OVERRIDE_OPUS_ALLOC_SCRATCH static OPUS_INLINE void *opus_alloc_scratch (size_t size) { @@ -59,7 +66,7 @@ static OPUS_INLINE void *opus_alloc_scratch (size_t size) } #endif -/** Opus wrapper for free(). To do your own dynamic allocation, all you need to do is replace this function and opus_alloc */ +/** Opus wrapper for free(). To do your own dynamic allocation replace this function, opus_realloc, and opus_free */ #ifndef OVERRIDE_OPUS_FREE static OPUS_INLINE void opus_free (void *ptr) { diff --git a/code/opus-1.2.1/celt/pitch.c b/code/opus-1.5.2/celt/pitch.c similarity index 91% rename from code/opus-1.2.1/celt/pitch.c rename to code/opus-1.5.2/celt/pitch.c index 5b061d86..e33c60a3 100644 --- a/code/opus-1.2.1/celt/pitch.c +++ b/code/opus-1.5.2/celt/pitch.c @@ -102,11 +102,9 @@ static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len, } } -static void celt_fir5(const opus_val16 *x, +static void celt_fir5(opus_val16 *x, const opus_val16 *num, - opus_val16 *y, - int N, - opus_val16 *mem) + int N) { int i; opus_val16 num0, num1, num2, num3, num4; @@ -116,11 +114,11 @@ static void celt_fir5(const opus_val16 *x, num2=num[2]; num3=num[3]; num4=num[4]; - mem0=mem[0]; - mem1=mem[1]; - mem2=mem[2]; - mem3=mem[3]; - mem4=mem[4]; + mem0=0; + mem1=0; + mem2=0; + mem3=0; + mem4=0; for (i=0;i>1;i++) - x_lp[i] = SHR32(HALF32(HALF32(x[0][(2*i-1)]+x[0][(2*i+1)])+x[0][2*i]), shift); - x_lp[0] = SHR32(HALF32(HALF32(x[0][1])+x[0][0]), shift); + x_lp[i] = SHR32(x[0][(2*i-1)], shift+2) + SHR32(x[0][(2*i+1)], shift+2) + SHR32(x[0][2*i], shift+1); + x_lp[0] = SHR32(x[0][1], shift+2) + SHR32(x[0][0], shift+1); if (C==2) { for (i=1;i>1;i++) - x_lp[i] += SHR32(HALF32(HALF32(x[1][(2*i-1)]+x[1][(2*i+1)])+x[1][2*i]), shift); - x_lp[0] += SHR32(HALF32(HALF32(x[1][1])+x[1][0]), shift); + x_lp[i] += SHR32(x[1][(2*i-1)], shift+2) + SHR32(x[1][(2*i+1)], shift+2) + SHR32(x[1][2*i], shift+1); + x_lp[0] += SHR32(x[1][1], shift+2) + SHR32(x[1][0], shift+1); } - +#else + for (i=1;i>1;i++) + x_lp[i] = .25f*x[0][(2*i-1)] + .25f*x[0][(2*i+1)] + .5f*x[0][2*i]; + x_lp[0] = .25f*x[0][1] + .5f*x[0][0]; + if (C==2) + { + for (i=1;i>1;i++) + x_lp[i] += .25f*x[1][(2*i-1)] + .25f*x[1][(2*i+1)] + .5f*x[1][2*i]; + x_lp[0] += .25f*x[1][1] + .5f*x[1][0]; + } +#endif _celt_autocorr(x_lp, ac, NULL, 0, 4, len>>1, arch); @@ -211,7 +213,7 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x lpc2[2] = lpc[2] + MULT16_16_Q15(c1,lpc[1]); lpc2[3] = lpc[3] + MULT16_16_Q15(c1,lpc[2]); lpc2[4] = MULT16_16_Q15(c1,lpc[3]); - celt_fir5(x_lp, lpc2, x_lp, len>>1, mem); + celt_fir5(x_lp, lpc2, len>>1); } /* Pure C implementation. */ @@ -256,11 +258,20 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 maxcorr=1; #endif celt_assert(max_pitch>0); - celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); + celt_sig_assert(((size_t)_x&3)==0); for (i=0;i=15) { diff --git a/code/opus-1.2.1/celt/quant_bands.h b/code/opus-1.5.2/celt/quant_bands.h similarity index 100% rename from code/opus-1.2.1/celt/quant_bands.h rename to code/opus-1.5.2/celt/quant_bands.h diff --git a/code/opus-1.2.1/celt/rate.c b/code/opus-1.5.2/celt/rate.c similarity index 98% rename from code/opus-1.2.1/celt/rate.c rename to code/opus-1.5.2/celt/rate.c index ca4cc870..7f7ad3fa 100644 --- a/code/opus-1.2.1/celt/rate.c +++ b/code/opus-1.5.2/celt/rate.c @@ -356,6 +356,8 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, else depth_threshold = 0; #ifdef FUZZING + (void)signalBandwidth; + (void)depth_threshold; if ((rand()&0x1) == 0) #else if (codedBands<=start+2 || (band_bits > (depth_threshold*band_width<>4 && j<=signalBandwidth)) @@ -529,7 +531,7 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, return codedBands; } -int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) { int lo, hi, len, j; diff --git a/code/opus-1.2.1/celt/rate.h b/code/opus-1.5.2/celt/rate.h similarity index 95% rename from code/opus-1.2.1/celt/rate.h rename to code/opus-1.5.2/celt/rate.h index 515f7687..fad5e412 100644 --- a/code/opus-1.2.1/celt/rate.h +++ b/code/opus-1.5.2/celt/rate.h @@ -95,7 +95,7 @@ static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int puls @param pulses Number of pulses per band (returned) @return Total number of bits allocated */ -int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stero, +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth); #endif diff --git a/code/opus-1.2.1/celt/stack_alloc.h b/code/opus-1.5.2/celt/stack_alloc.h similarity index 96% rename from code/opus-1.2.1/celt/stack_alloc.h rename to code/opus-1.5.2/celt/stack_alloc.h index 2b51c8d8..e2739bdf 100644 --- a/code/opus-1.2.1/celt/stack_alloc.h +++ b/code/opus-1.5.2/celt/stack_alloc.h @@ -40,7 +40,7 @@ #endif #ifdef USE_ALLOCA -# ifdef WIN32 +# ifdef _WIN32 # include # else # ifdef HAVE_ALLOCA_H @@ -102,7 +102,7 @@ #define VARDECL(type, var) type *var -# ifdef WIN32 +# ifdef _WIN32 # define ALLOC(var, size, type) var = ((type*)_alloca(sizeof(type)*(size))) # else # define ALLOC(var, size, type) var = ((type*)alloca(sizeof(type)*(size))) @@ -141,7 +141,7 @@ extern char *global_stack_top; #else #define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) -#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/sizeof(char)),(stack)+=(size)*(sizeof(type)/sizeof(char)),(type*)((stack)-(size)*(sizeof(type)/sizeof(char)))) +#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/(sizeof(char))),(stack)+=(size)*(sizeof(type)/(sizeof(char))),(type*)((stack)-(size)*(sizeof(type)/(sizeof(char))))) #if 0 /* Set this to 1 to instrument pseudostack usage */ #define RESTORE_STACK (printf("%ld %s:%d\n", global_stack-scratch_ptr, __FILE__, __LINE__),global_stack = _saved_stack) #else diff --git a/code/opus-1.2.1/celt/static_modes_fixed.h b/code/opus-1.5.2/celt/static_modes_fixed.h similarity index 100% rename from code/opus-1.2.1/celt/static_modes_fixed.h rename to code/opus-1.5.2/celt/static_modes_fixed.h diff --git a/code/opus-1.2.1/celt/static_modes_fixed_arm_ne10.h b/code/opus-1.5.2/celt/static_modes_fixed_arm_ne10.h similarity index 100% rename from code/opus-1.2.1/celt/static_modes_fixed_arm_ne10.h rename to code/opus-1.5.2/celt/static_modes_fixed_arm_ne10.h diff --git a/code/opus-1.2.1/celt/static_modes_float.h b/code/opus-1.5.2/celt/static_modes_float.h similarity index 100% rename from code/opus-1.2.1/celt/static_modes_float.h rename to code/opus-1.5.2/celt/static_modes_float.h diff --git a/code/opus-1.2.1/celt/static_modes_float_arm_ne10.h b/code/opus-1.5.2/celt/static_modes_float_arm_ne10.h similarity index 100% rename from code/opus-1.2.1/celt/static_modes_float_arm_ne10.h rename to code/opus-1.5.2/celt/static_modes_float_arm_ne10.h diff --git a/code/opus-1.5.2/celt/tests/test_unit_cwrs32.c b/code/opus-1.5.2/celt/tests/test_unit_cwrs32.c new file mode 100644 index 00000000..f6b8ac4b --- /dev/null +++ b/code/opus-1.5.2/celt/tests/test_unit_cwrs32.c @@ -0,0 +1,162 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#else +#define TEST_CUSTOM_MODES +#endif + +#define CELT_C +#include "stack_alloc.h" +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "cwrs.c" +#include "mathops.c" +#include "rate.h" + +#define NMAX (240) +#define KMAX (128) + +#ifdef TEST_CUSTOM_MODES + +#define NDIMS (44) +static const int pn[NDIMS]={ + 2, 3, 4, 5, 6, 7, 8, 9, 10, + 11, 12, 13, 14, 15, 16, 18, 20, 22, + 24, 26, 28, 30, 32, 36, 40, 44, 48, + 52, 56, 60, 64, 72, 80, 88, 96, 104, + 112, 120, 128, 144, 160, 176, 192, 208 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 128, 88, 52, 36, 26, 22, + 18, 16, 15, 13, 12, 12, 11, 10, 9, + 9, 8, 8, 7, 7, 7, 7, 6, 6, + 6, 6, 6, 5, 5, 5, 5, 5, 5, + 4, 4, 4, 4, 4, 4, 4, 4 +}; + +#else /* TEST_CUSTOM_MODES */ + +#define NDIMS (22) +static const int pn[NDIMS]={ + 2, 3, 4, 6, 8, 9, 11, 12, 16, + 18, 22, 24, 32, 36, 44, 48, 64, 72, + 88, 96, 144, 176 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 88, 36, 26, 18, 16, 12, + 11, 9, 9, 7, 7, 6, 6, 5, 5, + 5, 5, 4, 4 +}; + +#endif + +int main(void){ + int t; + int n; + ALLOC_STACK; + for(t=0;tpkmax[t])break; + printf("Testing CWRS with N=%i, K=%i...\n",n,k); +#if defined(SMALL_FOOTPRINT) + nc=ncwrs_urow(n,k,uu); +#else + nc=CELT_PVQ_V(n,k); +#endif + inc=nc/20000; + if(inc<1)inc=1; + for(i=0;i");*/ +#if defined(SMALL_FOOTPRINT) + ii=icwrs(n,k,&v,y,u); +#else + ii=icwrs(n,y); + v=CELT_PVQ_V(n,k); +#endif + if(ii!=i){ + fprintf(stderr,"Combination-index mismatch (%lu!=%lu).\n", + (long)ii,(long)i); + return 1; + } + if(v!=nc){ + fprintf(stderr,"Combination count mismatch (%lu!=%lu).\n", + (long)v,(long)nc); + return 2; + } + /*printf(" %6u\n",i);*/ + } + /*printf("\n");*/ + } + } + RESTORE_STACK; + return 0; +} diff --git a/code/opus-1.5.2/celt/tests/test_unit_dft.c b/code/opus-1.5.2/celt/tests/test_unit_dft.c new file mode 100644 index 00000000..ad6c60a0 --- /dev/null +++ b/code/opus-1.5.2/celt/tests/test_unit_dft.c @@ -0,0 +1,181 @@ +/* Copyright (c) 2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include + +#include "stack_alloc.h" +#include "kiss_fft.h" +#include "mathops.h" +#include "modes.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; + +void check(kiss_fft_cpx * in,kiss_fft_cpx * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0, snr; + + for (bin=0;binmdct.kfft[id]; +#endif + + in = (kiss_fft_cpx*)malloc(buflen); + out = (kiss_fft_cpx*)malloc(buflen); + + for (k=0;k1) { + int k; + for (k=1;k +#include +#include +#include +#define CELT_C +#include "entcode.h" +#include "entenc.h" +#include "entdec.h" +#include + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" + +#ifndef M_LOG2E +# define M_LOG2E 1.4426950408889634074 +#endif +#define DATA_SIZE 10000000 +#define DATA_SIZE2 10000 + +int main(int _argc,char **_argv){ + ec_enc enc; + ec_dec dec; + long nbits; + long nbits2; + double entropy; + int ft; + int ftb; + int sz; + int i; + int ret; + unsigned int sym; + unsigned int seed; + unsigned char *ptr; + const char *env_seed; + ret=0; + entropy=0; + if (_argc > 2) { + fprintf(stderr, "Usage: %s []\n", _argv[0]); + return 1; + } + env_seed = getenv("SEED"); + if (_argc > 1) + seed = atoi(_argv[1]); + else if (env_seed) + seed = atoi(env_seed); + else + seed = time(NULL); + /*Testing encoding of raw bit values.*/ + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr, DATA_SIZE); + for(ft=2;ft<1024;ft++){ + for(i=0;i>(rand()%11U))+1U)+10; + sz=rand()/((RAND_MAX>>(rand()%9U))+1U); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + zeros = rand()%13==0; + tell[0]=ec_tell_frac(&enc); + for(j=0;j>(rand()%9U))+1U); + logp1=(unsigned *)malloc(sz*sizeof(*logp1)); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + enc_method=(unsigned *)malloc(sz*sizeof(*enc_method)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + tell[0]=ec_tell_frac(&enc); + for(j=0;j>1)+1); + logp1[j]=(rand()%15)+1; + enc_method[j]=rand()/((RAND_MAX>>2)+1); + switch(enc_method[j]){ + case 0:{ + ec_encode(&enc,data[j]?(1<>2)+1); + switch(dec_method){ + case 0:{ + fs=ec_decode(&dec,1<=(1<=(1< +#include +#define CELT_C +#include "laplace.h" +#include "stack_alloc.h" + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "laplace.c" + +#define DATA_SIZE 40000 + +int ec_laplace_get_start_freq(int decay) +{ + opus_uint32 ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN+1); + int fs = (ft*(16384-decay))/(16384+decay); + return fs+LAPLACE_MINP; +} + +int main(void) +{ + int i; + int ret = 0; + ec_enc enc; + ec_dec dec; + unsigned char *ptr; + int val[10000], decay[10000]; + ALLOC_STACK; + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr,DATA_SIZE); + + val[0] = 3; decay[0] = 6000; + val[1] = 0; decay[1] = 5800; + val[2] = -1; decay[2] = 5600; + for (i=3;i<10000;i++) + { + val[i] = rand()%15-7; + decay[i] = rand()%11000+5000; + } + for (i=0;i<10000;i++) + ec_laplace_encode(&enc, &val[i], + ec_laplace_get_start_freq(decay[i]), decay[i]); + + ec_enc_done(&enc); + + ec_dec_init(&dec,ec_get_buffer(&enc),ec_range_bytes(&enc)); + + for (i=0;i<10000;i++) + { + int d = ec_laplace_decode(&dec, + ec_laplace_get_start_freq(decay[i]), decay[i]); + if (d != val[i]) + { + fprintf (stderr, "Got %d instead of %d\n", d, val[i]); + ret = 1; + } + } + + free(ptr); + RESTORE_STACK; + return ret; +} diff --git a/code/opus-1.5.2/celt/tests/test_unit_mathops.c b/code/opus-1.5.2/celt/tests/test_unit_mathops.c new file mode 100644 index 00000000..0615448d --- /dev/null +++ b/code/opus-1.5.2/celt/tests/test_unit_mathops.c @@ -0,0 +1,266 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#include +#include +#include "mathops.h" +#include "bands.h" + +#ifdef FIXED_POINT +#define WORD "%d" +#else +#define WORD "%f" +#endif + +int ret = 0; + +void testdiv(void) +{ + opus_int32 i; + for (i=1;i<=327670;i++) + { + double prod; + opus_val32 val; + val = celt_rcp(i); +#ifdef FIXED_POINT + prod = (1./32768./65526.)*val*i; +#else + prod = val*i; +#endif + if (fabs(prod-1) > .00025) + { + fprintf (stderr, "div failed: 1/%d="WORD" (product = %f)\n", i, val, prod); + ret = 1; + } + } +} + +void testsqrt(void) +{ + opus_int32 i; + for (i=1;i<=1000000000;i++) + { + double ratio; + opus_val16 val; + val = celt_sqrt(i); + ratio = val/sqrt(i); + if (fabs(ratio - 1) > .0005 && fabs(val-sqrt(i)) > 2) + { + fprintf (stderr, "sqrt failed: sqrt(%d)="WORD" (ratio = %f)\n", i, val, ratio); + ret = 1; + } + i+= i>>10; + } +} + +void testbitexactcos(void) +{ + int i; + opus_int32 min_d,max_d,last,chk; + chk=max_d=0; + last=min_d=32767; + for(i=64;i<=16320;i++) + { + opus_int32 d; + opus_int32 q=bitexact_cos(i); + chk ^= q*i; + d = last - q; + if (d>max_d)max_d=d; + if (dmax_d)max_d=d; + if (d0.0009) + { + fprintf (stderr, "celt_log2 failed: fabs((1.442695040888963387*log(x))-celt_log2(x))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007f) + { + float error = fabs(x-(1.442695040888963387*log(celt_exp2(x)))); + if (error>0.0002) + { + fprintf (stderr, "celt_exp2 failed: fabs(x-(1.442695040888963387*log(celt_exp2(x))))>0.0005 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007f) + { + float error = fabs(x-(celt_log2(celt_exp2(x)))); + if (error>0.001) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_log2(celt_exp2(x))))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} +#else +void testlog2(void) +{ + opus_val32 x; + for (x=8;x<1073741824;x+=(x>>3)) + { + float error = fabs((1.442695040888963387*log(x/16384.0))-celt_log2(x)/1024.0); + if (error>0.003) + { + fprintf (stderr, "celt_log2 failed: x = %ld, error = %f\n", (long)x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + opus_val16 x; + for (x=-32768;x<15360;x++) + { + float error1 = fabs(x/1024.0-(1.442695040888963387*log(celt_exp2(x)/65536.0))); + float error2 = fabs(exp(0.6931471805599453094*x/1024.0)-celt_exp2(x)/65536.0); + if (error1>0.0002&&error2>0.00004) + { + fprintf (stderr, "celt_exp2 failed: x = "WORD", error1 = %f, error2 = %f\n", x,error1,error2); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + opus_val32 x; + for (x=8;x<65536;x+=(x>>3)) + { + float error = fabs(x-0.25*celt_exp2(celt_log2(x)))/16384; + if (error>0.004) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_exp2(celt_log2(x))))>0.001 (x = %ld, error = %f)\n", (long)x,error); + ret = 1; + } + } +} + +void testilog2(void) +{ + opus_val32 x; + for (x=1;x<=268435455;x+=127) + { + opus_val32 lg; + opus_val32 y; + + lg = celt_ilog2(x); + if (lg<0 || lg>=31) + { + printf("celt_ilog2 failed: 0<=celt_ilog2(x)<31 (x = %d, celt_ilog2(x) = %d)\n",x,lg); + ret = 1; + } + y = 1<>1)>=y) + { + printf("celt_ilog2 failed: 2**celt_ilog2(x)<=x<2**(celt_ilog2(x)+1) (x = %d, 2**celt_ilog2(x) = %d)\n",x,y); + ret = 1; + } + } +} +#endif + +int main(void) +{ + testbitexactcos(); + testbitexactlog2tan(); + testdiv(); + testsqrt(); + testlog2(); + testexp2(); + testexp2log2(); +#ifdef FIXED_POINT + testilog2(); +#endif + return ret; +} diff --git a/code/opus-1.5.2/celt/tests/test_unit_mdct.c b/code/opus-1.5.2/celt/tests/test_unit_mdct.c new file mode 100644 index 00000000..70dc042e --- /dev/null +++ b/code/opus-1.5.2/celt/tests/test_unit_mdct.c @@ -0,0 +1,229 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include + +#include "mdct.h" +#include "stack_alloc.h" +#include "kiss_fft.h" +#include "mdct.h" +#include "modes.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; +void check(kiss_fft_scalar * in,kiss_fft_scalar * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0; + double snr; + for (bin=0;binmdct; +#endif + + in = (kiss_fft_scalar*)malloc(buflen); + in_copy = (kiss_fft_scalar*)malloc(buflen); + out = (kiss_fft_scalar*)malloc(buflen); + window = (opus_val16*)malloc(sizeof(opus_val16)*nfft/2); + + for (k=0;k1) { + int k; + for (k=1;k +#include +#include "vq.h" +#include "bands.h" +#include "stack_alloc.h" +#include + + +#define MAX_SIZE 100 + +int ret=0; +void test_rotation(int N, int K) +{ + int i; + double err = 0, ener = 0, snr, snr0; + opus_val16 x0[MAX_SIZE]; + opus_val16 x1[MAX_SIZE]; + for (i=0;i 20) + { + fprintf(stderr, "FAIL!\n"); + ret = 1; + } +} + +int main(void) +{ + ALLOC_STACK; + test_rotation(15, 3); + test_rotation(23, 5); + test_rotation(50, 3); + test_rotation(80, 1); + RESTORE_STACK; + return ret; +} diff --git a/code/opus-1.2.1/celt/arm/armopts.s.in b/code/opus-1.5.2/celt/tests/test_unit_types.c similarity index 74% rename from code/opus-1.2.1/celt/arm/armopts.s.in rename to code/opus-1.5.2/celt/tests/test_unit_types.c index 3d8aaf27..67a0fb8e 100644 --- a/code/opus-1.2.1/celt/arm/armopts.s.in +++ b/code/opus-1.5.2/celt/tests/test_unit_types.c @@ -1,4 +1,5 @@ -/* Copyright (C) 2013 Mozilla Corporation */ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ /* Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -24,14 +25,26 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -; Set the following to 1 if we have EDSP instructions -; (LDRD/STRD, etc., ARMv5E and later). -OPUS_ARM_MAY_HAVE_EDSP * @OPUS_ARM_MAY_HAVE_EDSP@ +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif -; Set the following to 1 if we have ARMv6 media instructions. -OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@ +#include "opus_types.h" +#include -; Set the following to 1 if we have NEON (some ARMv7) -OPUS_ARM_MAY_HAVE_NEON * @OPUS_ARM_MAY_HAVE_NEON@ - -END +int main(void) +{ + opus_int16 i = 1; + i <<= 14; + if (i>>14 != 1) + { + fprintf(stderr, "opus_int16 isn't 16 bits\n"); + return 1; + } + if (sizeof(opus_int16)*2 != sizeof(opus_int32)) + { + fprintf(stderr, "16*2 != 32\n"); + return 1; + } + return 0; +} diff --git a/code/opus-1.2.1/celt/vq.c b/code/opus-1.5.2/celt/vq.c similarity index 98% rename from code/opus-1.2.1/celt/vq.c rename to code/opus-1.5.2/celt/vq.c index 8ef80e50..8011e225 100644 --- a/code/opus-1.2.1/celt/vq.c +++ b/code/opus-1.5.2/celt/vq.c @@ -39,6 +39,10 @@ #include "rate.h" #include "pitch.h" +#if defined(MIPSr1_ASM) +#include "mips/vq_mipsr1.h" +#endif + #ifndef OVERRIDE_vq_exp_rotation1 static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) { @@ -230,12 +234,12 @@ opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch) pulsesLeft -= iy[j]; } while (++j=0, "Allocated too many pulses in the quick pass"); + celt_sig_assert(pulsesLeft>=0); /* This should never happen, but just in case it does (e.g. on silence) we fill the first bin with pulses. */ #ifdef FIXED_POINT_DEBUG - celt_assert2(pulsesLeft<=N+3, "Not enough pulses in the quick pass"); + celt_sig_assert(pulsesLeft<=N+3); #endif if (pulsesLeft > N+3) { diff --git a/code/opus-1.2.1/celt/vq.h b/code/opus-1.5.2/celt/vq.h similarity index 98% rename from code/opus-1.2.1/celt/vq.h rename to code/opus-1.5.2/celt/vq.h index 0dfe6af0..45ec5591 100644 --- a/code/opus-1.2.1/celt/vq.h +++ b/code/opus-1.5.2/celt/vq.h @@ -41,10 +41,6 @@ #include "x86/vq_sse.h" #endif -#if defined(MIPSr1_ASM) -#include "mips/vq_mipsr1.h" -#endif - void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread); opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch); diff --git a/code/opus-1.2.1/celt/x86/celt_lpc_sse.h b/code/opus-1.5.2/celt/x86/celt_lpc_sse.h similarity index 97% rename from code/opus-1.2.1/celt/x86/celt_lpc_sse.h rename to code/opus-1.5.2/celt/x86/celt_lpc_sse.h index 7d1ecf75..90e69ecf 100644 --- a/code/opus-1.2.1/celt/x86/celt_lpc_sse.h +++ b/code/opus-1.5.2/celt/x86/celt_lpc_sse.h @@ -33,7 +33,6 @@ #endif #if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) -#define OVERRIDE_CELT_FIR void celt_fir_sse4_1( const opus_val16 *x, @@ -44,10 +43,11 @@ void celt_fir_sse4_1( int arch); #if defined(OPUS_X86_PRESUME_SSE4_1) +#define OVERRIDE_CELT_FIR #define celt_fir(x, num, y, N, ord, arch) \ ((void)arch, celt_fir_sse4_1(x, num, y, N, ord, arch)) -#else +#elif defined(OPUS_HAVE_RTCD) extern void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( const opus_val16 *x, @@ -57,6 +57,7 @@ extern void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( int ord, int arch); +#define OVERRIDE_CELT_FIR # define celt_fir(x, num, y, N, ord, arch) \ ((*CELT_FIR_IMPL[(arch) & OPUS_ARCHMASK])(x, num, y, N, ord, arch)) diff --git a/code/opus-1.2.1/celt/x86/celt_lpc_sse.c b/code/opus-1.5.2/celt/x86/celt_lpc_sse4_1.c similarity index 90% rename from code/opus-1.2.1/celt/x86/celt_lpc_sse.c rename to code/opus-1.5.2/celt/x86/celt_lpc_sse4_1.c index 54785688..daf59d24 100644 --- a/code/opus-1.2.1/celt/x86/celt_lpc_sse.c +++ b/code/opus-1.5.2/celt/x86/celt_lpc_sse4_1.c @@ -64,9 +64,16 @@ void celt_fir_sse4_1(const opus_val16 *x, { opus_val32 sums[4] = {0}; __m128i vecSum, vecX; - - xcorr_kernel(rnum, x+i-ord, sums, ord, arch); - +#if defined(OPUS_CHECK_ASM) + { + opus_val32 sums_c[4] = {0}; + xcorr_kernel_c(rnum, x+i-ord, sums_c, ord); +#endif + xcorr_kernel(rnum, x+i-ord, sums, ord, arch); +#if defined(OPUS_CHECK_ASM) + celt_assert(memcmp(sums, sums_c, sizeof(sums)) == 0); + } +#endif vecSum = _mm_loadu_si128((__m128i *)sums); vecSum = _mm_add_epi32(vecSum, vecNoA); vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT); diff --git a/code/opus-1.5.2/celt/x86/pitch_avx.c b/code/opus-1.5.2/celt/x86/pitch_avx.c new file mode 100644 index 00000000..f731762d --- /dev/null +++ b/code/opus-1.5.2/celt/x86/pitch_avx.c @@ -0,0 +1,101 @@ +/* Copyright (c) 2023 Amazon */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + + +#include +#include "x86cpu.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(FIXED_POINT) + +/* Like the "regular" xcorr_kernel(), but computes 8 results at a time. */ +static void xcorr_kernel_avx(const float *x, const float *y, float sum[8], int len) +{ + __m256 xsum0, xsum1, xsum2, xsum3, xsum4, xsum5, xsum6, xsum7; + xsum7 = xsum6 = xsum5 = xsum4 = xsum3 = xsum2 = xsum1 = xsum0 = _mm256_setzero_ps(); + int i; + __m256 x0; + /* Compute 8 inner products using partial sums. */ + for (i=0;i0); + (void)arch; + for (i=0;i= 3); sum0 = _mm_setzero_si128(); @@ -177,19 +185,56 @@ void xcorr_kernel_sse4_1(const opus_val16 * x, const opus_val16 * y, opus_val32 vecSum = _mm_add_epi32(vecSum, sum2); } - for (;j=0, "Allocated too many pulses in the quick pass"); + celt_sig_assert(pulsesLeft>=0); /* This should never happen, but just in case it does (e.g. on silence) we fill the first bin with pulses. */ @@ -205,10 +205,10 @@ opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch) { __m128i y4; __m128i s4; - y4 = _mm_loadu_si128((__m128i*)&iy[j]); + y4 = _mm_loadu_si128((__m128i*)(void*)&iy[j]); s4 = _mm_castps_si128(_mm_loadu_ps(&signy[j])); y4 = _mm_xor_si128(_mm_add_epi32(y4, s4), s4); - _mm_storeu_si128((__m128i*)&iy[j], y4); + _mm_storeu_si128((__m128i*)(void*)&iy[j], y4); } RESTORE_STACK; return yy; diff --git a/code/opus-1.2.1/src/mlp.h b/code/opus-1.5.2/celt/x86/x86_arch_macros.h similarity index 80% rename from code/opus-1.2.1/src/mlp.h rename to code/opus-1.5.2/celt/x86/x86_arch_macros.h index 618e246e..975b443e 100644 --- a/code/opus-1.2.1/src/mlp.h +++ b/code/opus-1.5.2/celt/x86/x86_arch_macros.h @@ -1,5 +1,4 @@ -/* Copyright (c) 2008-2011 Octasic Inc. - Written by Jean-Marc Valin */ +/* Copyright (c) 2023 Amazon */ /* Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -25,19 +24,24 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef _MLP_H_ -#define _MLP_H_ +#ifdef _MSC_VER -#include "arch.h" +# ifdef OPUS_X86_MAY_HAVE_SSE +# ifndef __SSE__ +# define __SSE__ +# endif +# endif -typedef struct { - int layers; - const int *topo; - const float *weights; -} MLP; +# ifdef OPUS_X86_MAY_HAVE_SSE2 +# ifndef __SSE2__ +# define __SSE2__ +# endif +# endif -extern const MLP net; +# ifdef OPUS_X86_MAY_HAVE_SSE4_1 +# ifndef __SSE4_1__ +# define __SSE4_1__ +# endif +# endif -void mlp_process(const MLP *m, const float *in, float *out); - -#endif /* _MLP_H_ */ +#endif diff --git a/code/opus-1.2.1/celt/x86/x86_celt_map.c b/code/opus-1.5.2/celt/x86/x86_celt_map.c similarity index 92% rename from code/opus-1.2.1/celt/x86/x86_celt_map.c rename to code/opus-1.5.2/celt/x86/x86_celt_map.c index d39d88ed..ba8eafe6 100644 --- a/code/opus-1.2.1/celt/x86/x86_celt_map.c +++ b/code/opus-1.5.2/celt/x86/x86_celt_map.c @@ -90,6 +90,26 @@ opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( # else +#if defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_PRESUME_AVX2) + +void (*const PITCH_XCORR_IMPL[OPUS_ARCHMASK + 1])( + const float *_x, + const float *_y, + float *xcorr, + int len, + int max_pitch, + int arch +) = { + celt_pitch_xcorr_c, /* non-sse */ + celt_pitch_xcorr_c, + celt_pitch_xcorr_c, + celt_pitch_xcorr_c, + MAY_HAVE_AVX2(celt_pitch_xcorr) +}; + +#endif + + #if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE) void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( diff --git a/code/opus-1.2.1/celt/x86/x86cpu.c b/code/opus-1.5.2/celt/x86/x86cpu.c similarity index 68% rename from code/opus-1.2.1/celt/x86/x86cpu.c rename to code/opus-1.5.2/celt/x86/x86cpu.c index 080eb25e..2e7c32ae 100644 --- a/code/opus-1.2.1/celt/x86/x86cpu.c +++ b/code/opus-1.5.2/celt/x86/x86cpu.c @@ -35,11 +35,11 @@ #include "pitch.h" #include "x86cpu.h" -#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ +#if defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ - (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX)) - + (defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_PRESUME_AVX2))) #if defined(_MSC_VER) @@ -68,7 +68,8 @@ static void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) "=r" (CPUInfo[1]), "=c" (CPUInfo[2]), "=d" (CPUInfo[3]) : - "0" (InfoType) + /* We clear ECX to avoid a valgrind false-positive prior to v3.17.0. */ + "0" (InfoType), "2" (0) ); #else __asm__ __volatile__ ( @@ -77,11 +78,22 @@ static void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) "=b" (CPUInfo[1]), "=c" (CPUInfo[2]), "=d" (CPUInfo[3]) : - "0" (InfoType) + /* We clear ECX to avoid a valgrind false-positive prior to v3.17.0. */ + "0" (InfoType), "2" (0) ); #endif #elif defined(CPU_INFO_BY_C) - __get_cpuid(InfoType, &(CPUInfo[0]), &(CPUInfo[1]), &(CPUInfo[2]), &(CPUInfo[3])); + /* We use __get_cpuid_count to clear ECX to avoid a valgrind false-positive + prior to v3.17.0.*/ + if (!__get_cpuid_count(InfoType, 0, &(CPUInfo[0]), &(CPUInfo[1]), &(CPUInfo[2]), &(CPUInfo[3]))) { + /* Our function cannot fail, but __get_cpuid{_count} can. + Returning all zeroes will effectively disable all SIMD, which is + what we want on CPUs that don't support CPUID. */ + CPUInfo[3] = CPUInfo[2] = CPUInfo[1] = CPUInfo[0] = 0; + } +#else +# error "Configured to use x86 RTCD, but no CPU detection method available. " \ + "Reconfigure with --disable-rtcd (or send patches)." #endif } @@ -93,12 +105,12 @@ typedef struct CPU_Feature{ int HW_SSE2; int HW_SSE41; /* SIMD: 256-bit */ - int HW_AVX; + int HW_AVX2; } CPU_Feature; static void opus_cpu_feature_check(CPU_Feature *cpu_feature) { - unsigned int info[4] = {0}; + unsigned int info[4]; unsigned int nIds = 0; cpuid(info, 0); @@ -109,17 +121,23 @@ static void opus_cpu_feature_check(CPU_Feature *cpu_feature) cpu_feature->HW_SSE = (info[3] & (1 << 25)) != 0; cpu_feature->HW_SSE2 = (info[3] & (1 << 26)) != 0; cpu_feature->HW_SSE41 = (info[2] & (1 << 19)) != 0; - cpu_feature->HW_AVX = (info[2] & (1 << 28)) != 0; + cpu_feature->HW_AVX2 = (info[2] & (1 << 28)) != 0 && (info[2] & (1 << 12)) != 0; + if (cpu_feature->HW_AVX2 && nIds >= 7) { + cpuid(info, 7); + cpu_feature->HW_AVX2 = cpu_feature->HW_AVX2 && (info[1] & (1 << 5)) != 0; + } else { + cpu_feature->HW_AVX2 = 0; + } } else { cpu_feature->HW_SSE = 0; cpu_feature->HW_SSE2 = 0; cpu_feature->HW_SSE41 = 0; - cpu_feature->HW_AVX = 0; + cpu_feature->HW_AVX2 = 0; } } -int opus_select_arch(void) +static int opus_select_arch_impl(void) { CPU_Feature cpu_feature; int arch; @@ -145,7 +163,7 @@ int opus_select_arch(void) } arch++; - if (!cpu_feature.HW_AVX) + if (!cpu_feature.HW_AVX2) { return arch; } @@ -154,4 +172,13 @@ int opus_select_arch(void) return arch; } +int opus_select_arch(void) { + int arch = opus_select_arch_impl(); +#ifdef FUZZING + /* Randomly downgrade the architecture. */ + arch = rand()%(arch+1); +#endif + return arch; +} + #endif diff --git a/code/opus-1.5.2/celt/x86/x86cpu.h b/code/opus-1.5.2/celt/x86/x86cpu.h new file mode 100644 index 00000000..1e5b6a4c --- /dev/null +++ b/code/opus-1.5.2/celt/x86/x86cpu.h @@ -0,0 +1,112 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(X86CPU_H) +# define X86CPU_H + +# if defined(OPUS_X86_MAY_HAVE_SSE) +# define MAY_HAVE_SSE(name) name ## _sse +# else +# define MAY_HAVE_SSE(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE2) +# define MAY_HAVE_SSE2(name) name ## _sse2 +# else +# define MAY_HAVE_SSE2(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) +# define MAY_HAVE_SSE4_1(name) name ## _sse4_1 +# else +# define MAY_HAVE_SSE4_1(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_AVX2) +# define MAY_HAVE_AVX2(name) name ## _avx2 +# else +# define MAY_HAVE_AVX2(name) name ## _c +# endif + +# if defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_PRESUME_AVX2))) +int opus_select_arch(void); +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE2) +# include "opus_defines.h" + +/*MOVD should not impose any alignment restrictions, but the C standard does, + and UBSan will report errors if we actually make unaligned accesses. + Use this to work around those restrictions (which should hopefully all get + optimized to a single MOVD instruction). + GCC implemented _mm_loadu_si32() since GCC 11; HOWEVER, there is a bug! + https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99754 + LLVM implemented _mm_loadu_si32() since Clang 8.0, however the + __clang_major__ version number macro is unreliable, as vendors + (specifically, Apple) will use different numbering schemes than upstream. + Clang's advice is "use feature detection", but they do not provide feature + detection support for specific SIMD functions. + We follow the approach from the SIMDe project and instead detect unrelated + features that should be available in the version we want (see + ).*/ +# if defined(__clang__) +# if __has_warning("-Wextra-semi-stmt") || \ + __has_builtin(__builtin_rotateleft32) +# define OPUS_CLANG_8 (1) +# endif +# endif +# if !defined(_MSC_VER) && !OPUS_GNUC_PREREQ(11,3) && !defined(OPUS_CLANG_8) +# include +# include + +# ifdef _mm_loadu_si32 +# undef _mm_loadu_si32 +# endif +# define _mm_loadu_si32 WORKAROUND_mm_loadu_si32 +static inline __m128i WORKAROUND_mm_loadu_si32(void const* mem_addr) { + int val; + memcpy(&val, mem_addr, sizeof(val)); + return _mm_cvtsi32_si128(val); +} +# elif defined(_MSC_VER) + /* MSVC needs this for _mm_loadu_si32 */ +# include +# endif + +# define OP_CVTEPI8_EPI32_M32(x) \ + (_mm_cvtepi8_epi32(_mm_loadu_si32(x))) + +# define OP_CVTEPI16_EPI32_M64(x) \ + (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(void*)(x)))) + +# endif + +#endif diff --git a/code/opus-1.2.1/include/opus.h b/code/opus-1.5.2/include/opus.h similarity index 87% rename from code/opus-1.2.1/include/opus.h rename to code/opus-1.5.2/include/opus.h index 5be73ddf..eadeda75 100644 --- a/code/opus-1.2.1/include/opus.h +++ b/code/opus-1.5.2/include/opus.h @@ -103,7 +103,7 @@ extern "C" { * @endcode * * where opus_encoder_get_size() returns the required size for the encoder state. Note that - * future versions of this code may change the size, so no assuptions should be made about it. + * future versions of this code may change the size, so no assumptions should be made about it. * * The encoder state is always continuous in memory and only a shallow copy is sufficient * to copy it (e.g. memcpy()) @@ -198,7 +198,7 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_encoder_get_size(int channels); * This must be one of 8000, 12000, 16000, * 24000, or 48000. * @param [in] channels int: Number of channels (1 or 2) in input signal - * @param [in] application int: Coding mode (@ref OPUS_APPLICATION_VOIP/@ref OPUS_APPLICATION_AUDIO/@ref OPUS_APPLICATION_RESTRICTED_LOWDELAY) + * @param [in] application int: Coding mode (one of @ref OPUS_APPLICATION_VOIP, @ref OPUS_APPLICATION_AUDIO, or @ref OPUS_APPLICATION_RESTRICTED_LOWDELAY) * @param [out] error int*: @ref opus_errorcodes * @note Regardless of the sampling rate and number channels selected, the Opus encoder * can switch to a lower audio bandwidth or number of channels if the bitrate @@ -222,7 +222,7 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusEncoder *opus_encoder_create( * This must be one of 8000, 12000, 16000, * 24000, or 48000. * @param [in] channels int: Number of channels (1 or 2) in input signal - * @param [in] application int: Coding mode (OPUS_APPLICATION_VOIP/OPUS_APPLICATION_AUDIO/OPUS_APPLICATION_RESTRICTED_LOWDELAY) + * @param [in] application int: Coding mode (one of OPUS_APPLICATION_VOIP, OPUS_APPLICATION_AUDIO, or OPUS_APPLICATION_RESTRICTED_LOWDELAY) * @retval #OPUS_OK Success or @ref opus_errorcodes */ OPUS_EXPORT int opus_encoder_init( @@ -357,7 +357,7 @@ OPUS_EXPORT int opus_encoder_ctl(OpusEncoder *st, int request, ...) OPUS_ARG_NON * error = opus_decoder_init(dec, Fs, channels); * @endcode * where opus_decoder_get_size() returns the required size for the decoder state. Note that - * future versions of this code may change the size, so no assuptions should be made about it. + * future versions of this code may change the size, so no assumptions should be made about it. * * The decoder state is always continuous in memory and only a shallow copy is sufficient * to copy it (e.g. memcpy()) @@ -398,6 +398,21 @@ OPUS_EXPORT int opus_encoder_ctl(OpusEncoder *st, int request, ...) OPUS_ARG_NON */ typedef struct OpusDecoder OpusDecoder; +/** Opus DRED decoder. + * This contains the complete state of an Opus DRED decoder. + * It is position independent and can be freely copied. + * @see opus_dred_decoder_create,opus_dred_decoder_init + */ +typedef struct OpusDREDDecoder OpusDREDDecoder; + + +/** Opus DRED state. + * This contains the complete state of an Opus DRED packet. + * It is position independent and can be freely copied. + * @see opus_dred_create,opus_dred_init + */ +typedef struct OpusDRED OpusDRED; + /** Gets the size of an OpusDecoder structure. * @param [in] channels int: Number of channels. * This must be 1 or 2. @@ -511,6 +526,101 @@ OPUS_EXPORT int opus_decoder_ctl(OpusDecoder *st, int request, ...) OPUS_ARG_NON */ OPUS_EXPORT void opus_decoder_destroy(OpusDecoder *st); +/** Gets the size of an OpusDREDDecoder structure. + * @returns The size in bytes. + */ +OPUS_EXPORT int opus_dred_decoder_get_size(void); + +/** Allocates and initializes an OpusDREDDecoder state. + * @param [out] error int*: #OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_EXPORT OpusDREDDecoder *opus_dred_decoder_create(int *error); + +/** Initializes an OpusDREDDecoder state. + * @param[in] dec OpusDREDDecoder*: State to be initialized. + */ +OPUS_EXPORT int opus_dred_decoder_init(OpusDREDDecoder *dec); + +/** Frees an OpusDREDDecoder allocated by opus_dred_decoder_create(). + * @param[in] dec OpusDREDDecoder*: State to be freed. + */ +OPUS_EXPORT void opus_dred_decoder_destroy(OpusDREDDecoder *dec); + +/** Perform a CTL function on an Opus DRED decoder. + * + * Generally the request and subsequent arguments are generated + * by a convenience macro. + * @param dred_dec OpusDREDDecoder*: DRED Decoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls or + * @ref opus_decoderctls. + * @see opus_genericctls + * @see opus_decoderctls + */ +OPUS_EXPORT int opus_dred_decoder_ctl(OpusDREDDecoder *dred_dec, int request, ...); + +/** Gets the size of an OpusDRED structure. + * @returns The size in bytes. + */ +OPUS_EXPORT int opus_dred_get_size(void); + +/** Allocates and initializes a DRED state. + * @param [out] error int*: #OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_EXPORT OpusDRED *opus_dred_alloc(int *error); + +/** Frees an OpusDRED allocated by opus_dred_create(). + * @param[in] dec OpusDRED*: State to be freed. + */ +OPUS_EXPORT void opus_dred_free(OpusDRED *dec); + +/** Decode an Opus DRED packet. + * @param [in] dred_dec OpusDRED*: DRED Decoder state + * @param [in] dred OpusDRED*: DRED state + * @param [in] data char*: Input payload + * @param [in] len opus_int32: Number of bytes in payload + * @param [in] max_dred_samples opus_int32: Maximum number of DRED samples that may be needed (if available in the packet). + * @param [in] sampling_rate opus_int32: Sampling rate used for max_dred_samples argument. Needs not match the actual sampling rate of the decoder. + * @param [out] dred_end opus_int32*: Number of non-encoded (silence) samples between the DRED timestamp and the last DRED sample. + * @param [in] defer_processing int: Flag (0 or 1). If set to one, the CPU-intensive part of the DRED decoding is deferred until opus_dred_process() is called. + * @returns Offset (positive) of the first decoded DRED samples, zero if no DRED is present, or @ref opus_errorcodes + */ +OPUS_EXPORT int opus_dred_parse(OpusDREDDecoder *dred_dec, OpusDRED *dred, const unsigned char *data, opus_int32 len, opus_int32 max_dred_samples, opus_int32 sampling_rate, int *dred_end, int defer_processing) OPUS_ARG_NONNULL(1); + +/** Finish decoding an Opus DRED packet. The function only needs to be called if opus_dred_parse() was called with defer_processing=1. + * The source and destination will often be the same DRED state. + * @param [in] dred_dec OpusDRED*: DRED Decoder state + * @param [in] src OpusDRED*: Source DRED state to start the processing from. + * @param [out] dst OpusDRED*: Destination DRED state to store the updated state after processing. + * @returns @ref opus_errorcodes + */ +OPUS_EXPORT int opus_dred_process(OpusDREDDecoder *dred_dec, const OpusDRED *src, OpusDRED *dst); + +/** Decode audio from an Opus DRED packet with floating point output. + * @param [in] st OpusDecoder*: Decoder state + * @param [in] dred OpusDRED*: DRED state + * @param [in] dred_offset opus_int32: position of the redundancy to decode (in samples before the beginning of the real audio data in the packet). + * @param [out] pcm opus_int16*: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(opus_int16) + * @param [in] frame_size Number of samples per channel to decode in \a pcm. + * frame_size must be a multiple of 2.5 ms. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_EXPORT int opus_decoder_dred_decode(OpusDecoder *st, const OpusDRED *dred, opus_int32 dred_offset, opus_int16 *pcm, opus_int32 frame_size); + +/** Decode audio from an Opus DRED packet with floating point output. + * @param [in] st OpusDecoder*: Decoder state + * @param [in] dred OpusDRED*: DRED state + * @param [in] dred_offset opus_int32: position of the redundancy to decode (in samples before the beginning of the real audio data in the packet). + * @param [out] pcm float*: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(float) + * @param [in] frame_size Number of samples per channel to decode in \a pcm. + * frame_size must be a multiple of 2.5 ms. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_EXPORT int opus_decoder_dred_decode_float(OpusDecoder *st, const OpusDRED *dred, opus_int32 dred_offset, float *pcm, opus_int32 frame_size); + + /** Parse an opus packet into one or more frames. * Opus_decode will perform this operation internally so most applications do * not need to use this function. @@ -531,7 +641,7 @@ OPUS_EXPORT int opus_packet_parse( const unsigned char *frames[48], opus_int16 size[48], int *payload_offset -) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(5); /** Gets the bandwidth of an Opus packet. * @param [in] data char*: Opus packet @@ -583,6 +693,14 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_frames(const unsigned */ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_samples(const unsigned char packet[], opus_int32 len, opus_int32 Fs) OPUS_ARG_NONNULL(1); +/** Checks whether an Opus packet has LBRR. + * @param [in] packet char*: Opus packet + * @param [in] len opus_int32: Length of packet + * @returns 1 is LBRR is present, 0 otherwise + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_has_lbrr(const unsigned char packet[], opus_int32 len); + /** Gets the number of samples of an Opus packet. * @param [in] dec OpusDecoder*: Decoder state * @param [in] packet char*: Opus packet diff --git a/code/opus-1.2.1/include/opus_custom.h b/code/opus-1.5.2/include/opus_custom.h similarity index 98% rename from code/opus-1.2.1/include/opus_custom.h rename to code/opus-1.5.2/include/opus_custom.h index 41f36bf2..2f22d4b3 100644 --- a/code/opus-1.2.1/include/opus_custom.h +++ b/code/opus-1.5.2/include/opus_custom.h @@ -104,7 +104,8 @@ typedef struct OpusCustomDecoder OpusCustomDecoder; /** The mode contains all the information necessary to create an encoder. Both the encoder and decoder need to be initialized with exactly the same mode, otherwise the output will be - corrupted. + corrupted. The mode MUST NOT BE DESTROYED until the encoders and + decoders that use it are destroyed as well. @brief Mode configuration */ typedef struct OpusCustomMode OpusCustomMode; @@ -178,7 +179,7 @@ OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomEncoder *opus_custom_encode ) OPUS_ARG_NONNULL(1); -/** Destroys a an encoder state. +/** Destroys an encoder state. * @param[in] st OpusCustomEncoder*: State to be freed. */ OPUS_CUSTOM_EXPORT void opus_custom_encoder_destroy(OpusCustomEncoder *st); @@ -286,7 +287,7 @@ OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomDecoder *opus_custom_decode int *error ) OPUS_ARG_NONNULL(1); -/** Destroys a an decoder state. +/** Destroys a decoder state. * @param[in] st OpusCustomDecoder*: State to be freed. */ OPUS_CUSTOM_EXPORT void opus_custom_decoder_destroy(OpusCustomDecoder *st); diff --git a/code/opus-1.2.1/include/opus_defines.h b/code/opus-1.5.2/include/opus_defines.h similarity index 93% rename from code/opus-1.2.1/include/opus_defines.h rename to code/opus-1.5.2/include/opus_defines.h index 33c5acdb..cd8f4dde 100644 --- a/code/opus-1.2.1/include/opus_defines.h +++ b/code/opus-1.5.2/include/opus_defines.h @@ -64,7 +64,7 @@ extern "C" { /**Export control for opus functions */ #ifndef OPUS_EXPORT -# if defined(WIN32) +# if defined(_WIN32) # if defined(OPUS_BUILD) && defined(DLL_EXPORT) # define OPUS_EXPORT __declspec(dllexport) # else @@ -168,12 +168,33 @@ extern "C" { /* Don't use 4045, it's already taken by OPUS_GET_GAIN_REQUEST */ #define OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST 4046 #define OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST 4047 +#define OPUS_GET_IN_DTX_REQUEST 4049 +#define OPUS_SET_DRED_DURATION_REQUEST 4050 +#define OPUS_GET_DRED_DURATION_REQUEST 4051 +#define OPUS_SET_DNN_BLOB_REQUEST 4052 +/*#define OPUS_GET_DNN_BLOB_REQUEST 4053 */ + +/** Defines for the presence of extended APIs. */ +#define OPUS_HAVE_OPUS_PROJECTION_H /* Macros to trigger compilation errors when the wrong types are provided to a CTL */ #define __opus_check_int(x) (((void)((x) == (opus_int32)0)), (opus_int32)(x)) + +#ifdef DISABLE_PTR_CHECK +/* Disable checks to prevent ubsan from complaining about NULL checks + in test_opus_api. */ +#define __opus_check_int_ptr(ptr) (ptr) +#define __opus_check_uint_ptr(ptr) (ptr) +#define __opus_check_uint8_ptr(ptr) (ptr) +#define __opus_check_val16_ptr(ptr) (ptr) +#define __opus_check_void_ptr(ptr) (ptr) +#else #define __opus_check_int_ptr(ptr) ((ptr) + ((ptr) - (opus_int32*)(ptr))) #define __opus_check_uint_ptr(ptr) ((ptr) + ((ptr) - (opus_uint32*)(ptr))) +#define __opus_check_uint8_ptr(ptr) ((ptr) + ((ptr) - (opus_uint8*)(ptr))) #define __opus_check_val16_ptr(ptr) ((ptr) + ((ptr) - (opus_val16*)(ptr))) +#define __opus_check_void_ptr(x) ((void)((void *)0 == (x)), (x)) +#endif /** @endcond */ /** @defgroup opus_ctlvalues Pre-defined values for CTL interface @@ -478,7 +499,8 @@ extern "C" { * @param[in] x opus_int32: Allowed values: *
*
0
Disable inband FEC (default).
- *
1
Enable inband FEC.
+ *
1
Inband FEC enabled. If the packet loss rate is sufficiently high, Opus will automatically switch to SILK even at high rates to enable use of that FEC.
+ *
2
Inband FEC enabled, but does not necessarily switch to SILK if we have music.
*
* @hideinitializer */ #define OPUS_SET_INBAND_FEC(x) OPUS_SET_INBAND_FEC_REQUEST, __opus_check_int(x) @@ -487,7 +509,8 @@ extern "C" { * @param[out] x opus_int32 *: Returns one of the following values: *
*
0
Inband FEC disabled (default).
- *
1
Inband FEC enabled.
+ *
1
Inband FEC enabled. If the packet loss rate is sufficiently high, Opus will automatically switch to SILK even at high rates to enable use of that FEC.
+ *
2
Inband FEC enabled, but does not necessarily switch to SILK if we have music.
*
* @hideinitializer */ #define OPUS_GET_INBAND_FEC(x) OPUS_GET_INBAND_FEC_REQUEST, __opus_check_int_ptr(x) @@ -614,6 +637,18 @@ extern "C" { * @hideinitializer */ #define OPUS_GET_PREDICTION_DISABLED(x) OPUS_GET_PREDICTION_DISABLED_REQUEST, __opus_check_int_ptr(x) +/** If non-zero, enables Deep Redundancy (DRED) and use the specified maximum number of 10-ms redundant frames + * @hideinitializer */ +#define OPUS_SET_DRED_DURATION(x) OPUS_SET_DRED_DURATION_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured Deep Redundancy (DRED) maximum number of frames. + * @hideinitializer */ +#define OPUS_GET_DRED_DURATION(x) OPUS_GET_DRED_DURATION_REQUEST, __opus_check_int_ptr(x) + +/** Provide external DNN weights from binary object (only when explicitly built without the weights) + * @hideinitializer */ +#define OPUS_SET_DNN_BLOB(data, len) OPUS_SET_DNN_BLOB_REQUEST, __opus_check_void_ptr(data), __opus_check_int(len) + + /**@}*/ /** @defgroup opus_genericctls Generic CTLs @@ -712,6 +747,16 @@ extern "C" { * * @hideinitializer */ #define OPUS_GET_PHASE_INVERSION_DISABLED(x) OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST, __opus_check_int_ptr(x) +/** Gets the DTX state of the encoder. + * Returns whether the last encoded frame was either a comfort noise update + * during DTX or not encoded because of DTX. + * @param[out] x opus_int32 *: Returns one of the following values: + *
+ *
0
The encoder is not in DTX.
+ *
1
The encoder is in DTX.
+ *
+ * @hideinitializer */ +#define OPUS_GET_IN_DTX(x) OPUS_GET_IN_DTX_REQUEST, __opus_check_int_ptr(x) /**@}*/ diff --git a/code/opus-1.2.1/include/opus_multistream.h b/code/opus-1.5.2/include/opus_multistream.h similarity index 99% rename from code/opus-1.2.1/include/opus_multistream.h rename to code/opus-1.5.2/include/opus_multistream.h index babcee69..824cc55a 100644 --- a/code/opus-1.2.1/include/opus_multistream.h +++ b/code/opus-1.5.2/include/opus_multistream.h @@ -143,7 +143,7 @@ extern "C" { * Vorbis * channel ordering. A decoder may wish to apply an additional permutation * to the mapping the encoder used to achieve a different output channel - * order (e.g. for outputing in WAV order). + * order (e.g. for outputting in WAV order). * * Each multistream packet contains an Opus packet for each stream, and all of * the Opus packets in a single multistream packet must have the same diff --git a/code/opus-1.5.2/include/opus_projection.h b/code/opus-1.5.2/include/opus_projection.h new file mode 100644 index 00000000..9dabf4e8 --- /dev/null +++ b/code/opus-1.5.2/include/opus_projection.h @@ -0,0 +1,568 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file opus_projection.h + * @brief Opus projection reference API + */ + +#ifndef OPUS_PROJECTION_H +#define OPUS_PROJECTION_H + +#include "opus_multistream.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** @cond OPUS_INTERNAL_DOC */ + +/** These are the actual encoder and decoder CTL ID numbers. + * They should not be used directly by applications.c + * In general, SETs should be even and GETs should be odd.*/ +/**@{*/ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST 6001 +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST 6003 +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST 6005 +/**@}*/ + + +/** @endcond */ + +/** @defgroup opus_projection_ctls Projection specific encoder and decoder CTLs + * + * These are convenience macros that are specific to the + * opus_projection_encoder_ctl() and opus_projection_decoder_ctl() + * interface. + * The CTLs from @ref opus_genericctls, @ref opus_encoderctls, + * @ref opus_decoderctls, and @ref opus_multistream_ctls may be applied to a + * projection encoder or decoder as well. + */ +/**@{*/ + +/** Gets the gain (in dB. S7.8-format) of the demixing matrix from the encoder. + * @param[out] x opus_int32 *: Returns the gain (in dB. S7.8-format) + * of the demixing matrix. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN(x) OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST, __opus_check_int_ptr(x) + + +/** Gets the size in bytes of the demixing matrix from the encoder. + * @param[out] x opus_int32 *: Returns the size in bytes of the + * demixing matrix. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE(x) OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST, __opus_check_int_ptr(x) + + +/** Copies the demixing matrix to the supplied pointer location. + * @param[out] x unsigned char *: Returns the demixing matrix to the + * supplied pointer location. + * @param y opus_int32: The size in bytes of the reserved memory at the + * pointer location. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX(x,y) OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST, x, __opus_check_int(y) + + +/**@}*/ + +/** Opus projection encoder state. + * This contains the complete state of a projection Opus encoder. + * It is position independent and can be freely copied. + * @see opus_projection_ambisonics_encoder_create + */ +typedef struct OpusProjectionEncoder OpusProjectionEncoder; + + +/** Opus projection decoder state. + * This contains the complete state of a projection Opus decoder. + * It is position independent and can be freely copied. + * @see opus_projection_decoder_create + * @see opus_projection_decoder_init + */ +typedef struct OpusProjectionDecoder OpusProjectionDecoder; + + +/**\name Projection encoder functions */ +/**@{*/ + +/** Gets the size of an OpusProjectionEncoder structure. + * @param channels int: The total number of input channels to encode. + * This must be no more than 255. + * @param mapping_family int: The mapping family to use for selecting + * the appropriate projection. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_projection_ambisonics_encoder_get_size( + int channels, + int mapping_family +); + + +/** Allocates and initializes a projection encoder state. + * Call opus_projection_encoder_destroy() to release + * this object when finished. + * @param Fs opus_int32: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels int: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (streams + + * coupled_streams). + * @param mapping_family int: The mapping family to use for selecting + * the appropriate projection. + * @param[out] streams int *: The total number of streams that will + * be encoded from the input. + * @param[out] coupled_streams int *: Number of coupled (2 channel) + * streams that will be encoded from the input. + * @param application int: The target encoder application. + * This must be one of the following: + *
+ *
#OPUS_APPLICATION_VOIP
+ *
Process signal for improved speech intelligibility.
+ *
#OPUS_APPLICATION_AUDIO
+ *
Favor faithfulness to the original input.
+ *
#OPUS_APPLICATION_RESTRICTED_LOWDELAY
+ *
Configure the minimum possible coding delay by disabling certain modes + * of operation.
+ *
+ * @param[out] error int *: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusProjectionEncoder *opus_projection_ambisonics_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + int application, + int *error +) OPUS_ARG_NONNULL(4) OPUS_ARG_NONNULL(5); + + +/** Initialize a previously allocated projection encoder state. + * The memory pointed to by \a st must be at least the size returned by + * opus_projection_ambisonics_encoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_projection_ambisonics_encoder_create + * @see opus_projection_ambisonics_encoder_get_size + * @param st OpusProjectionEncoder*: Projection encoder state to initialize. + * @param Fs opus_int32: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels int: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (streams + + * coupled_streams). + * @param streams int: The total number of streams to encode from the + * input. + * This must be no more than the number of channels. + * @param coupled_streams int: Number of coupled (2 channel) streams + * to encode. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * encoded channels (streams + + * coupled_streams) must be no + * more than the number of input channels. + * @param application int: The target encoder application. + * This must be one of the following: + *
+ *
#OPUS_APPLICATION_VOIP
+ *
Process signal for improved speech intelligibility.
+ *
#OPUS_APPLICATION_AUDIO
+ *
Favor faithfulness to the original input.
+ *
#OPUS_APPLICATION_RESTRICTED_LOWDELAY
+ *
Configure the minimum possible coding delay by disabling certain modes + * of operation.
+ *
+ * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_projection_ambisonics_encoder_init( + OpusProjectionEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + int application +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(5) OPUS_ARG_NONNULL(6); + + +/** Encodes a projection Opus frame. + * @param st OpusProjectionEncoder*: Projection encoder state. + * @param[in] pcm const opus_int16*: The input signal as interleaved + * samples. + * This must contain + * frame_size*channels + * samples. + * @param frame_size int: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data unsigned char*: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes opus_int32: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_encode( + OpusProjectionEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + + +/** Encodes a projection Opus frame from floating point input. + * @param st OpusProjectionEncoder*: Projection encoder state. + * @param[in] pcm const float*: The input signal as interleaved + * samples with a normal range of + * +/-1.0. + * Samples with a range beyond +/-1.0 + * are supported but will be clipped by + * decoders using the integer API and + * should only be used if it is known + * that the far end supports extended + * dynamic range. + * This must contain + * frame_size*channels + * samples. + * @param frame_size int: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data unsigned char*: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes opus_int32: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_encode_float( + OpusProjectionEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + + +/** Frees an OpusProjectionEncoder allocated by + * opus_projection_ambisonics_encoder_create(). + * @param st OpusProjectionEncoder*: Projection encoder state to be freed. + */ +OPUS_EXPORT void opus_projection_encoder_destroy(OpusProjectionEncoder *st); + + +/** Perform a CTL function on a projection Opus encoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st OpusProjectionEncoder*: Projection encoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_encoderctls, @ref opus_multistream_ctls, or + * @ref opus_projection_ctls + * @see opus_genericctls + * @see opus_encoderctls + * @see opus_multistream_ctls + * @see opus_projection_ctls + */ +OPUS_EXPORT int opus_projection_encoder_ctl(OpusProjectionEncoder *st, int request, ...) OPUS_ARG_NONNULL(1); + + +/**@}*/ + +/**\name Projection decoder functions */ +/**@{*/ + +/** Gets the size of an OpusProjectionDecoder structure. + * @param channels int: The total number of output channels. + * This must be no more than 255. + * @param streams int: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams int: Number streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (streams + + * coupled_streams) must be no + * more than 255. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_projection_decoder_get_size( + int channels, + int streams, + int coupled_streams +); + + +/** Allocates and initializes a projection decoder state. + * Call opus_projection_decoder_destroy() to release + * this object when finished. + * @param Fs opus_int32: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels int: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (streams + + * coupled_streams). + * @param streams int: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams int: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (streams + + * coupled_streams) must be no + * more than 255. + * @param[in] demixing_matrix const unsigned char[demixing_matrix_size]: Demixing matrix + * that mapping from coded channels to output channels, + * as described in @ref opus_projection and + * @ref opus_projection_ctls. + * @param demixing_matrix_size opus_int32: The size in bytes of the + * demixing matrix, as + * described in @ref + * opus_projection_ctls. + * @param[out] error int *: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusProjectionDecoder *opus_projection_decoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + unsigned char *demixing_matrix, + opus_int32 demixing_matrix_size, + int *error +) OPUS_ARG_NONNULL(5); + + +/** Intialize a previously allocated projection decoder state object. + * The memory pointed to by \a st must be at least the size returned by + * opus_projection_decoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_projection_decoder_create + * @see opus_projection_deocder_get_size + * @param st OpusProjectionDecoder*: Projection encoder state to initialize. + * @param Fs opus_int32: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels int: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (streams + + * coupled_streams). + * @param streams int: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams int: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (streams + + * coupled_streams) must be no + * more than 255. + * @param[in] demixing_matrix const unsigned char[demixing_matrix_size]: Demixing matrix + * that mapping from coded channels to output channels, + * as described in @ref opus_projection and + * @ref opus_projection_ctls. + * @param demixing_matrix_size opus_int32: The size in bytes of the + * demixing matrix, as + * described in @ref + * opus_projection_ctls. + * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_projection_decoder_init( + OpusProjectionDecoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + unsigned char *demixing_matrix, + opus_int32 demixing_matrix_size +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); + + +/** Decode a projection Opus packet. + * @param st OpusProjectionDecoder*: Projection decoder state. + * @param[in] data const unsigned char*: Input payload. + * Use a NULL + * pointer to indicate packet + * loss. + * @param len opus_int32: Number of bytes in payload. + * @param[out] pcm opus_int16*: Output signal, with interleaved + * samples. + * This must contain room for + * frame_size*channels + * samples. + * @param frame_size int: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * must be a multiple of 2.5 ms. + * @param decode_fec int: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_decode( + OpusProjectionDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + + +/** Decode a projection Opus packet with floating point output. + * @param st OpusProjectionDecoder*: Projection decoder state. + * @param[in] data const unsigned char*: Input payload. + * Use a NULL + * pointer to indicate packet + * loss. + * @param len opus_int32: Number of bytes in payload. + * @param[out] pcm opus_int16*: Output signal, with interleaved + * samples. + * This must contain room for + * frame_size*channels + * samples. + * @param frame_size int: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * must be a multiple of 2.5 ms. + * @param decode_fec int: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_decode_float( + OpusProjectionDecoder *st, + const unsigned char *data, + opus_int32 len, + float *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + + +/** Perform a CTL function on a projection Opus decoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st OpusProjectionDecoder*: Projection decoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_decoderctls, @ref opus_multistream_ctls, or + * @ref opus_projection_ctls. + * @see opus_genericctls + * @see opus_decoderctls + * @see opus_multistream_ctls + * @see opus_projection_ctls + */ +OPUS_EXPORT int opus_projection_decoder_ctl(OpusProjectionDecoder *st, int request, ...) OPUS_ARG_NONNULL(1); + + +/** Frees an OpusProjectionDecoder allocated by + * opus_projection_decoder_create(). + * @param st OpusProjectionDecoder: Projection decoder state to be freed. + */ +OPUS_EXPORT void opus_projection_decoder_destroy(OpusProjectionDecoder *st); + + +/**@}*/ + +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_PROJECTION_H */ diff --git a/code/opus-1.2.1/include/opus_types.h b/code/opus-1.5.2/include/opus_types.h similarity index 96% rename from code/opus-1.2.1/include/opus_types.h rename to code/opus-1.5.2/include/opus_types.h index 71808266..7cf67558 100644 --- a/code/opus-1.2.1/include/opus_types.h +++ b/code/opus-1.5.2/include/opus_types.h @@ -33,14 +33,29 @@ #ifndef OPUS_TYPES_H #define OPUS_TYPES_H +#define opus_int int /* used for counters etc; at least 16 bits */ +#define opus_int64 long long +#define opus_int8 signed char + +#define opus_uint unsigned int /* used for counters etc; at least 16 bits */ +#define opus_uint64 unsigned long long +#define opus_uint8 unsigned char + /* Use the real stdint.h if it's there (taken from Paul Hsieh's pstdint.h) */ #if (defined(__STDC__) && __STDC__ && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L) || (defined(__GNUC__) && (defined(_STDINT_H) || defined(_STDINT_H_)) || defined (HAVE_STDINT_H)) #include - +# undef opus_int64 +# undef opus_int8 +# undef opus_uint64 +# undef opus_uint8 + typedef int8_t opus_int8; + typedef uint8_t opus_uint8; typedef int16_t opus_int16; typedef uint16_t opus_uint16; typedef int32_t opus_int32; typedef uint32_t opus_uint32; + typedef int64_t opus_int64; + typedef uint64_t opus_uint64; #elif defined(_WIN32) # if defined(__CYGWIN__) @@ -148,12 +163,4 @@ #endif -#define opus_int int /* used for counters etc; at least 16 bits */ -#define opus_int64 long long -#define opus_int8 signed char - -#define opus_uint unsigned int /* used for counters etc; at least 16 bits */ -#define opus_uint64 unsigned long long -#define opus_uint8 unsigned char - #endif /* OPUS_TYPES_H */ diff --git a/code/opus-1.2.1/silk/A2NLSF.c b/code/opus-1.5.2/silk/A2NLSF.c similarity index 100% rename from code/opus-1.2.1/silk/A2NLSF.c rename to code/opus-1.5.2/silk/A2NLSF.c diff --git a/code/opus-1.2.1/silk/API.h b/code/opus-1.5.2/silk/API.h similarity index 86% rename from code/opus-1.2.1/silk/API.h rename to code/opus-1.5.2/silk/API.h index 0131acbb..878965c7 100644 --- a/code/opus-1.2.1/silk/API.h +++ b/code/opus-1.5.2/silk/API.h @@ -34,6 +34,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "entenc.h" #include "entdec.h" +#ifdef ENABLE_DEEP_PLC +#include "lpcnet_private.h" +#endif + #ifdef __cplusplus extern "C" { @@ -80,13 +84,24 @@ opus_int silk_Encode( /* O Returns error co opus_int nSamplesIn, /* I Number of samples in input vector */ ec_enc *psRangeEnc, /* I/O Compressor data structure */ opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */ - const opus_int prefillFlag /* I Flag to indicate prefilling buffers no coding */ + const opus_int prefillFlag, /* I Flag to indicate prefilling buffers no coding */ + int activity /* I Decision of Opus voice activity detector */ ); /****************************************/ /* Decoder functions */ /****************************************/ + +/***********************************************/ +/* Load OSCE models from external data pointer */ +/***********************************************/ +opus_int silk_LoadOSCEModels( + void *decState, /* O I/O State */ + const unsigned char *data, /* I pointer to binary blob */ + int len /* I length of binary blob data */ +); + /***********************************************/ /* Get size in bytes of the Silk decoder state */ /***********************************************/ @@ -95,8 +110,12 @@ opus_int silk_Get_Decoder_Size( /* O Returns error co ); /*************************/ -/* Init or Reset decoder */ +/* Init and Reset decoder */ /*************************/ +opus_int silk_ResetDecoder( /* O Returns error code */ + void *decState /* I/O State */ +); + opus_int silk_InitDecoder( /* O Returns error code */ void *decState /* I/O State */ ); @@ -112,6 +131,9 @@ opus_int silk_Decode( /* O Returns error co ec_dec *psRangeDec, /* I/O Compressor data structure */ opus_int16 *samplesOut, /* O Decoded output speech vector */ opus_int32 *nSamplesOut, /* O Number of samples decoded */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ); diff --git a/code/opus-1.2.1/silk/CNG.c b/code/opus-1.5.2/silk/CNG.c similarity index 95% rename from code/opus-1.2.1/silk/CNG.c rename to code/opus-1.5.2/silk/CNG.c index e6d9b861..2a910099 100644 --- a/code/opus-1.2.1/silk/CNG.c +++ b/code/opus-1.5.2/silk/CNG.c @@ -118,6 +118,10 @@ void silk_CNG( /* Smooth gains */ for( i = 0; i < psDec->nb_subfr; i++ ) { psCNG->CNG_smth_Gain_Q16 += silk_SMULWB( psDecCtrl->Gains_Q16[ i ] - psCNG->CNG_smth_Gain_Q16, CNG_GAIN_SMTH_Q16 ); + /* If the smoothed gain is 3 dB greater than this subframe's gain, use this subframe's gain to adapt faster. */ + if( silk_SMULWW( psCNG->CNG_smth_Gain_Q16, CNG_GAIN_SMTH_THRESHOLD_Q16 ) > psDecCtrl->Gains_Q16[ i ] ) { + psCNG->CNG_smth_Gain_Q16 = psDecCtrl->Gains_Q16[ i ]; + } } } @@ -146,8 +150,8 @@ void silk_CNG( /* Generate CNG signal, by synthesis filtering */ silk_memcpy( CNG_sig_Q14, psCNG->CNG_synth_state, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + celt_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); for( i = 0; i < length; i++ ) { - silk_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12[ 0 ] ); diff --git a/code/opus-1.2.1/silk/HP_variable_cutoff.c b/code/opus-1.5.2/silk/HP_variable_cutoff.c similarity index 100% rename from code/opus-1.2.1/silk/HP_variable_cutoff.c rename to code/opus-1.5.2/silk/HP_variable_cutoff.c diff --git a/code/opus-1.2.1/silk/Inlines.h b/code/opus-1.5.2/silk/Inlines.h similarity index 100% rename from code/opus-1.2.1/silk/Inlines.h rename to code/opus-1.5.2/silk/Inlines.h diff --git a/code/opus-1.2.1/silk/LPC_analysis_filter.c b/code/opus-1.5.2/silk/LPC_analysis_filter.c similarity index 97% rename from code/opus-1.2.1/silk/LPC_analysis_filter.c rename to code/opus-1.5.2/silk/LPC_analysis_filter.c index 7715f70f..d34b5eb7 100644 --- a/code/opus-1.2.1/silk/LPC_analysis_filter.c +++ b/code/opus-1.5.2/silk/LPC_analysis_filter.c @@ -64,12 +64,12 @@ void silk_LPC_analysis_filter( const opus_int16 *in_ptr; #endif - silk_assert( d >= 6 ); - silk_assert( (d & 1) == 0 ); - silk_assert( d <= len ); + celt_assert( d >= 6 ); + celt_assert( (d & 1) == 0 ); + celt_assert( d <= len ); #if defined(FIXED_POINT) && USE_CELT_FIR - silk_assert( d <= SILK_MAX_ORDER_LPC ); + celt_assert( d <= SILK_MAX_ORDER_LPC ); for ( j = 0; j < d; j++ ) { num[ j ] = -B[ j ]; } diff --git a/code/opus-1.2.1/silk/LPC_fit.c b/code/opus-1.5.2/silk/LPC_fit.c similarity index 97% rename from code/opus-1.2.1/silk/LPC_fit.c rename to code/opus-1.5.2/silk/LPC_fit.c index cdea4f3a..c0690a1f 100644 --- a/code/opus-1.2.1/silk/LPC_fit.c +++ b/code/opus-1.5.2/silk/LPC_fit.c @@ -31,7 +31,8 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" -/* Convert int32 coefficients to int16 coefs and make sure there's no wrap-around */ +/* Convert int32 coefficients to int16 coefs and make sure there's no wrap-around. + This logic is reused in _celt_lpc(). Any bug fixes should also be applied there. */ void silk_LPC_fit( opus_int16 *a_QOUT, /* O Output signal */ opus_int32 *a_QIN, /* I/O Input signal */ diff --git a/code/opus-1.2.1/silk/LPC_inv_pred_gain.c b/code/opus-1.5.2/silk/LPC_inv_pred_gain.c similarity index 100% rename from code/opus-1.2.1/silk/LPC_inv_pred_gain.c rename to code/opus-1.5.2/silk/LPC_inv_pred_gain.c diff --git a/code/opus-1.2.1/silk/LP_variable_cutoff.c b/code/opus-1.5.2/silk/LP_variable_cutoff.c similarity index 100% rename from code/opus-1.2.1/silk/LP_variable_cutoff.c rename to code/opus-1.5.2/silk/LP_variable_cutoff.c diff --git a/code/opus-1.2.1/silk/MacroCount.h b/code/opus-1.5.2/silk/MacroCount.h similarity index 100% rename from code/opus-1.2.1/silk/MacroCount.h rename to code/opus-1.5.2/silk/MacroCount.h index 78100ffe..dab2f57a 100644 --- a/code/opus-1.2.1/silk/MacroCount.h +++ b/code/opus-1.5.2/silk/MacroCount.h @@ -27,9 +27,9 @@ POSSIBILITY OF SUCH DAMAGE. #ifndef SIGPROCFIX_API_MACROCOUNT_H #define SIGPROCFIX_API_MACROCOUNT_H -#include #ifdef silk_MACRO_COUNT +#include #define varDefine opus_int64 ops_count = 0; extern opus_int64 ops_count; diff --git a/code/opus-1.2.1/silk/MacroDebug.h b/code/opus-1.5.2/silk/MacroDebug.h similarity index 96% rename from code/opus-1.2.1/silk/MacroDebug.h rename to code/opus-1.5.2/silk/MacroDebug.h index 8dd4ce2e..3110da9a 100644 --- a/code/opus-1.2.1/silk/MacroDebug.h +++ b/code/opus-1.5.2/silk/MacroDebug.h @@ -55,7 +55,7 @@ static OPUS_INLINE opus_int16 silk_ADD16_(opus_int16 a, opus_int16 b, char *file static OPUS_INLINE opus_int32 silk_ADD32_(opus_int32 a, opus_int32 b, char *file, int line){ opus_int32 ret; - ret = a + b; + ret = (opus_int32)((opus_uint32)a + (opus_uint32)b); if ( ret != silk_ADD_SAT32( a, b ) ) { fprintf (stderr, "silk_ADD32(%d, %d) in %s: line %d\n", a, b, file, line); @@ -101,9 +101,9 @@ static OPUS_INLINE opus_int16 silk_SUB16_(opus_int16 a, opus_int16 b, char *file #undef silk_SUB32 #define silk_SUB32(a,b) silk_SUB32_((a), (b), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_SUB32_(opus_int32 a, opus_int32 b, char *file, int line){ - opus_int32 ret; + opus_int64 ret; - ret = a - b; + ret = a - (opus_int64)b; if ( ret != silk_SUB_SAT32( a, b ) ) { fprintf (stderr, "silk_SUB32(%d, %d) in %s: line %d\n", a, b, file, line); @@ -257,7 +257,7 @@ static OPUS_INLINE opus_int64 silk_SUB_SAT64_( opus_int64 a64, opus_int64 b64, c static OPUS_INLINE opus_int32 silk_MUL_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 ret; opus_int64 ret64; - ret = a32 * b32; + ret = (opus_int32)((opus_uint32)a32 * (opus_uint32)b32); ret64 = (opus_int64)a32 * (opus_int64)b32; if ( (opus_int64)ret != ret64 ) { @@ -333,8 +333,8 @@ static OPUS_INLINE opus_int32 silk_SMULWB_(opus_int32 a32, opus_int32 b32, char #define silk_SMLAWB(a,b,c) silk_SMLAWB_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_SMLAWB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; - ret = silk_ADD32( a32, silk_SMULWB( b32, c32 ) ); - if ( silk_ADD32( a32, silk_SMULWB( b32, c32 ) ) != silk_ADD_SAT32( a32, silk_SMULWB( b32, c32 ) ) ) + ret = silk_ADD32_ovflw( a32, silk_SMULWB( b32, c32 ) ); + if ( ret != silk_ADD_SAT32( a32, silk_SMULWB( b32, c32 ) ) ) { fprintf (stderr, "silk_SMLAWB(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); #ifdef FIXED_DEBUG_ASSERT @@ -465,7 +465,7 @@ static OPUS_INLINE opus_int32 silk_SMULWW_(opus_int32 a32, opus_int32 b32, char if ( fail ) { - fprintf (stderr, "silk_SMULWT(%d, %d) in %s: line %d\n", a32, b32, file, line); + fprintf (stderr, "silk_SMULWW(%d, %d) in %s: line %d\n", a32, b32, file, line); #ifdef FIXED_DEBUG_ASSERT silk_assert( 0 ); #endif @@ -491,12 +491,6 @@ static OPUS_INLINE opus_int32 silk_SMLAWW_(opus_int32 a32, opus_int32 b32, opus_ return ret; } -/* Multiply-accumulate macros that allow overflow in the addition (ie, no asserts in debug mode) */ -#undef silk_MLA_ovflw -#define silk_MLA_ovflw(a32, b32, c32) ((a32) + ((b32) * (c32))) -#undef silk_SMLABB_ovflw -#define silk_SMLABB_ovflw(a32, b32, c32) ((a32) + ((opus_int32)((opus_int16)(b32))) * (opus_int32)((opus_int16)(c32))) - /* no checking needed for silk_SMULL no checking needed for silk_SMLAL no checking needed for silk_SMLALBB @@ -546,10 +540,10 @@ static OPUS_INLINE opus_int32 silk_DIV32_16_(opus_int32 a32, opus_int32 b32, cha static OPUS_INLINE opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ opus_int8 ret; int fail = 0; - ret = a << shift; + ret = (opus_int8)((opus_uint8)a << shift); fail |= shift < 0; fail |= shift >= 8; - fail |= (opus_int64)ret != ((opus_int64)a) << shift; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); if ( fail ) { fprintf (stderr, "silk_LSHIFT8(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -565,10 +559,10 @@ static OPUS_INLINE opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char * static OPUS_INLINE opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ opus_int16 ret; int fail = 0; - ret = a << shift; + ret = (opus_int16)((opus_uint16)a << shift); fail |= shift < 0; fail |= shift >= 16; - fail |= (opus_int64)ret != ((opus_int64)a) << shift; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); if ( fail ) { fprintf (stderr, "silk_LSHIFT16(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -584,10 +578,10 @@ static OPUS_INLINE opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, cha static OPUS_INLINE opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ opus_int32 ret; int fail = 0; - ret = a << shift; + ret = (opus_int32)((opus_uint32)a << shift); fail |= shift < 0; fail |= shift >= 32; - fail |= (opus_int64)ret != ((opus_int64)a) << shift; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); if ( fail ) { fprintf (stderr, "silk_LSHIFT32(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -603,7 +597,7 @@ static OPUS_INLINE opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, cha static OPUS_INLINE opus_int64 silk_LSHIFT64_(opus_int64 a, opus_int shift, char *file, int line){ opus_int64 ret; int fail = 0; - ret = a << shift; + ret = (opus_int64)((opus_uint64)a << shift); fail |= shift < 0; fail |= shift >= 64; fail |= (ret>>shift) != ((opus_int64)a); @@ -714,8 +708,8 @@ static OPUS_INLINE opus_uint32 silk_RSHIFT_uint_(opus_uint32 a, opus_int32 shift #define silk_ADD_LSHIFT(a,b,c) silk_ADD_LSHIFT_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int line){ opus_int16 ret; - ret = a + (b << shift); - if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) + ret = a + (opus_int16)((opus_uint16)b << shift); + if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (opus_int64)(((opus_uint64)b) << shift)) ) { fprintf (stderr, "silk_ADD_LSHIFT(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); #ifdef FIXED_DEBUG_ASSERT @@ -729,8 +723,8 @@ static OPUS_INLINE int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int #define silk_ADD_LSHIFT32(a,b,c) silk_ADD_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_ADD_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; - ret = a + (b << shift); - if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) + ret = silk_ADD32_ovflw(a, (opus_int32)((opus_uint32)b << shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (opus_int64)(((opus_uint64)b) << shift)) ) { fprintf (stderr, "silk_ADD_LSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); #ifdef FIXED_DEBUG_ASSERT @@ -774,7 +768,7 @@ static OPUS_INLINE int silk_ADD_RSHIFT_(int a, int b, int shift, char *file, int #define silk_ADD_RSHIFT32(a,b,c) silk_ADD_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_ADD_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; - ret = a + (b >> shift); + ret = silk_ADD32_ovflw(a, (b >> shift)); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) { fprintf (stderr, "silk_ADD_RSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); @@ -804,8 +798,8 @@ static OPUS_INLINE opus_uint32 silk_ADD_RSHIFT_uint_(opus_uint32 a, opus_uint32 #define silk_SUB_LSHIFT32(a,b,c) silk_SUB_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; - ret = a - (b << shift); - if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (((opus_int64)b) << shift)) ) + ret = silk_SUB32_ovflw(a, (opus_int32)((opus_uint32)b << shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (opus_int64)(((opus_uint64)b) << shift)) ) { fprintf (stderr, "silk_SUB_LSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); #ifdef FIXED_DEBUG_ASSERT @@ -819,7 +813,7 @@ static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opu #define silk_SUB_RSHIFT32(a,b,c) silk_SUB_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; - ret = a - (b >> shift); + ret = silk_SUB32_ovflw(a, (b >> shift)); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (((opus_int64)b) >> shift)) ) { fprintf (stderr, "silk_SUB_RSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); @@ -835,7 +829,7 @@ static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opu static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; - /* the marco definition can't handle a shift of zero */ + /* the macro definition can't handle a shift of zero */ if ( (shift <= 0) || (shift>31) || ((opus_int64)ret != ((opus_int64)a + ((opus_int64)1 << (shift - 1))) >> shift) ) { fprintf (stderr, "silk_RSHIFT_ROUND(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -850,7 +844,7 @@ static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, #define silk_RSHIFT_ROUND64(a,b) silk_RSHIFT_ROUND64_((a), (b), __FILE__, __LINE__) static OPUS_INLINE opus_int64 silk_RSHIFT_ROUND64_(opus_int64 a, opus_int32 shift, char *file, int line){ opus_int64 ret; - /* the marco definition can't handle a shift of zero */ + /* the macro definition can't handle a shift of zero */ if ( (shift <= 0) || (shift>=64) ) { fprintf (stderr, "silk_RSHIFT_ROUND64(%lld, %d) in %s: line %d\n", (long long)a, shift, file, line); diff --git a/code/opus-1.2.1/silk/NLSF2A.c b/code/opus-1.5.2/silk/NLSF2A.c similarity index 99% rename from code/opus-1.2.1/silk/NLSF2A.c rename to code/opus-1.5.2/silk/NLSF2A.c index 116b465b..d5b77306 100644 --- a/code/opus-1.2.1/silk/NLSF2A.c +++ b/code/opus-1.5.2/silk/NLSF2A.c @@ -86,7 +86,7 @@ void silk_NLSF2A( opus_int32 a32_QA1[ SILK_MAX_ORDER_LPC ]; silk_assert( LSF_COS_TAB_SZ_FIX == 128 ); - silk_assert( d==10 || d==16 ); + celt_assert( d==10 || d==16 ); /* convert LSFs to 2*cos(LSF), using piecewise linear curve from table */ ordering = d == 16 ? ordering16 : ordering10; diff --git a/code/opus-1.2.1/silk/NLSF_VQ.c b/code/opus-1.5.2/silk/NLSF_VQ.c similarity index 98% rename from code/opus-1.2.1/silk/NLSF_VQ.c rename to code/opus-1.5.2/silk/NLSF_VQ.c index 452f3dcb..b83182a7 100644 --- a/code/opus-1.2.1/silk/NLSF_VQ.c +++ b/code/opus-1.5.2/silk/NLSF_VQ.c @@ -46,7 +46,7 @@ void silk_NLSF_VQ( const opus_int16 *w_Q9_ptr; const opus_uint8 *cb_Q8_ptr; - silk_assert( ( LPC_order & 1 ) == 0 ); + celt_assert( ( LPC_order & 1 ) == 0 ); /* Loop over codebook */ cb_Q8_ptr = pCB_Q8; diff --git a/code/opus-1.2.1/silk/NLSF_VQ_weights_laroia.c b/code/opus-1.5.2/silk/NLSF_VQ_weights_laroia.c similarity index 98% rename from code/opus-1.2.1/silk/NLSF_VQ_weights_laroia.c rename to code/opus-1.5.2/silk/NLSF_VQ_weights_laroia.c index 04894c59..9873bcde 100644 --- a/code/opus-1.2.1/silk/NLSF_VQ_weights_laroia.c +++ b/code/opus-1.5.2/silk/NLSF_VQ_weights_laroia.c @@ -48,8 +48,8 @@ void silk_NLSF_VQ_weights_laroia( opus_int k; opus_int32 tmp1_int, tmp2_int; - silk_assert( D > 0 ); - silk_assert( ( D & 1 ) == 0 ); + celt_assert( D > 0 ); + celt_assert( ( D & 1 ) == 0 ); /* First value */ tmp1_int = silk_max_int( pNLSF_Q15[ 0 ], 1 ); diff --git a/code/opus-1.2.1/silk/NLSF_decode.c b/code/opus-1.5.2/silk/NLSF_decode.c similarity index 100% rename from code/opus-1.2.1/silk/NLSF_decode.c rename to code/opus-1.5.2/silk/NLSF_decode.c diff --git a/code/opus-1.2.1/silk/NLSF_del_dec_quant.c b/code/opus-1.5.2/silk/NLSF_del_dec_quant.c similarity index 100% rename from code/opus-1.2.1/silk/NLSF_del_dec_quant.c rename to code/opus-1.5.2/silk/NLSF_del_dec_quant.c diff --git a/code/opus-1.2.1/silk/NLSF_encode.c b/code/opus-1.5.2/silk/NLSF_encode.c similarity index 99% rename from code/opus-1.2.1/silk/NLSF_encode.c rename to code/opus-1.5.2/silk/NLSF_encode.c index 268b9a19..01ac7db7 100644 --- a/code/opus-1.2.1/silk/NLSF_encode.c +++ b/code/opus-1.5.2/silk/NLSF_encode.c @@ -60,7 +60,7 @@ opus_int32 silk_NLSF_encode( /* O Returns const opus_int16 *pCB_Wght_Q9; SAVE_STACK; - silk_assert( signalType >= 0 && signalType <= 2 ); + celt_assert( signalType >= 0 && signalType <= 2 ); silk_assert( NLSF_mu_Q20 <= 32767 && NLSF_mu_Q20 >= 0 ); /* NLSF stabilization */ diff --git a/code/opus-1.2.1/silk/NLSF_stabilize.c b/code/opus-1.5.2/silk/NLSF_stabilize.c similarity index 100% rename from code/opus-1.2.1/silk/NLSF_stabilize.c rename to code/opus-1.5.2/silk/NLSF_stabilize.c diff --git a/code/opus-1.2.1/silk/NLSF_unpack.c b/code/opus-1.5.2/silk/NLSF_unpack.c similarity index 100% rename from code/opus-1.2.1/silk/NLSF_unpack.c rename to code/opus-1.5.2/silk/NLSF_unpack.c diff --git a/code/opus-1.2.1/silk/NSQ.c b/code/opus-1.5.2/silk/NSQ.c similarity index 92% rename from code/opus-1.2.1/silk/NSQ.c rename to code/opus-1.5.2/silk/NSQ.c index 617a19fe..1caa829b 100644 --- a/code/opus-1.2.1/silk/NSQ.c +++ b/code/opus-1.5.2/silk/NSQ.c @@ -75,21 +75,21 @@ static OPUS_INLINE void silk_noise_shape_quantizer( void silk_NSQ_c ( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int16 x16[], /* I Input */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) { opus_int k, lag, start_idx, LSF_interpolation_flag; @@ -143,7 +143,7 @@ void silk_NSQ_c if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { /* Rewhiten with new A coefs */ start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; - silk_assert( start_idx > 0 ); + celt_assert( start_idx > 0 ); silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); @@ -173,9 +173,9 @@ void silk_NSQ_c RESTORE_STACK; } -/***********************************/ -/* silk_noise_shape_quantizer */ -/***********************************/ +/******************************/ +/* silk_noise_shape_quantizer */ +/******************************/ #if !defined(OPUS_X86_MAY_HAVE_SSE4_1) static OPUS_INLINE @@ -247,7 +247,7 @@ void silk_noise_shape_quantizer( } /* Noise shape feedback */ - silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ n_AR_Q12 = silk_NSQ_noise_shape_feedback_loop(&NSQ->sDiff_shp_Q14, NSQ->sAR2_Q14, AR_shp_Q13, shapingLPCOrder, arch); n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sLF_AR_shp_Q14, Tilt_Q14 ); @@ -255,20 +255,20 @@ void silk_noise_shape_quantizer( n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); n_LF_Q12 = silk_SMLAWT( n_LF_Q12, NSQ->sLF_AR_shp_Q14, LF_shp_Q14 ); - silk_assert( lag > 0 || signalType != TYPE_VOICED ); + celt_assert( lag > 0 || signalType != TYPE_VOICED ); /* Combine prediction and noise shaping signals */ - tmp1 = silk_SUB32( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ - tmp1 = silk_SUB32( tmp1, n_LF_Q12 ); /* Q12 */ + tmp1 = silk_SUB32_ovflw( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ + tmp1 = silk_SUB32_ovflw( tmp1, n_LF_Q12 ); /* Q12 */ if( lag > 0 ) { /* Symmetric, packed FIR coefficients */ - n_LTP_Q13 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); n_LTP_Q13 = silk_SMLAWT( n_LTP_Q13, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); n_LTP_Q13 = silk_LSHIFT( n_LTP_Q13, 1 ); shp_lag_ptr++; tmp2 = silk_SUB32( LTP_pred_Q13, n_LTP_Q13 ); /* Q13 */ - tmp1 = silk_ADD_LSHIFT32( tmp2, tmp1, 1 ); /* Q13 */ + tmp1 = silk_ADD32_ovflw( tmp2, silk_LSHIFT32( tmp1, 1 ) ); /* Q13 */ tmp1 = silk_RSHIFT_ROUND( tmp1, 3 ); /* Q10 */ } else { tmp1 = silk_RSHIFT_ROUND( tmp1, 2 ); /* Q10 */ @@ -340,7 +340,7 @@ void silk_noise_shape_quantizer( /* Add predictions */ LPC_exc_Q14 = silk_ADD_LSHIFT32( exc_Q14, LTP_pred_Q13, 1 ); - xq_Q14 = silk_ADD_LSHIFT32( LPC_exc_Q14, LPC_pred_Q10, 4 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, silk_LSHIFT32( LPC_pred_Q10, 4 ) ); /* Scale XQ back to normal level before saving */ xq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( xq_Q14, Gain_Q10 ), 8 ) ); @@ -349,10 +349,10 @@ void silk_noise_shape_quantizer( psLPC_Q14++; *psLPC_Q14 = xq_Q14; NSQ->sDiff_shp_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_sc_Q10[ i ], 4 ); - sLF_AR_shp_Q14 = silk_SUB_LSHIFT32( NSQ->sDiff_shp_Q14, n_AR_Q12, 2 ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( NSQ->sDiff_shp_Q14, silk_LSHIFT32( n_AR_Q12, 2 ) ); NSQ->sLF_AR_shp_Q14 = sLF_AR_shp_Q14; - NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB_LSHIFT32( sLF_AR_shp_Q14, n_LF_Q12, 2 ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB32_ovflw(sLF_AR_shp_Q14, silk_LSHIFT32(n_LF_Q12, 2)); sLTP_Q15[ NSQ->sLTP_buf_idx ] = silk_LSHIFT( LPC_exc_Q14, 1 ); NSQ->sLTP_shp_buf_idx++; NSQ->sLTP_buf_idx++; diff --git a/code/opus-1.2.1/silk/NSQ.h b/code/opus-1.5.2/silk/NSQ.h similarity index 100% rename from code/opus-1.2.1/silk/NSQ.h rename to code/opus-1.5.2/silk/NSQ.h diff --git a/code/opus-1.2.1/silk/NSQ_del_dec.c b/code/opus-1.5.2/silk/NSQ_del_dec.c similarity index 93% rename from code/opus-1.2.1/silk/NSQ_del_dec.c rename to code/opus-1.5.2/silk/NSQ_del_dec.c index 1cd29d96..e8dadf15 100644 --- a/code/opus-1.2.1/silk/NSQ_del_dec.c +++ b/code/opus-1.5.2/silk/NSQ_del_dec.c @@ -115,21 +115,21 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( ); void silk_NSQ_del_dec_c( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int16 x16[], /* I Input */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) { opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; @@ -250,7 +250,7 @@ void silk_NSQ_del_dec_c( /* Rewhiten with new A coefs */ start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; - silk_assert( start_idx > 0 ); + celt_assert( start_idx > 0 ); silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); @@ -361,7 +361,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( NSQ_sample_struct *psSS; SAVE_STACK; - silk_assert( nStatesDelayedDecision > 0 ); + celt_assert( nStatesDelayedDecision > 0 ); ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; @@ -394,8 +394,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( /* Long-term shaping */ if( lag > 0 ) { /* Symmetric, packed FIR coefficients */ - n_LTP_Q14 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); - n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ shp_lag_ptr++; } else { @@ -419,22 +419,22 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ /* Noise shape feedback */ - silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ /* Output of lowpass section */ tmp2 = silk_SMLAWB( psDD->Diff_Q14, psDD->sAR2_Q14[ 0 ], warping_Q16 ); /* Output of allpass section */ - tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], psDD->sAR2_Q14[ 1 ] - tmp2, warping_Q16 ); + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ 1 ], tmp2), warping_Q16 ); psDD->sAR2_Q14[ 0 ] = tmp2; n_AR_Q14 = silk_RSHIFT( shapingLPCOrder, 1 ); n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ 0 ] ); /* Loop over allpass sections */ for( j = 2; j < shapingLPCOrder; j += 2 ) { /* Output of allpass section */ - tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], psDD->sAR2_Q14[ j + 0 ] - tmp1, warping_Q16 ); + tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 0 ], tmp1), warping_Q16 ); psDD->sAR2_Q14[ j - 1 ] = tmp1; n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ j - 1 ] ); /* Output of allpass section */ - tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], psDD->sAR2_Q14[ j + 1 ] - tmp2, warping_Q16 ); + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 1 ], tmp2), warping_Q16 ); psDD->sAR2_Q14[ j + 0 ] = tmp2; n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ j ] ); } @@ -451,9 +451,9 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( /* Input minus prediction plus noise feedback */ /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ - tmp1 = silk_ADD32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ - tmp2 = silk_ADD32( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ - tmp1 = silk_SUB32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_ADD_SAT32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32_ovflw( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB_SAT32( tmp2, tmp1 ); /* Q13 */ tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ @@ -530,12 +530,12 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( /* Add predictions */ LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); - xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); /* Update states */ - psSS[ 0 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); - sLF_AR_shp_Q14 = silk_SUB32( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); - psSS[ 0 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; psSS[ 0 ].xq_Q14 = xq_Q14; @@ -550,12 +550,12 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( /* Add predictions */ LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); - xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); /* Update states */ - psSS[ 1 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); - sLF_AR_shp_Q14 = silk_SUB32( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); - psSS[ 1 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; psSS[ 1 ].xq_Q14 = xq_Q14; diff --git a/code/opus-1.2.1/silk/PLC.c b/code/opus-1.5.2/silk/PLC.c similarity index 90% rename from code/opus-1.2.1/silk/PLC.c rename to code/opus-1.5.2/silk/PLC.c index a3e55ea7..b35bf750 100644 --- a/code/opus-1.2.1/silk/PLC.c +++ b/code/opus-1.5.2/silk/PLC.c @@ -33,6 +33,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "stack_alloc.h" #include "PLC.h" +#ifdef ENABLE_DEEP_PLC +#include "lpcnet.h" +#endif + #define NB_ATT 2 static const opus_int16 HARM_ATT_Q15[NB_ATT] = { 32440, 31130 }; /* 0.99, 0.95 */ static const opus_int16 PLC_RAND_ATTENUATE_V_Q15[NB_ATT] = { 31130, 26214 }; /* 0.95, 0.8 */ @@ -47,6 +51,9 @@ static OPUS_INLINE void silk_PLC_conceal( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[], /* O LPC residual signal */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ); @@ -67,6 +74,9 @@ void silk_PLC( silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[], /* I/O signal */ opus_int lost, /* I Loss flag */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ) { @@ -80,7 +90,11 @@ void silk_PLC( /****************************/ /* Generate Signal */ /****************************/ - silk_PLC_conceal( psDec, psDecCtrl, frame, arch ); + silk_PLC_conceal( psDec, psDecCtrl, frame, +#ifdef ENABLE_DEEP_PLC + lpcnet, +#endif + arch ); psDec->lossCnt++; } else { @@ -88,6 +102,14 @@ void silk_PLC( /* Update state */ /****************************/ silk_PLC_update( psDec, psDecCtrl ); +#ifdef ENABLE_DEEP_PLC + if ( lpcnet != NULL && psDec->sPLC.fs_kHz == 16 ) { + int k; + for( k = 0; k < psDec->nb_subfr; k += 2 ) { + lpcnet_plc_update( lpcnet, frame + k * psDec->subfr_length ); + } + } +#endif } } @@ -195,6 +217,9 @@ static OPUS_INLINE void silk_PLC_conceal( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[], /* O LPC residual signal */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ) { @@ -291,7 +316,7 @@ static OPUS_INLINE void silk_PLC_conceal( /* Rewhiten LTP state */ idx = psDec->ltp_mem_length - lag - psDec->LPC_order - LTP_ORDER / 2; - silk_assert( idx > 0 ); + celt_assert( idx > 0 ); silk_LPC_analysis_filter( &sLTP[ idx ], &psDec->outBuf[ idx ], A_Q12, psDec->ltp_mem_length - idx, psDec->LPC_order, arch ); /* Scale LTP state */ inv_gain_Q30 = silk_INVERSE32_varQ( psPLC->prevGain_Q16[ 1 ], 46 ); @@ -328,10 +353,8 @@ static OPUS_INLINE void silk_PLC_conceal( for( j = 0; j < LTP_ORDER; j++ ) { B_Q14[ j ] = silk_RSHIFT( silk_SMULBB( harm_Gain_Q15, B_Q14[ j ] ), 15 ); } - if ( psDec->indices.signalType != TYPE_NO_VOICE_ACTIVITY ) { - /* Gradually reduce excitation gain */ - rand_scale_Q14 = silk_RSHIFT( silk_SMULBB( rand_scale_Q14, rand_Gain_Q15 ), 15 ); - } + /* Gradually reduce excitation gain */ + rand_scale_Q14 = silk_RSHIFT( silk_SMULBB( rand_scale_Q14, rand_Gain_Q15 ), 15 ); /* Slowly increase pitch lag */ psPLC->pitchL_Q8 = silk_SMLAWB( psPLC->pitchL_Q8, psPLC->pitchL_Q8, PITCH_DRIFT_FAC_Q16 ); @@ -347,7 +370,7 @@ static OPUS_INLINE void silk_PLC_conceal( /* Copy LPC state */ silk_memcpy( sLPC_Q14_ptr, psDec->sLPC_Q14_buf, MAX_LPC_ORDER * sizeof( opus_int32 ) ); - silk_assert( psDec->LPC_order >= 10 ); /* check that unrolling works */ + celt_assert( psDec->LPC_order >= 10 ); /* check that unrolling works */ for( i = 0; i < psDec->frame_length; i++ ) { /* partly unrolled */ /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ @@ -373,6 +396,24 @@ static OPUS_INLINE void silk_PLC_conceal( /* Scale with Gain */ frame[ i ] = (opus_int16)silk_SAT16( silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], prevGain_Q10[ 1 ] ), 8 ) ) ); } +#ifdef ENABLE_DEEP_PLC + if ( lpcnet != NULL && lpcnet->loaded && psDec->sPLC.fs_kHz == 16 ) { + int run_deep_plc = psDec->sPLC.enable_deep_plc || lpcnet->fec_fill_pos != 0; + if( run_deep_plc ) { + for( k = 0; k < psDec->nb_subfr; k += 2 ) { + lpcnet_plc_conceal( lpcnet, frame + k * psDec->subfr_length ); + } + /* We *should* be able to copy only from psDec->frame_length-MAX_LPC_ORDER, i.e. the last MAX_LPC_ORDER samples. */ + for( i = 0; i < psDec->frame_length; i++ ) { + sLPC_Q14_ptr[ MAX_LPC_ORDER + i ] = (int)floor(.5 + frame[ i ] * (float)(1 << 24) / prevGain_Q10[ 1 ] ); + } + } else { + for( k = 0; k < psDec->nb_subfr; k += 2 ) { + lpcnet_plc_update( lpcnet, frame + k * psDec->subfr_length ); + } + } + } +#endif /* Save LPC state */ silk_memcpy( psDec->sLPC_Q14_buf, &sLPC_Q14_ptr[ psDec->frame_length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); @@ -433,12 +474,16 @@ void silk_PLC_glue_frames( slope_Q16 = silk_DIV32_16( ( (opus_int32)1 << 16 ) - gain_Q16, length ); /* Make slope 4x steeper to avoid missing onsets after DTX */ slope_Q16 = silk_LSHIFT( slope_Q16, 2 ); - - for( i = 0; i < length; i++ ) { - frame[ i ] = silk_SMULWB( gain_Q16, frame[ i ] ); - gain_Q16 += slope_Q16; - if( gain_Q16 > (opus_int32)1 << 16 ) { - break; +#ifdef ENABLE_DEEP_PLC + if ( psDec->sPLC.fs_kHz != 16 ) +#endif + { + for( i = 0; i < length; i++ ) { + frame[ i ] = silk_SMULWB( gain_Q16, frame[ i ] ); + gain_Q16 += slope_Q16; + if( gain_Q16 > (opus_int32)1 << 16 ) { + break; + } } } } diff --git a/code/opus-1.2.1/silk/PLC.h b/code/opus-1.5.2/silk/PLC.h similarity index 97% rename from code/opus-1.2.1/silk/PLC.h rename to code/opus-1.5.2/silk/PLC.h index 6438f516..1bebb786 100644 --- a/code/opus-1.2.1/silk/PLC.h +++ b/code/opus-1.5.2/silk/PLC.h @@ -49,6 +49,9 @@ void silk_PLC( silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[], /* I/O signal */ opus_int lost, /* I Loss flag */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ); diff --git a/code/opus-1.2.1/silk/SigProc_FIX.h b/code/opus-1.5.2/silk/SigProc_FIX.h similarity index 99% rename from code/opus-1.2.1/silk/SigProc_FIX.h rename to code/opus-1.5.2/silk/SigProc_FIX.h index f9ae3263..fbdfa82e 100644 --- a/code/opus-1.2.1/silk/SigProc_FIX.h +++ b/code/opus-1.5.2/silk/SigProc_FIX.h @@ -381,7 +381,7 @@ opus_int32 silk_inner_prod_aligned_scale( const opus_int len /* I vector lengths */ ); -opus_int64 silk_inner_prod16_aligned_64_c( +opus_int64 silk_inner_prod16_c( const opus_int16 *inVec1, /* I input vector 1 */ const opus_int16 *inVec2, /* I input vector 2 */ const opus_int len /* I vector lengths */ @@ -609,12 +609,14 @@ static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b) /* the following seems faster on x86 */ #define silk_SMMUL(a32, b32) (opus_int32)silk_RSHIFT64(silk_SMULL((a32), (b32)), 32) -#if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +#if !defined(OVERRIDE_silk_burg_modified) #define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ ((void)(arch), silk_burg_modified_c(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) +#endif -#define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \ - ((void)(arch),silk_inner_prod16_aligned_64_c(inVec1, inVec2, len)) +#if !defined(OVERRIDE_silk_inner_prod16) +#define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((void)(arch),silk_inner_prod16_c(inVec1, inVec2, len)) #endif #include "Inlines.h" diff --git a/code/opus-1.2.1/silk/VAD.c b/code/opus-1.5.2/silk/VAD.c similarity index 96% rename from code/opus-1.2.1/silk/VAD.c rename to code/opus-1.5.2/silk/VAD.c index 0a782af2..d0cda521 100644 --- a/code/opus-1.2.1/silk/VAD.c +++ b/code/opus-1.5.2/silk/VAD.c @@ -101,9 +101,9 @@ opus_int silk_VAD_GetSA_Q8_c( /* O Return v /* Safety checks */ silk_assert( VAD_N_BANDS == 4 ); - silk_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); - silk_assert( psEncC->frame_length <= 512 ); - silk_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); + celt_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); + celt_assert( psEncC->frame_length <= 512 ); + celt_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); /***********************/ /* Filter and Decimate */ @@ -252,15 +252,14 @@ opus_int silk_VAD_GetSA_Q8_c( /* O Return v speech_nrg += ( b + 1 ) * silk_RSHIFT( Xnrg[ b ] - psSilk_VAD->NL[ b ], 4 ); } + if( psEncC->frame_length == 20 * psEncC->fs_kHz ) { + speech_nrg = silk_RSHIFT32( speech_nrg, 1 ); + } /* Power scaling */ if( speech_nrg <= 0 ) { SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); - } else if( speech_nrg < 32768 ) { - if( psEncC->frame_length == 10 * psEncC->fs_kHz ) { - speech_nrg = silk_LSHIFT_SAT32( speech_nrg, 16 ); - } else { - speech_nrg = silk_LSHIFT_SAT32( speech_nrg, 15 ); - } + } else if( speech_nrg < 16384 ) { + speech_nrg = silk_LSHIFT32( speech_nrg, 16 ); /* square-root */ speech_nrg = silk_SQRT_APPROX( speech_nrg ); @@ -313,6 +312,8 @@ void silk_VAD_GetNoiseLevels( /* Initially faster smoothing */ if( psSilk_VAD->counter < 1000 ) { /* 1000 = 20 sec */ min_coef = silk_DIV32_16( silk_int16_MAX, silk_RSHIFT( psSilk_VAD->counter, 4 ) + 1 ); + /* Increment frame counter */ + psSilk_VAD->counter++; } else { min_coef = 0; } @@ -356,7 +357,4 @@ void silk_VAD_GetNoiseLevels( /* Store as part of state */ psSilk_VAD->NL[ k ] = nl; } - - /* Increment frame counter */ - psSilk_VAD->counter++; } diff --git a/code/opus-1.2.1/silk/VQ_WMat_EC.c b/code/opus-1.5.2/silk/VQ_WMat_EC.c similarity index 98% rename from code/opus-1.2.1/silk/VQ_WMat_EC.c rename to code/opus-1.5.2/silk/VQ_WMat_EC.c index 0f3d545c..245a7e4b 100644 --- a/code/opus-1.2.1/silk/VQ_WMat_EC.c +++ b/code/opus-1.5.2/silk/VQ_WMat_EC.c @@ -64,7 +64,7 @@ void silk_VQ_WMat_EC_c( *rate_dist_Q8 = silk_int32_MAX; *res_nrg_Q15 = silk_int32_MAX; cb_row_Q7 = cb_Q7; - /* In things go really bad, at least *ind is set to something safe. */ + /* If things go really bad, at least *ind is set to something safe. */ *ind = 0; for( k = 0; k < L; k++ ) { opus_int32 penalty; @@ -115,7 +115,7 @@ void silk_VQ_WMat_EC_c( if( sum1_Q15 >= 0 ) { /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */ bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) ); - /* In the following line we reduce the codelength component by half ("-1"); seems to slghtly improve quality */ + /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */ bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 ); if( bits_tot_Q8 <= *rate_dist_Q8 ) { *rate_dist_Q8 = bits_tot_Q8; diff --git a/code/opus-1.2.1/silk/ana_filt_bank_1.c b/code/opus-1.5.2/silk/ana_filt_bank_1.c similarity index 100% rename from code/opus-1.2.1/silk/ana_filt_bank_1.c rename to code/opus-1.5.2/silk/ana_filt_bank_1.c diff --git a/code/opus-1.2.1/silk/arm/LPC_inv_pred_gain_arm.h b/code/opus-1.5.2/silk/arm/LPC_inv_pred_gain_arm.h similarity index 100% rename from code/opus-1.2.1/silk/arm/LPC_inv_pred_gain_arm.h rename to code/opus-1.5.2/silk/arm/LPC_inv_pred_gain_arm.h diff --git a/code/opus-1.2.1/silk/arm/LPC_inv_pred_gain_neon_intr.c b/code/opus-1.5.2/silk/arm/LPC_inv_pred_gain_neon_intr.c similarity index 93% rename from code/opus-1.2.1/silk/arm/LPC_inv_pred_gain_neon_intr.c rename to code/opus-1.5.2/silk/arm/LPC_inv_pred_gain_neon_intr.c index 27142f34..726e6667 100644 --- a/code/opus-1.2.1/silk/arm/LPC_inv_pred_gain_neon_intr.c +++ b/code/opus-1.5.2/silk/arm/LPC_inv_pred_gain_neon_intr.c @@ -210,28 +210,32 @@ opus_int32 silk_LPC_inverse_pred_gain_neon( /* O Returns inverse predi /* Increase Q domain of the AR coefficients */ t0_s16x8 = vld1q_s16( A_Q12 + 0 ); t1_s16x8 = vld1q_s16( A_Q12 + 8 ); - t2_s16x8 = vld1q_s16( A_Q12 + 16 ); + if ( order > 16 ) { + t2_s16x8 = vld1q_s16( A_Q12 + 16 ); + } t0_s32x4 = vpaddlq_s16( t0_s16x8 ); switch( order - leftover ) { case 24: t0_s32x4 = vpadalq_s16( t0_s32x4, t2_s16x8 ); - /* Intend to fall through */ + vst1q_s32( Atmp_QA + 16, vshll_n_s16( vget_low_s16 ( t2_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 20, vshll_n_s16( vget_high_s16( t2_s16x8 ), QA - 12 ) ); + /* FALLTHROUGH */ case 16: t0_s32x4 = vpadalq_s16( t0_s32x4, t1_s16x8 ); - vst1q_s32( Atmp_QA + 16, vshll_n_s16( vget_low_s16 ( t2_s16x8 ), QA - 12 ) ); - vst1q_s32( Atmp_QA + 20, vshll_n_s16( vget_high_s16( t2_s16x8 ), QA - 12 ) ); - /* Intend to fall through */ + vst1q_s32( Atmp_QA + 8, vshll_n_s16( vget_low_s16 ( t1_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 12, vshll_n_s16( vget_high_s16( t1_s16x8 ), QA - 12 ) ); + /* FALLTHROUGH */ case 8: { const int32x2_t t_s32x2 = vpadd_s32( vget_low_s32( t0_s32x4 ), vget_high_s32( t0_s32x4 ) ); const int64x1_t t_s64x1 = vpaddl_s32( t_s32x2 ); DC_resp = vget_lane_s32( vreinterpret_s32_s64( t_s64x1 ), 0 ); - vst1q_s32( Atmp_QA + 8, vshll_n_s16( vget_low_s16 ( t1_s16x8 ), QA - 12 ) ); - vst1q_s32( Atmp_QA + 12, vshll_n_s16( vget_high_s16( t1_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 0, vshll_n_s16( vget_low_s16 ( t0_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 4, vshll_n_s16( vget_high_s16( t0_s16x8 ), QA - 12 ) ); } break; @@ -246,17 +250,23 @@ opus_int32 silk_LPC_inverse_pred_gain_neon( /* O Returns inverse predi case 6: DC_resp += (opus_int32)A_Q12[ 5 ]; DC_resp += (opus_int32)A_Q12[ 4 ]; - /* Intend to fall through */ + Atmp_QA[ order - leftover + 5 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 5 ], QA - 12 ); + Atmp_QA[ order - leftover + 4 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 4 ], QA - 12 ); + /* FALLTHROUGH */ case 4: DC_resp += (opus_int32)A_Q12[ 3 ]; DC_resp += (opus_int32)A_Q12[ 2 ]; - /* Intend to fall through */ + Atmp_QA[ order - leftover + 3 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 3 ], QA - 12 ); + Atmp_QA[ order - leftover + 2 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 2 ], QA - 12 ); + /* FALLTHROUGH */ case 2: DC_resp += (opus_int32)A_Q12[ 1 ]; DC_resp += (opus_int32)A_Q12[ 0 ]; - /* Intend to fall through */ + Atmp_QA[ order - leftover + 1 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 1 ], QA - 12 ); + Atmp_QA[ order - leftover + 0 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 0 ], QA - 12 ); + /* FALLTHROUGH */ default: break; @@ -266,8 +276,6 @@ opus_int32 silk_LPC_inverse_pred_gain_neon( /* O Returns inverse predi if( DC_resp >= 4096 ) { invGain_Q30 = 0; } else { - vst1q_s32( Atmp_QA + 0, vshll_n_s16( vget_low_s16 ( t0_s16x8 ), QA - 12 ) ); - vst1q_s32( Atmp_QA + 4, vshll_n_s16( vget_high_s16( t0_s16x8 ), QA - 12 ) ); invGain_Q30 = LPC_inverse_pred_gain_QA_neon( Atmp_QA, order ); } } diff --git a/code/opus-1.2.1/silk/arm/NSQ_del_dec_arm.h b/code/opus-1.5.2/silk/arm/NSQ_del_dec_arm.h similarity index 98% rename from code/opus-1.2.1/silk/arm/NSQ_del_dec_arm.h rename to code/opus-1.5.2/silk/arm/NSQ_del_dec_arm.h index 9e76e169..0c4fcfcc 100644 --- a/code/opus-1.2.1/silk/arm/NSQ_del_dec_arm.h +++ b/code/opus-1.5.2/silk/arm/NSQ_del_dec_arm.h @@ -34,7 +34,7 @@ POSSIBILITY OF SUCH DAMAGE. void silk_NSQ_del_dec_neon( const silk_encoder_state *psEncC, silk_nsq_state *NSQ, SideInfoIndices *psIndices, const opus_int16 x16[], opus_int8 pulses[], - const opus_int16 PredCoef_Q12[2 * MAX_LPC_ORDER], + const opus_int16 *PredCoef_Q12, const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], @@ -65,7 +65,7 @@ void silk_NSQ_del_dec_neon( extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( const silk_encoder_state *psEncC, silk_nsq_state *NSQ, SideInfoIndices *psIndices, const opus_int16 x16[], opus_int8 pulses[], - const opus_int16 PredCoef_Q12[2 * MAX_LPC_ORDER], + const opus_int16 *PredCoef_Q12, const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], diff --git a/code/opus-1.2.1/silk/arm/NSQ_del_dec_neon_intr.c b/code/opus-1.5.2/silk/arm/NSQ_del_dec_neon_intr.c similarity index 97% rename from code/opus-1.2.1/silk/arm/NSQ_del_dec_neon_intr.c rename to code/opus-1.5.2/silk/arm/NSQ_del_dec_neon_intr.c index 212410f3..668dde6d 100644 --- a/code/opus-1.2.1/silk/arm/NSQ_del_dec_neon_intr.c +++ b/code/opus-1.5.2/silk/arm/NSQ_del_dec_neon_intr.c @@ -35,6 +35,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" #include "stack_alloc.h" +#include "os_support.h" /* NEON intrinsics optimization now can only parallelize up to 4 delay decision states. */ /* If there are more states, C function is called, and this optimization must be expanded. */ @@ -220,7 +221,7 @@ void silk_NSQ_del_dec_neon( SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int16 x16[], /* I Input */ opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ @@ -279,6 +280,7 @@ void silk_NSQ_del_dec_neon( /* Initialize delayed decision states */ ALLOC( psDelDec, 1, NSQ_del_decs_struct ); + OPUS_CLEAR(psDelDec, 1); /* Only RandState and RD_Q10 need to be initialized to 0. */ silk_memset( psDelDec->RandState, 0, sizeof( psDelDec->RandState ) ); vst1q_s32( psDelDec->RD_Q10, vdupq_n_s32( 0 ) ); @@ -587,6 +589,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( silk_assert( nStatesDelayedDecision > 0 ); silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ ALLOC( psSampleState, 2, NSQ_samples_struct ); + OPUS_CLEAR(psSampleState, 2); shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; @@ -711,23 +714,26 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( const int rdo_offset = Lambda_Q10/2 - 512; const uint16x4_t greaterThanRdo = vcgt_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ); const uint16x4_t lessThanMinusRdo = vclt_s16( q1_Q10_s16x4, vdup_n_s16( -rdo_offset ) ); + int16x4_t signed_offset = vbsl_s16( greaterThanRdo, vdup_n_s16( -rdo_offset ), vdup_n_s16( 0 ) ); + signed_offset = vbsl_s16( lessThanMinusRdo, vdup_n_s16( rdo_offset ), signed_offset ); /* If Lambda_Q10 > 32767, then q1_Q0, q1_Q10 and q2_Q10 must change to 32-bit. */ silk_assert( Lambda_Q10 <= 32767 ); q1_Q0_s16x4 = vreinterpret_s16_u16( vclt_s16( q1_Q10_s16x4, vdup_n_s16( 0 ) ) ); - q1_Q0_s16x4 = vbsl_s16( greaterThanRdo, vsub_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ), q1_Q0_s16x4 ); - q1_Q0_s16x4 = vbsl_s16( lessThanMinusRdo, vadd_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ), q1_Q0_s16x4 ); + q1_Q0_s16x4 = vbsl_s16(vorr_u16(greaterThanRdo, lessThanMinusRdo), vadd_s16( q1_Q10_s16x4 , signed_offset), q1_Q0_s16x4); q1_Q0_s16x4 = vshr_n_s16( q1_Q0_s16x4, 10 ); } { const uint16x4_t equal0_u16x4 = vceq_s16( q1_Q0_s16x4, vdup_n_s16( 0 ) ); const uint16x4_t equalMinus1_u16x4 = vceq_s16( q1_Q0_s16x4, vdup_n_s16( -1 ) ); const uint16x4_t lessThanMinus1_u16x4 = vclt_s16( q1_Q0_s16x4, vdup_n_s16( -1 ) ); - int16x4_t tmp1_s16x4, tmp2_s16x4; + int16x4_t tmp1_s16x4, tmp2_s16x4, tmp_summand_s16x4; q1_Q10_s16x4 = vshl_n_s16( q1_Q0_s16x4, 10 ); - tmp1_s16x4 = vadd_s16( q1_Q10_s16x4, vdup_n_s16( offset_Q10 - QUANT_LEVEL_ADJUST_Q10 ) ); - q1_Q10_s16x4 = vadd_s16( q1_Q10_s16x4, vdup_n_s16( offset_Q10 + QUANT_LEVEL_ADJUST_Q10 ) ); + tmp_summand_s16x4 = vand_s16( vreinterpret_s16_u16(vcge_s16(q1_Q0_s16x4, vdup_n_s16(0))), vdup_n_s16( offset_Q10 - QUANT_LEVEL_ADJUST_Q10 ) ); + tmp1_s16x4 = vadd_s16( q1_Q10_s16x4, tmp_summand_s16x4 ); + tmp_summand_s16x4 = vbsl_s16( lessThanMinus1_u16x4, vdup_n_s16( offset_Q10 + QUANT_LEVEL_ADJUST_Q10 ), vdup_n_s16(0) ); + q1_Q10_s16x4 = vadd_s16( q1_Q10_s16x4, tmp_summand_s16x4); q1_Q10_s16x4 = vbsl_s16( lessThanMinus1_u16x4, q1_Q10_s16x4, tmp1_s16x4 ); q1_Q10_s16x4 = vbsl_s16( equal0_u16x4, vdup_n_s16( offset_Q10 ), q1_Q10_s16x4 ); q1_Q10_s16x4 = vbsl_s16( equalMinus1_u16x4, vdup_n_s16( offset_Q10 - ( 1024 - QUANT_LEVEL_ADJUST_Q10 ) ), q1_Q10_s16x4 ); @@ -818,6 +824,13 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( } } + /* clear unused part of RD_Q10 to avoid overflows */ + if( nStatesDelayedDecision < NEON_MAX_DEL_DEC_STATES ) + { + OPUS_CLEAR(psSampleState[0].RD_Q10 + nStatesDelayedDecision, NEON_MAX_DEL_DEC_STATES - nStatesDelayedDecision); + OPUS_CLEAR(psSampleState[1].RD_Q10 + nStatesDelayedDecision, NEON_MAX_DEL_DEC_STATES - nStatesDelayedDecision); + } + /* Increase RD values of expired states */ { uint32x4_t t_u32x4; @@ -896,7 +909,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( vst1q_s32( psDelDec->Pred_Q15[ *smpl_buf_idx ], vshlq_n_s32( vld1q_s32( psSampleState[ 0 ].LPC_exc_Q14 ), 1 ) ); vst1q_s32( psDelDec->Shape_Q14[ *smpl_buf_idx ], vld1q_s32( psSampleState[ 0 ].sLTP_shp_Q14 ) ); tmp1_s32x4 = vrshrq_n_s32( tmp1_s32x4, 10 ); - tmp1_s32x4 = vaddq_s32( vld1q_s32( psDelDec->Seed ), tmp1_s32x4 ); + tmp1_s32x4 = vreinterpretq_s32_u32( vaddq_u32( vreinterpretq_u32_s32( + vld1q_s32( psDelDec->Seed ) ), vreinterpretq_u32_s32( tmp1_s32x4 ) ) ); vst1q_s32( psDelDec->Seed, tmp1_s32x4 ); vst1q_s32( psDelDec->RandState[ *smpl_buf_idx ], tmp1_s32x4 ); vst1q_s32( psDelDec->RD_Q10, vld1q_s32( psSampleState[ 0 ].RD_Q10 ) ); diff --git a/code/opus-1.2.1/silk/arm/NSQ_neon.c b/code/opus-1.5.2/silk/arm/NSQ_neon.c similarity index 100% rename from code/opus-1.2.1/silk/arm/NSQ_neon.c rename to code/opus-1.5.2/silk/arm/NSQ_neon.c diff --git a/code/opus-1.2.1/silk/arm/NSQ_neon.h b/code/opus-1.5.2/silk/arm/NSQ_neon.h similarity index 97% rename from code/opus-1.2.1/silk/arm/NSQ_neon.h rename to code/opus-1.5.2/silk/arm/NSQ_neon.h index b31d9442..f03d8ddd 100644 --- a/code/opus-1.2.1/silk/arm/NSQ_neon.h +++ b/code/opus-1.5.2/silk/arm/NSQ_neon.h @@ -73,7 +73,7 @@ static OPUS_INLINE void silk_short_prediction_create_arch_coef_neon(opus_int32 * #elif defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_MAY_HAVE_NEON_INTR) #define silk_short_prediction_create_arch_coef(out, in, order) \ - do { if (arch == OPUS_ARCH_ARM_NEON) { silk_short_prediction_create_arch_coef_neon(out, in, order); } } while (0) + do { if (arch >= OPUS_ARCH_ARM_NEON) { silk_short_prediction_create_arch_coef_neon(out, in, order); } } while (0) #endif @@ -95,7 +95,7 @@ opus_int32 silk_NSQ_noise_shape_feedback_loop_neon(const opus_int32 *data0, opus (coef vs. coefRev) so can't use the usual IMPL table implementation */ #undef silk_noise_shape_quantizer_short_prediction #define silk_noise_shape_quantizer_short_prediction(in, coef, coefRev, order, arch) \ - (arch == OPUS_ARCH_ARM_NEON ? \ + (arch >= OPUS_ARCH_ARM_NEON ? \ silk_noise_shape_quantizer_short_prediction_neon(in, coefRev, order) : \ silk_noise_shape_quantizer_short_prediction_c(in, coef, order)) diff --git a/code/opus-1.2.1/silk/arm/SigProc_FIX_armv4.h b/code/opus-1.5.2/silk/arm/SigProc_FIX_armv4.h similarity index 100% rename from code/opus-1.2.1/silk/arm/SigProc_FIX_armv4.h rename to code/opus-1.5.2/silk/arm/SigProc_FIX_armv4.h diff --git a/code/opus-1.2.1/silk/arm/SigProc_FIX_armv5e.h b/code/opus-1.5.2/silk/arm/SigProc_FIX_armv5e.h similarity index 100% rename from code/opus-1.2.1/silk/arm/SigProc_FIX_armv5e.h rename to code/opus-1.5.2/silk/arm/SigProc_FIX_armv5e.h diff --git a/code/opus-1.2.1/silk/arm/arm_silk_map.c b/code/opus-1.5.2/silk/arm/arm_silk_map.c similarity index 95% rename from code/opus-1.2.1/silk/arm/arm_silk_map.c rename to code/opus-1.5.2/silk/arm/arm_silk_map.c index 0b9bfec2..a91f79b5 100644 --- a/code/opus-1.2.1/silk/arm/arm_silk_map.c +++ b/code/opus-1.5.2/silk/arm/arm_silk_map.c @@ -49,6 +49,7 @@ void (*const SILK_BIQUAD_ALT_STRIDE2_IMPL[OPUS_ARCHMASK + 1])( silk_biquad_alt_stride2_c, /* EDSP */ silk_biquad_alt_stride2_c, /* Media */ silk_biquad_alt_stride2_neon, /* Neon */ + silk_biquad_alt_stride2_neon, /* dotprod */ }; opus_int32 (*const SILK_LPC_INVERSE_PRED_GAIN_IMPL[OPUS_ARCHMASK + 1])( /* O Returns inverse prediction gain in energy domain, Q30 */ @@ -59,6 +60,7 @@ opus_int32 (*const SILK_LPC_INVERSE_PRED_GAIN_IMPL[OPUS_ARCHMASK + 1])( /* O R silk_LPC_inverse_pred_gain_c, /* EDSP */ silk_LPC_inverse_pred_gain_c, /* Media */ silk_LPC_inverse_pred_gain_neon, /* Neon */ + silk_LPC_inverse_pred_gain_neon, /* dotprod */ }; void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( @@ -67,7 +69,7 @@ void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int16 x16[], /* I Input */ opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ @@ -82,6 +84,7 @@ void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( silk_NSQ_del_dec_c, /* EDSP */ silk_NSQ_del_dec_c, /* Media */ silk_NSQ_del_dec_neon, /* Neon */ + silk_NSQ_del_dec_neon, /* dotprod */ }; /*There is no table for silk_noise_shape_quantizer_short_prediction because the @@ -97,6 +100,7 @@ opus_int32 silk_NSQ_noise_shape_feedback_loop_c, /* EDSP */ silk_NSQ_noise_shape_feedback_loop_c, /* Media */ silk_NSQ_noise_shape_feedback_loop_neon, /* NEON */ + silk_NSQ_noise_shape_feedback_loop_neon, /* dotprod */ }; # endif @@ -116,6 +120,7 @@ void (*const SILK_WARPED_AUTOCORRELATION_FIX_IMPL[OPUS_ARCHMASK + 1])( silk_warped_autocorrelation_FIX_c, /* EDSP */ silk_warped_autocorrelation_FIX_c, /* Media */ silk_warped_autocorrelation_FIX_neon, /* Neon */ + silk_warped_autocorrelation_FIX_neon, /* dotprod */ }; # endif diff --git a/code/opus-1.2.1/silk/arm/biquad_alt_arm.h b/code/opus-1.5.2/silk/arm/biquad_alt_arm.h similarity index 100% rename from code/opus-1.2.1/silk/arm/biquad_alt_arm.h rename to code/opus-1.5.2/silk/arm/biquad_alt_arm.h diff --git a/code/opus-1.2.1/silk/arm/biquad_alt_neon_intr.c b/code/opus-1.5.2/silk/arm/biquad_alt_neon_intr.c similarity index 100% rename from code/opus-1.2.1/silk/arm/biquad_alt_neon_intr.c rename to code/opus-1.5.2/silk/arm/biquad_alt_neon_intr.c diff --git a/code/opus-1.2.1/silk/arm/macros_arm64.h b/code/opus-1.5.2/silk/arm/macros_arm64.h similarity index 100% rename from code/opus-1.2.1/silk/arm/macros_arm64.h rename to code/opus-1.5.2/silk/arm/macros_arm64.h diff --git a/code/opus-1.2.1/silk/arm/macros_armv4.h b/code/opus-1.5.2/silk/arm/macros_armv4.h similarity index 100% rename from code/opus-1.2.1/silk/arm/macros_armv4.h rename to code/opus-1.5.2/silk/arm/macros_armv4.h diff --git a/code/opus-1.2.1/silk/arm/macros_armv5e.h b/code/opus-1.5.2/silk/arm/macros_armv5e.h similarity index 100% rename from code/opus-1.2.1/silk/arm/macros_armv5e.h rename to code/opus-1.5.2/silk/arm/macros_armv5e.h diff --git a/code/opus-1.2.1/silk/biquad_alt.c b/code/opus-1.5.2/silk/biquad_alt.c similarity index 100% rename from code/opus-1.2.1/silk/biquad_alt.c rename to code/opus-1.5.2/silk/biquad_alt.c diff --git a/code/opus-1.2.1/silk/bwexpander.c b/code/opus-1.5.2/silk/bwexpander.c similarity index 100% rename from code/opus-1.2.1/silk/bwexpander.c rename to code/opus-1.5.2/silk/bwexpander.c diff --git a/code/opus-1.2.1/silk/bwexpander_32.c b/code/opus-1.5.2/silk/bwexpander_32.c similarity index 94% rename from code/opus-1.2.1/silk/bwexpander_32.c rename to code/opus-1.5.2/silk/bwexpander_32.c index d0010f73..0f32b9df 100644 --- a/code/opus-1.2.1/silk/bwexpander_32.c +++ b/code/opus-1.5.2/silk/bwexpander_32.c @@ -31,7 +31,8 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" -/* Chirp (bandwidth expand) LP AR filter */ +/* Chirp (bandwidth expand) LP AR filter. + This logic is reused in _celt_lpc(). Any bug fixes should also be applied there. */ void silk_bwexpander_32( opus_int32 *ar, /* I/O AR filter to be expanded (without leading 1) */ const opus_int d, /* I Length of ar */ diff --git a/code/opus-1.2.1/silk/check_control_input.c b/code/opus-1.5.2/silk/check_control_input.c similarity index 93% rename from code/opus-1.2.1/silk/check_control_input.c rename to code/opus-1.5.2/silk/check_control_input.c index b5de9ce4..739fb01f 100644 --- a/code/opus-1.2.1/silk/check_control_input.c +++ b/code/opus-1.5.2/silk/check_control_input.c @@ -38,7 +38,7 @@ opus_int check_control_input( silk_EncControlStruct *encControl /* I Control structure */ ) { - silk_assert( encControl != NULL ); + celt_assert( encControl != NULL ); if( ( ( encControl->API_sampleRate != 8000 ) && ( encControl->API_sampleRate != 12000 ) && @@ -59,46 +59,46 @@ opus_int check_control_input( ( encControl->minInternalSampleRate > encControl->desiredInternalSampleRate ) || ( encControl->maxInternalSampleRate < encControl->desiredInternalSampleRate ) || ( encControl->minInternalSampleRate > encControl->maxInternalSampleRate ) ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_FS_NOT_SUPPORTED; } if( encControl->payloadSize_ms != 10 && encControl->payloadSize_ms != 20 && encControl->payloadSize_ms != 40 && encControl->payloadSize_ms != 60 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_PACKET_SIZE_NOT_SUPPORTED; } if( encControl->packetLossPercentage < 0 || encControl->packetLossPercentage > 100 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_LOSS_RATE; } if( encControl->useDTX < 0 || encControl->useDTX > 1 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_DTX_SETTING; } if( encControl->useCBR < 0 || encControl->useCBR > 1 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_CBR_SETTING; } if( encControl->useInBandFEC < 0 || encControl->useInBandFEC > 1 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_INBAND_FEC_SETTING; } if( encControl->nChannelsAPI < 1 || encControl->nChannelsAPI > ENCODER_NUM_CHANNELS ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; } if( encControl->nChannelsInternal < 1 || encControl->nChannelsInternal > ENCODER_NUM_CHANNELS ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; } if( encControl->nChannelsInternal > encControl->nChannelsAPI ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; } if( encControl->complexity < 0 || encControl->complexity > 10 ) { - silk_assert( 0 ); + celt_assert( 0 ); return SILK_ENC_INVALID_COMPLEXITY_SETTING; } diff --git a/code/opus-1.2.1/silk/code_signs.c b/code/opus-1.5.2/silk/code_signs.c similarity index 100% rename from code/opus-1.2.1/silk/code_signs.c rename to code/opus-1.5.2/silk/code_signs.c diff --git a/code/opus-1.2.1/silk/control.h b/code/opus-1.5.2/silk/control.h similarity index 95% rename from code/opus-1.2.1/silk/control.h rename to code/opus-1.5.2/silk/control.h index b76ec33c..f5633e62 100644 --- a/code/opus-1.2.1/silk/control.h +++ b/code/opus-1.5.2/silk/control.h @@ -77,6 +77,9 @@ typedef struct { /* I: Flag to enable in-band Forward Error Correction (FEC); 0/1 */ opus_int useInBandFEC; + /* I: Flag to enable in-band Deep REDundancy (DRED); 0/1 */ + opus_int useDRED; + /* I: Flag to actually code in-band Forward Error Correction (FEC) in the current packet; 0/1 */ opus_int LBRR_coded; @@ -141,6 +144,14 @@ typedef struct { /* O: Pitch lag of previous frame (0 if unvoiced), measured in samples at 48 kHz */ opus_int prevPitchLag; + + /* I: Enable Deep PLC */ + opus_int enable_deep_plc; + +#ifdef ENABLE_OSCE + /* I: OSCE method */ + opus_int osce_method; +#endif } silk_DecControlStruct; #ifdef __cplusplus diff --git a/code/opus-1.5.2/silk/control_SNR.c b/code/opus-1.5.2/silk/control_SNR.c new file mode 100644 index 00000000..9a6db275 --- /dev/null +++ b/code/opus-1.5.2/silk/control_SNR.c @@ -0,0 +1,113 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +/* These tables hold SNR values divided by 21 (so they fit in 8 bits) + for different target bitrates spaced at 400 bps interval. The first + 10 values are omitted (0-4 kb/s) because they're all zeros. + These tables were obtained by running different SNRs through the + encoder and measuring the active bitrate. */ +static const unsigned char silk_TargetRate_NB_21[117 - 10] = { + 0, 15, 39, 52, 61, 68, + 74, 79, 84, 88, 92, 95, 99,102,105,108,111,114,117,119,122,124, + 126,129,131,133,135,137,139,142,143,145,147,149,151,153,155,157, + 158,160,162,163,165,167,168,170,171,173,174,176,177,179,180,182, + 183,185,186,187,189,190,192,193,194,196,197,199,200,201,203,204, + 205,207,208,209,211,212,213,215,216,217,219,220,221,223,224,225, + 227,228,230,231,232,234,235,236,238,239,241,242,243,245,246,248, + 249,250,252,253,255 +}; + +static const unsigned char silk_TargetRate_MB_21[165 - 10] = { + 0, 0, 28, 43, 52, 59, + 65, 70, 74, 78, 81, 85, 87, 90, 93, 95, 98,100,102,105,107,109, + 111,113,115,116,118,120,122,123,125,127,128,130,131,133,134,136, + 137,138,140,141,143,144,145,147,148,149,151,152,153,154,156,157, + 158,159,160,162,163,164,165,166,167,168,169,171,172,173,174,175, + 176,177,178,179,180,181,182,183,184,185,186,187,188,188,189,190, + 191,192,193,194,195,196,197,198,199,200,201,202,203,203,204,205, + 206,207,208,209,210,211,212,213,214,214,215,216,217,218,219,220, + 221,222,223,224,224,225,226,227,228,229,230,231,232,233,234,235, + 236,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250, + 251,252,253,254,255 +}; + +static const unsigned char silk_TargetRate_WB_21[201 - 10] = { + 0, 0, 0, 8, 29, 41, + 49, 56, 62, 66, 70, 74, 77, 80, 83, 86, 88, 91, 93, 95, 97, 99, + 101,103,105,107,108,110,112,113,115,116,118,119,121,122,123,125, + 126,127,129,130,131,132,134,135,136,137,138,140,141,142,143,144, + 145,146,147,148,149,150,151,152,153,154,156,157,158,159,159,160, + 161,162,163,164,165,166,167,168,169,170,171,171,172,173,174,175, + 176,177,177,178,179,180,181,181,182,183,184,185,185,186,187,188, + 189,189,190,191,192,192,193,194,195,195,196,197,198,198,199,200, + 200,201,202,203,203,204,205,206,206,207,208,209,209,210,211,211, + 212,213,214,214,215,216,216,217,218,219,219,220,221,221,222,223, + 224,224,225,226,226,227,228,229,229,230,231,232,232,233,234,234, + 235,236,237,237,238,239,240,240,241,242,243,243,244,245,246,246, + 247,248,249,249,250,251,252,253,255 +}; + +/* Control SNR of redidual quantizer */ +opus_int silk_control_SNR( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + opus_int32 TargetRate_bps /* I Target max bitrate (bps) */ +) +{ + int id; + int bound; + const unsigned char *snr_table; + + psEncC->TargetRate_bps = TargetRate_bps; + if( psEncC->nb_subfr == 2 ) { + TargetRate_bps -= 2000 + psEncC->fs_kHz/16; + } + if( psEncC->fs_kHz == 8 ) { + bound = sizeof(silk_TargetRate_NB_21); + snr_table = silk_TargetRate_NB_21; + } else if( psEncC->fs_kHz == 12 ) { + bound = sizeof(silk_TargetRate_MB_21); + snr_table = silk_TargetRate_MB_21; + } else { + bound = sizeof(silk_TargetRate_WB_21); + snr_table = silk_TargetRate_WB_21; + } + id = (TargetRate_bps+200)/400; + id = silk_min(id - 10, bound-1); + if( id <= 0 ) { + psEncC->SNR_dB_Q7 = 0; + } else { + psEncC->SNR_dB_Q7 = snr_table[id]*21; + } + return SILK_NO_ERROR; +} diff --git a/code/opus-1.2.1/silk/control_audio_bandwidth.c b/code/opus-1.5.2/silk/control_audio_bandwidth.c similarity index 91% rename from code/opus-1.2.1/silk/control_audio_bandwidth.c rename to code/opus-1.5.2/silk/control_audio_bandwidth.c index 4f9bc5cb..f6d22d83 100644 --- a/code/opus-1.2.1/silk/control_audio_bandwidth.c +++ b/code/opus-1.5.2/silk/control_audio_bandwidth.c @@ -39,9 +39,15 @@ opus_int silk_control_audio_bandwidth( ) { opus_int fs_kHz; + opus_int orig_kHz; opus_int32 fs_Hz; - fs_kHz = psEncC->fs_kHz; + orig_kHz = psEncC->fs_kHz; + /* Handle a bandwidth-switching reset where we need to be aware what the last sampling rate was. */ + if( orig_kHz == 0 ) { + orig_kHz = psEncC->sLP.saved_fs_kHz; + } + fs_kHz = orig_kHz; fs_Hz = silk_SMULBB( fs_kHz, 1000 ); if( fs_Hz == 0 ) { /* Encoder has just been initialized */ @@ -61,7 +67,7 @@ opus_int silk_control_audio_bandwidth( } if( psEncC->allow_bandwidth_switch || encControl->opusCanSwitch ) { /* Check if we should switch down */ - if( silk_SMULBB( psEncC->fs_kHz, 1000 ) > psEncC->desiredInternal_fs_Hz ) + if( silk_SMULBB( orig_kHz, 1000 ) > psEncC->desiredInternal_fs_Hz ) { /* Switch down */ if( psEncC->sLP.mode == 0 ) { @@ -76,7 +82,7 @@ opus_int silk_control_audio_bandwidth( psEncC->sLP.mode = 0; /* Switch to a lower sample frequency */ - fs_kHz = psEncC->fs_kHz == 16 ? 12 : 8; + fs_kHz = orig_kHz == 16 ? 12 : 8; } else { if( psEncC->sLP.transition_frame_no <= 0 ) { encControl->switchReady = 1; @@ -90,12 +96,12 @@ opus_int silk_control_audio_bandwidth( } else /* Check if we should switch up */ - if( silk_SMULBB( psEncC->fs_kHz, 1000 ) < psEncC->desiredInternal_fs_Hz ) + if( silk_SMULBB( orig_kHz, 1000 ) < psEncC->desiredInternal_fs_Hz ) { /* Switch up */ if( encControl->opusCanSwitch ) { /* Switch to a higher sample frequency */ - fs_kHz = psEncC->fs_kHz == 8 ? 12 : 16; + fs_kHz = orig_kHz == 8 ? 12 : 16; /* New transition */ psEncC->sLP.transition_frame_no = 0; diff --git a/code/opus-1.2.1/silk/control_codec.c b/code/opus-1.5.2/silk/control_codec.c similarity index 96% rename from code/opus-1.2.1/silk/control_codec.c rename to code/opus-1.5.2/silk/control_codec.c index 9350fd4b..784ffe66 100644 --- a/code/opus-1.2.1/silk/control_codec.c +++ b/code/opus-1.5.2/silk/control_codec.c @@ -238,8 +238,8 @@ static opus_int silk_setup_fs( } /* Set internal sampling frequency */ - silk_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); - silk_assert( psEnc->sCmn.nb_subfr == 2 || psEnc->sCmn.nb_subfr == 4 ); + celt_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); + celt_assert( psEnc->sCmn.nb_subfr == 2 || psEnc->sCmn.nb_subfr == 4 ); if( psEnc->sCmn.fs_kHz != fs_kHz ) { /* reset part of the state */ silk_memset( &psEnc->sShape, 0, sizeof( psEnc->sShape ) ); @@ -299,7 +299,7 @@ static opus_int silk_setup_fs( } /* Check that settings are valid */ - silk_assert( ( psEnc->sCmn.subfr_length * psEnc->sCmn.nb_subfr ) == psEnc->sCmn.frame_length ); + celt_assert( ( psEnc->sCmn.subfr_length * psEnc->sCmn.nb_subfr ) == psEnc->sCmn.frame_length ); return ret; } @@ -312,7 +312,7 @@ static opus_int silk_setup_complexity( opus_int ret = 0; /* Set encoding complexity */ - silk_assert( Complexity >= 0 && Complexity <= 10 ); + celt_assert( Complexity >= 0 && Complexity <= 10 ); if( Complexity < 1 ) { psEncC->pitchEstimationComplexity = SILK_PE_MIN_COMPLEX; psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.8, 16 ); @@ -390,12 +390,12 @@ static opus_int silk_setup_complexity( psEncC->shapeWinLength = SUB_FRAME_LENGTH_MS * psEncC->fs_kHz + 2 * psEncC->la_shape; psEncC->Complexity = Complexity; - silk_assert( psEncC->pitchEstimationLPCOrder <= MAX_FIND_PITCH_LPC_ORDER ); - silk_assert( psEncC->shapingLPCOrder <= MAX_SHAPE_LPC_ORDER ); - silk_assert( psEncC->nStatesDelayedDecision <= MAX_DEL_DEC_STATES ); - silk_assert( psEncC->warping_Q16 <= 32767 ); - silk_assert( psEncC->la_shape <= LA_SHAPE_MAX ); - silk_assert( psEncC->shapeWinLength <= SHAPE_LPC_WIN_MAX ); + celt_assert( psEncC->pitchEstimationLPCOrder <= MAX_FIND_PITCH_LPC_ORDER ); + celt_assert( psEncC->shapingLPCOrder <= MAX_SHAPE_LPC_ORDER ); + celt_assert( psEncC->nStatesDelayedDecision <= MAX_DEL_DEC_STATES ); + celt_assert( psEncC->warping_Q16 <= 32767 ); + celt_assert( psEncC->la_shape <= LA_SHAPE_MAX ); + celt_assert( psEncC->shapeWinLength <= SHAPE_LPC_WIN_MAX ); return ret; } @@ -415,7 +415,7 @@ static OPUS_INLINE opus_int silk_setup_LBRR( /* Previous packet did not have LBRR, and was therefore coded at a higher bitrate */ psEncC->LBRR_GainIncreases = 7; } else { - psEncC->LBRR_GainIncreases = silk_max_int( 7 - silk_SMULWB( (opus_int32)psEncC->PacketLoss_perc, SILK_FIX_CONST( 0.4, 16 ) ), 2 ); + psEncC->LBRR_GainIncreases = silk_max_int( 7 - silk_SMULWB( (opus_int32)psEncC->PacketLoss_perc, SILK_FIX_CONST( 0.2, 16 ) ), 3 ); } } diff --git a/code/opus-1.2.1/silk/debug.c b/code/opus-1.5.2/silk/debug.c similarity index 98% rename from code/opus-1.2.1/silk/debug.c rename to code/opus-1.5.2/silk/debug.c index 9253faf7..46a24a47 100644 --- a/code/opus-1.2.1/silk/debug.c +++ b/code/opus-1.5.2/silk/debug.c @@ -29,19 +29,23 @@ POSSIBILITY OF SUCH DAMAGE. #include "config.h" #endif +typedef int prevent_empty_translation_unit_warning; + #include "debug.h" + +#if SILK_DEBUG || SILK_TIC_TOC #include "SigProc_FIX.h" +#endif #if SILK_TIC_TOC -#ifdef _WIN32 - #if (defined(_WIN32) || defined(_WINCE)) #include /* timer */ #else /* Linux or Mac*/ #include #endif +#ifdef _WIN32 unsigned long silk_GetHighResolutionTime(void) /* O time in usec*/ { /* Returns a time counter in microsec */ @@ -65,7 +69,7 @@ unsigned long GetHighResolutionTime(void) /* O time in usec*/ int silk_Timer_nTimers = 0; int silk_Timer_depth_ctr = 0; char silk_Timer_tags[silk_NUM_TIMERS_MAX][silk_NUM_TIMERS_MAX_TAG_LEN]; -#ifdef WIN32 +#ifdef _WIN32 LARGE_INTEGER silk_Timer_start[silk_NUM_TIMERS_MAX]; #else unsigned long silk_Timer_start[silk_NUM_TIMERS_MAX]; @@ -76,7 +80,7 @@ opus_int64 silk_Timer_sum[silk_NUM_TIMERS_MAX]; opus_int64 silk_Timer_max[silk_NUM_TIMERS_MAX]; opus_int64 silk_Timer_depth[silk_NUM_TIMERS_MAX]; -#ifdef WIN32 +#ifdef _WIN32 void silk_TimerSave(char *file_name) { if( silk_Timer_nTimers > 0 ) diff --git a/code/opus-1.2.1/silk/debug.h b/code/opus-1.5.2/silk/debug.h similarity index 99% rename from code/opus-1.2.1/silk/debug.h rename to code/opus-1.5.2/silk/debug.h index 6f68c1ca..36163e47 100644 --- a/code/opus-1.2.1/silk/debug.h +++ b/code/opus-1.5.2/silk/debug.h @@ -28,28 +28,29 @@ POSSIBILITY OF SUCH DAMAGE. #ifndef SILK_DEBUG_H #define SILK_DEBUG_H -#include "typedef.h" -#include /* file writing */ -#include /* strcpy, strcmp */ - -#ifdef __cplusplus -extern "C" -{ -#endif - -unsigned long GetHighResolutionTime(void); /* O time in usec*/ - /* Set to 1 to enable DEBUG_STORE_DATA() macros for dumping * intermediate signals from the codec. */ #define SILK_DEBUG 0 /* Flag for using timers */ -#define SILK_TIC_TOC 0 +#define SILK_TIC_TOC 0 +#if SILK_DEBUG || SILK_TIC_TOC +#include "typedef.h" +#include /* strcpy, strcmp */ +#include /* file writing */ +#endif + +#ifdef __cplusplus +extern "C" +{ +#endif #if SILK_TIC_TOC +unsigned long GetHighResolutionTime(void); /* O time in usec*/ + #if (defined(_WIN32) || defined(_WINCE)) #include /* timer */ #else /* Linux or Mac*/ diff --git a/code/opus-1.2.1/silk/dec_API.c b/code/opus-1.5.2/silk/dec_API.c similarity index 90% rename from code/opus-1.2.1/silk/dec_API.c rename to code/opus-1.5.2/silk/dec_API.c index b7d8ed48..c1091d13 100644 --- a/code/opus-1.2.1/silk/dec_API.c +++ b/code/opus-1.5.2/silk/dec_API.c @@ -33,6 +33,11 @@ POSSIBILITY OF SUCH DAMAGE. #include "stack_alloc.h" #include "os_support.h" +#ifdef ENABLE_OSCE +#include "osce.h" +#include "osce_structs.h" +#endif + /************************/ /* Decoder Super Struct */ /************************/ @@ -42,12 +47,33 @@ typedef struct { opus_int nChannelsAPI; opus_int nChannelsInternal; opus_int prev_decode_only_middle; +#ifdef ENABLE_OSCE + OSCEModel osce_model; +#endif } silk_decoder; /*********************/ /* Decoder functions */ /*********************/ + + +opus_int silk_LoadOSCEModels(void *decState, const unsigned char *data, int len) +{ +#ifdef ENABLE_OSCE + opus_int ret = SILK_NO_ERROR; + + ret = osce_load_models(&((silk_decoder *)decState)->osce_model, data, len); + ((silk_decoder *)decState)->osce_model.loaded = (ret == 0); + return ret; +#else + (void) decState; + (void) data; + (void) len; + return SILK_NO_ERROR; +#endif +} + opus_int silk_Get_Decoder_Size( /* O Returns error code */ opus_int *decSizeBytes /* O Number of bytes in SILK decoder state */ ) @@ -60,12 +86,37 @@ opus_int silk_Get_Decoder_Size( /* O Returns error co } /* Reset decoder state */ +opus_int silk_ResetDecoder( /* O Returns error code */ + void *decState /* I/O State */ +) +{ + opus_int n, ret = SILK_NO_ERROR; + silk_decoder_state *channel_state = ((silk_decoder *)decState)->channel_state; + + for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) { + ret = silk_reset_decoder( &channel_state[ n ] ); + } + silk_memset(&((silk_decoder *)decState)->sStereo, 0, sizeof(((silk_decoder *)decState)->sStereo)); + /* Not strictly needed, but it's cleaner that way */ + ((silk_decoder *)decState)->prev_decode_only_middle = 0; + + return ret; +} + + opus_int silk_InitDecoder( /* O Returns error code */ void *decState /* I/O State */ ) { opus_int n, ret = SILK_NO_ERROR; silk_decoder_state *channel_state = ((silk_decoder *)decState)->channel_state; +#ifdef ENABLE_OSCE + ((silk_decoder *)decState)->osce_model.loaded = 0; +#endif +#ifndef USE_WEIGHTS_FILE + /* load osce models */ + silk_LoadOSCEModels(decState, NULL, 0); +#endif for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) { ret = silk_init_decoder( &channel_state[ n ] ); @@ -86,6 +137,9 @@ opus_int silk_Decode( /* O Returns error co ec_dec *psRangeDec, /* I/O Compressor data structure */ opus_int16 *samplesOut, /* O Decoded output speech vector */ opus_int32 *nSamplesOut, /* O Number of samples decoded */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif int arch /* I Run-time architecture */ ) { @@ -104,7 +158,7 @@ opus_int silk_Decode( /* O Returns error co int delay_stack_alloc; SAVE_STACK; - silk_assert( decControl->nChannelsInternal == 1 || decControl->nChannelsInternal == 2 ); + celt_assert( decControl->nChannelsInternal == 1 || decControl->nChannelsInternal == 2 ); /**********************************/ /* Test if first frame in payload */ @@ -143,13 +197,13 @@ opus_int silk_Decode( /* O Returns error co channel_state[ n ].nFramesPerPacket = 3; channel_state[ n ].nb_subfr = 4; } else { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return SILK_DEC_INVALID_FRAME_SIZE; } fs_kHz_dec = ( decControl->internalSampleRate >> 10 ) + 1; if( fs_kHz_dec != 8 && fs_kHz_dec != 12 && fs_kHz_dec != 16 ) { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return SILK_DEC_INVALID_SAMPLING_FREQUENCY; } @@ -278,6 +332,7 @@ opus_int silk_Decode( /* O Returns error co has_side = !psDec->prev_decode_only_middle || (decControl->nChannelsInternal == 2 && lostFlag == FLAG_DECODE_LBRR && channel_state[1].LBRR_flags[ channel_state[1].nFramesDecoded ] == 1 ); } + channel_state[ 0 ].sPLC.enable_deep_plc = decControl->enable_deep_plc; /* Call decoder for one frame */ for( n = 0; n < decControl->nChannelsInternal; n++ ) { if( n == 0 || has_side ) { @@ -297,7 +352,19 @@ opus_int silk_Decode( /* O Returns error co } else { condCoding = CODE_CONDITIONALLY; } - ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding, arch); +#ifdef ENABLE_OSCE + if ( channel_state[n].osce.method != decControl->osce_method ) { + osce_reset( &channel_state[n].osce, decControl->osce_method ); + } +#endif + ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding, +#ifdef ENABLE_DEEP_PLC + n == 0 ? lpcnet : NULL, +#endif +#ifdef ENABLE_OSCE + &psDec->osce_model, +#endif + arch); } else { silk_memset( &samplesOut1_tmp[ n ][ 2 ], 0, nSamplesOutDec * sizeof( opus_int16 ) ); } diff --git a/code/opus-1.2.1/silk/decode_core.c b/code/opus-1.5.2/silk/decode_core.c similarity index 99% rename from code/opus-1.2.1/silk/decode_core.c rename to code/opus-1.5.2/silk/decode_core.c index fd4e857a..1c352a65 100644 --- a/code/opus-1.2.1/silk/decode_core.c +++ b/code/opus-1.5.2/silk/decode_core.c @@ -141,7 +141,7 @@ void silk_decode_core( if( k == 0 || ( k == 2 && NLSF_interpolation_flag ) ) { /* Rewhiten with new A coefs */ start_idx = psDec->ltp_mem_length - lag - psDec->LPC_order - LTP_ORDER / 2; - silk_assert( start_idx > 0 ); + celt_assert( start_idx > 0 ); if( k == 2 ) { silk_memcpy( &psDec->outBuf[ psDec->ltp_mem_length ], xq, 2 * psDec->subfr_length * sizeof( opus_int16 ) ); @@ -196,7 +196,7 @@ void silk_decode_core( for( i = 0; i < psDec->subfr_length; i++ ) { /* Short-term prediction */ - silk_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); + celt_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12_tmp[ 0 ] ); diff --git a/code/opus-1.2.1/silk/decode_frame.c b/code/opus-1.5.2/silk/decode_frame.c similarity index 73% rename from code/opus-1.2.1/silk/decode_frame.c rename to code/opus-1.5.2/silk/decode_frame.c index dfa73c4a..9bc4ca2b 100644 --- a/code/opus-1.2.1/silk/decode_frame.c +++ b/code/opus-1.5.2/silk/decode_frame.c @@ -33,6 +33,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "stack_alloc.h" #include "PLC.h" +#ifdef ENABLE_OSCE +#include "osce.h" +#endif + /****************/ /* Decode frame */ /****************/ @@ -43,6 +47,12 @@ opus_int silk_decode_frame( opus_int32 *pN, /* O Pointer to size of output frame */ opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ opus_int condCoding, /* I The type of conditional coding to use */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif +#ifdef ENABLE_OSCE + OSCEModel *osce_model, +#endif int arch /* I Run-time architecture */ ) { @@ -55,12 +65,16 @@ opus_int silk_decode_frame( psDecCtrl->LTP_scale_Q14 = 0; /* Safety checks */ - silk_assert( L > 0 && L <= MAX_FRAME_LENGTH ); + celt_assert( L > 0 && L <= MAX_FRAME_LENGTH ); if( lostFlag == FLAG_DECODE_NORMAL || ( lostFlag == FLAG_DECODE_LBRR && psDec->LBRR_flags[ psDec->nFramesDecoded ] == 1 ) ) { VARDECL( opus_int16, pulses ); +#ifdef ENABLE_OSCE + opus_int32 ec_start; + ec_start = ec_tell(psRangeDec); +#endif ALLOC( pulses, (L + SHELL_CODEC_FRAME_LENGTH - 1) & ~(SHELL_CODEC_FRAME_LENGTH - 1), opus_int16 ); /*********************************************/ @@ -84,30 +98,55 @@ opus_int silk_decode_frame( /********************************************************/ silk_decode_core( psDec, psDecCtrl, pOut, pulses, arch ); + /*************************/ + /* Update output buffer. */ + /*************************/ + celt_assert( psDec->ltp_mem_length >= psDec->frame_length ); + mv_len = psDec->ltp_mem_length - psDec->frame_length; + silk_memmove( psDec->outBuf, &psDec->outBuf[ psDec->frame_length ], mv_len * sizeof(opus_int16) ); + silk_memcpy( &psDec->outBuf[ mv_len ], pOut, psDec->frame_length * sizeof( opus_int16 ) ); + +#ifdef ENABLE_OSCE + /********************************************************/ + /* Run SILK enhancer */ + /********************************************************/ + osce_enhance_frame( osce_model, psDec, psDecCtrl, pOut, ec_tell(psRangeDec) - ec_start, arch ); +#endif + /********************************************************/ /* Update PLC state */ /********************************************************/ - silk_PLC( psDec, psDecCtrl, pOut, 0, arch ); + silk_PLC( psDec, psDecCtrl, pOut, 0, +#ifdef ENABLE_DEEP_PLC + lpcnet, +#endif + arch ); psDec->lossCnt = 0; psDec->prevSignalType = psDec->indices.signalType; - silk_assert( psDec->prevSignalType >= 0 && psDec->prevSignalType <= 2 ); + celt_assert( psDec->prevSignalType >= 0 && psDec->prevSignalType <= 2 ); /* A frame has been decoded without errors */ psDec->first_frame_after_reset = 0; } else { /* Handle packet loss by extrapolation */ - psDec->indices.signalType = psDec->prevSignalType; - silk_PLC( psDec, psDecCtrl, pOut, 1, arch ); - } + silk_PLC( psDec, psDecCtrl, pOut, 1, +#ifdef ENABLE_DEEP_PLC + lpcnet, +#endif + arch ); - /*************************/ - /* Update output buffer. */ - /*************************/ - silk_assert( psDec->ltp_mem_length >= psDec->frame_length ); - mv_len = psDec->ltp_mem_length - psDec->frame_length; - silk_memmove( psDec->outBuf, &psDec->outBuf[ psDec->frame_length ], mv_len * sizeof(opus_int16) ); - silk_memcpy( &psDec->outBuf[ mv_len ], pOut, psDec->frame_length * sizeof( opus_int16 ) ); +#ifdef ENABLE_OSCE + osce_reset( &psDec->osce, psDec->osce.method ); +#endif + /*************************/ + /* Update output buffer. */ + /*************************/ + celt_assert( psDec->ltp_mem_length >= psDec->frame_length ); + mv_len = psDec->ltp_mem_length - psDec->frame_length; + silk_memmove( psDec->outBuf, &psDec->outBuf[ psDec->frame_length ], mv_len * sizeof(opus_int16) ); + silk_memcpy( &psDec->outBuf[ mv_len ], pOut, psDec->frame_length * sizeof( opus_int16 ) ); + } /************************************************/ /* Comfort noise generation / estimation */ diff --git a/code/opus-1.2.1/silk/decode_indices.c b/code/opus-1.5.2/silk/decode_indices.c similarity index 99% rename from code/opus-1.2.1/silk/decode_indices.c rename to code/opus-1.5.2/silk/decode_indices.c index 7afe5c26..0bb4a997 100644 --- a/code/opus-1.2.1/silk/decode_indices.c +++ b/code/opus-1.5.2/silk/decode_indices.c @@ -79,7 +79,7 @@ void silk_decode_indices( /**********************/ psDec->indices.NLSFIndices[ 0 ] = (opus_int8)ec_dec_icdf( psRangeDec, &psDec->psNLSF_CB->CB1_iCDF[ ( psDec->indices.signalType >> 1 ) * psDec->psNLSF_CB->nVectors ], 8 ); silk_NLSF_unpack( ec_ix, pred_Q8, psDec->psNLSF_CB, psDec->indices.NLSFIndices[ 0 ] ); - silk_assert( psDec->psNLSF_CB->order == psDec->LPC_order ); + celt_assert( psDec->psNLSF_CB->order == psDec->LPC_order ); for( i = 0; i < psDec->psNLSF_CB->order; i++ ) { Ix = ec_dec_icdf( psRangeDec, &psDec->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); if( Ix == 0 ) { diff --git a/code/opus-1.2.1/silk/decode_parameters.c b/code/opus-1.5.2/silk/decode_parameters.c similarity index 100% rename from code/opus-1.2.1/silk/decode_parameters.c rename to code/opus-1.5.2/silk/decode_parameters.c diff --git a/code/opus-1.2.1/silk/decode_pitch.c b/code/opus-1.5.2/silk/decode_pitch.c similarity index 96% rename from code/opus-1.2.1/silk/decode_pitch.c rename to code/opus-1.5.2/silk/decode_pitch.c index fedbc6a5..fd1b6bf5 100644 --- a/code/opus-1.2.1/silk/decode_pitch.c +++ b/code/opus-1.5.2/silk/decode_pitch.c @@ -51,7 +51,7 @@ void silk_decode_pitch( Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ]; cbk_size = PE_NB_CBKS_STAGE2_EXT; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; cbk_size = PE_NB_CBKS_STAGE2_10MS; } @@ -60,7 +60,7 @@ void silk_decode_pitch( Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; cbk_size = PE_NB_CBKS_STAGE3_MAX; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; cbk_size = PE_NB_CBKS_STAGE3_10MS; } diff --git a/code/opus-1.2.1/silk/decode_pulses.c b/code/opus-1.5.2/silk/decode_pulses.c similarity index 98% rename from code/opus-1.2.1/silk/decode_pulses.c rename to code/opus-1.5.2/silk/decode_pulses.c index d6bbec92..a56d2d30 100644 --- a/code/opus-1.2.1/silk/decode_pulses.c +++ b/code/opus-1.5.2/silk/decode_pulses.c @@ -56,7 +56,7 @@ void silk_decode_pulses( silk_assert( 1 << LOG2_SHELL_CODEC_FRAME_LENGTH == SHELL_CODEC_FRAME_LENGTH ); iter = silk_RSHIFT( frame_length, LOG2_SHELL_CODEC_FRAME_LENGTH ); if( iter * SHELL_CODEC_FRAME_LENGTH < frame_length ) { - silk_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ + celt_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ iter++; } diff --git a/code/opus-1.2.1/silk/decoder_set_fs.c b/code/opus-1.5.2/silk/decoder_set_fs.c similarity index 95% rename from code/opus-1.2.1/silk/decoder_set_fs.c rename to code/opus-1.5.2/silk/decoder_set_fs.c index eef0fd25..d9a13d0f 100644 --- a/code/opus-1.2.1/silk/decoder_set_fs.c +++ b/code/opus-1.5.2/silk/decoder_set_fs.c @@ -40,8 +40,8 @@ opus_int silk_decoder_set_fs( { opus_int frame_length, ret = 0; - silk_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); - silk_assert( psDec->nb_subfr == MAX_NB_SUBFR || psDec->nb_subfr == MAX_NB_SUBFR/2 ); + celt_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); + celt_assert( psDec->nb_subfr == MAX_NB_SUBFR || psDec->nb_subfr == MAX_NB_SUBFR/2 ); /* New (sub)frame length */ psDec->subfr_length = silk_SMULBB( SUB_FRAME_LENGTH_MS, fs_kHz ); @@ -86,7 +86,7 @@ opus_int silk_decoder_set_fs( psDec->pitch_lag_low_bits_iCDF = silk_uniform4_iCDF; } else { /* unsupported sampling rate */ - silk_assert( 0 ); + celt_assert( 0 ); } psDec->first_frame_after_reset = 1; psDec->lagPrev = 100; @@ -101,7 +101,7 @@ opus_int silk_decoder_set_fs( } /* Check that settings are valid */ - silk_assert( psDec->frame_length > 0 && psDec->frame_length <= MAX_FRAME_LENGTH ); + celt_assert( psDec->frame_length > 0 && psDec->frame_length <= MAX_FRAME_LENGTH ); return ret; } diff --git a/code/opus-1.2.1/silk/define.h b/code/opus-1.5.2/silk/define.h similarity index 97% rename from code/opus-1.2.1/silk/define.h rename to code/opus-1.5.2/silk/define.h index 1286048e..491c86f3 100644 --- a/code/opus-1.2.1/silk/define.h +++ b/code/opus-1.5.2/silk/define.h @@ -46,7 +46,6 @@ extern "C" /* Limits on bitrate */ #define MIN_TARGET_RATE_BPS 5000 #define MAX_TARGET_RATE_BPS 80000 -#define TARGET_RATE_TAB_SZ 8 /* LBRR thresholds */ #define LBRR_NB_MIN_RATE_BPS 12000 @@ -58,6 +57,11 @@ extern "C" #define MAX_CONSECUTIVE_DTX 20 /* eq 400 ms */ #define DTX_ACTIVITY_THRESHOLD 0.1f +/* VAD decision */ +#define VAD_NO_DECISION -1 +#define VAD_NO_ACTIVITY 0 +#define VAD_ACTIVITY 1 + /* Maximum sampling frequency */ #define MAX_FS_KHZ 16 #define MAX_API_FS_KHZ 48 @@ -221,6 +225,7 @@ extern "C" /* Defines for CN generation */ #define CNG_BUF_MASK_MAX 255 /* 2^floor(log2(MAX_FRAME_LENGTH))-1 */ #define CNG_GAIN_SMTH_Q16 4634 /* 0.25^(1/4) */ +#define CNG_GAIN_SMTH_THRESHOLD_Q16 46396 /* -3 dB */ #define CNG_NLSF_SMTH_Q16 16348 /* 0.25 */ #ifdef __cplusplus diff --git a/code/opus-1.2.1/silk/enc_API.c b/code/opus-1.5.2/silk/enc_API.c similarity index 93% rename from code/opus-1.2.1/silk/enc_API.c rename to code/opus-1.5.2/silk/enc_API.c index 701c2905..369caddd 100644 --- a/code/opus-1.2.1/silk/enc_API.c +++ b/code/opus-1.5.2/silk/enc_API.c @@ -41,6 +41,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "main_FLP.h" #endif +#ifdef ENABLE_DRED +#include "dred_encoder.h" +#endif + /***************************************/ /* Read control structure from encoder */ /***************************************/ @@ -82,7 +86,7 @@ opus_int silk_InitEncoder( /* O Returns error co silk_memset( psEnc, 0, sizeof( silk_encoder ) ); for( n = 0; n < ENCODER_NUM_CHANNELS; n++ ) { if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) { - silk_assert( 0 ); + celt_assert( 0 ); } } @@ -91,7 +95,7 @@ opus_int silk_InitEncoder( /* O Returns error co /* Read control structure */ if( ret += silk_QueryEncoder( encState, encStatus ) ) { - silk_assert( 0 ); + celt_assert( 0 ); } return ret; @@ -144,7 +148,8 @@ opus_int silk_Encode( /* O Returns error co opus_int nSamplesIn, /* I Number of samples in input vector */ ec_enc *psRangeEnc, /* I/O Compressor data structure */ opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */ - const opus_int prefillFlag /* I Flag to indicate prefilling buffers no coding */ + const opus_int prefillFlag, /* I Flag to indicate prefilling buffers no coding */ + opus_int activity /* I Decision of Opus voice activity detector */ ) { opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0; @@ -166,7 +171,7 @@ opus_int silk_Encode( /* O Returns error co /* Check values in encoder control structure */ if( ( ret = check_control_input( encControl ) ) != 0 ) { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return ret; } @@ -199,16 +204,26 @@ opus_int silk_Encode( /* O Returns error co tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1; curr_block = 0; if( prefillFlag ) { + silk_LP_state save_LP; /* Only accept input length of 10 ms */ if( nBlocksOf10ms != 1 ) { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } + if ( prefillFlag == 2 ) { + save_LP = psEnc->state_Fxx[ 0 ].sCmn.sLP; + /* Save the sampling rate so the bandwidth switching code can keep handling transitions. */ + save_LP.saved_fs_kHz = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + } /* Reset Encoder */ for( n = 0; n < encControl->nChannelsInternal; n++ ) { ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch ); - silk_assert( !ret ); + /* Restore the variable LP state. */ + if ( prefillFlag == 2 ) { + psEnc->state_Fxx[ n ].sCmn.sLP = save_LP; + } + celt_assert( !ret ); } tmp_payloadSize_ms = encControl->payloadSize_ms; encControl->payloadSize_ms = 10; @@ -221,13 +236,13 @@ opus_int silk_Encode( /* O Returns error co } else { /* Only accept input lengths that are a multiple of 10 ms */ if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } /* Make sure no more than one packet can be produced */ if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) { - silk_assert( 0 ); + celt_assert( 0 ); RESTORE_STACK; return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } @@ -248,7 +263,7 @@ opus_int silk_Encode( /* O Returns error co } psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX; } - silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); /* Input buffering/resampling and encoding */ nSamplesToBufferMax = @@ -259,6 +274,7 @@ opus_int silk_Encode( /* O Returns error co psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); ALLOC( buf, nSamplesFromInputMax, opus_int16 ); while( 1 ) { + int curr_nBitsUsedLBRR = 0; nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx; nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax ); nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); @@ -306,7 +322,7 @@ opus_int silk_Encode( /* O Returns error co } psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; } else { - silk_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 ); + celt_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 ); silk_memcpy(buf, samplesIn, nSamplesFromInput*sizeof(opus_int16)); ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); @@ -322,8 +338,8 @@ opus_int silk_Encode( /* O Returns error co /* Silk encoder */ if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) { /* Enough data in input buffer, so encode */ - silk_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length ); - silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length ); + celt_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length ); + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length ); /* Deal with LBRR data */ if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) { @@ -331,6 +347,7 @@ opus_int silk_Encode( /* O Returns error co opus_uint8 iCDF[ 2 ] = { 0, 0 }; iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal ); ec_enc_icdf( psRangeEnc, 0, iCDF, 8 ); + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ); /* Encode any LBRR data from previous packet */ /* Encode LBRR flags */ @@ -375,8 +392,7 @@ opus_int silk_Encode( /* O Returns error co for( n = 0; n < encControl->nChannelsInternal; n++ ) { silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) ); } - - psEnc->nBitsUsedLBRR = ec_tell( psRangeEnc ); + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ) - curr_nBitsUsedLBRR; } silk_HP_variable_cutoff( psEnc->state_Fxx ); @@ -385,6 +401,16 @@ opus_int silk_Encode( /* O Returns error co nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 ); /* Subtract bits used for LBRR */ if( !prefillFlag ) { + /* psEnc->nBitsUsedLBRR is an exponential moving average of the LBRR usage, + except that for the first LBRR frame it does no averaging and for the first + frame after after LBRR, it goes back to zero immediately. */ + if ( curr_nBitsUsedLBRR < 10 ) { + psEnc->nBitsUsedLBRR = 0; + } else if ( psEnc->nBitsUsedLBRR < 10) { + psEnc->nBitsUsedLBRR = curr_nBitsUsedLBRR; + } else { + psEnc->nBitsUsedLBRR = ( psEnc->nBitsUsedLBRR + curr_nBitsUsedLBRR ) / 2; + } nBits -= psEnc->nBitsUsedLBRR; } /* Divide by number of uncoded frames left in packet */ @@ -425,7 +451,7 @@ opus_int silk_Encode( /* O Returns error co psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16 = 65536; psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1; } - silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ] ); + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ], activity ); } else { psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0; } @@ -440,7 +466,7 @@ opus_int silk_Encode( /* O Returns error co silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) ); silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) ); } - silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ] ); + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ], activity ); /* Encode */ for( n = 0; n < encControl->nChannelsInternal; n++ ) { diff --git a/code/opus-1.2.1/silk/encode_indices.c b/code/opus-1.5.2/silk/encode_indices.c similarity index 98% rename from code/opus-1.2.1/silk/encode_indices.c rename to code/opus-1.5.2/silk/encode_indices.c index 666c8c0b..4bcbc334 100644 --- a/code/opus-1.2.1/silk/encode_indices.c +++ b/code/opus-1.5.2/silk/encode_indices.c @@ -56,8 +56,8 @@ void silk_encode_indices( /* Encode signal type and quantizer offset */ /*******************************************/ typeOffset = 2 * psIndices->signalType + psIndices->quantOffsetType; - silk_assert( typeOffset >= 0 && typeOffset < 6 ); - silk_assert( encode_LBRR == 0 || typeOffset >= 2 ); + celt_assert( typeOffset >= 0 && typeOffset < 6 ); + celt_assert( encode_LBRR == 0 || typeOffset >= 2 ); if( encode_LBRR || typeOffset >= 2 ) { ec_enc_icdf( psRangeEnc, typeOffset - 2, silk_type_offset_VAD_iCDF, 8 ); } else { @@ -90,7 +90,7 @@ void silk_encode_indices( /****************/ ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ 0 ], &psEncC->psNLSF_CB->CB1_iCDF[ ( psIndices->signalType >> 1 ) * psEncC->psNLSF_CB->nVectors ], 8 ); silk_NLSF_unpack( ec_ix, pred_Q8, psEncC->psNLSF_CB, psIndices->NLSFIndices[ 0 ] ); - silk_assert( psEncC->psNLSF_CB->order == psEncC->predictLPCOrder ); + celt_assert( psEncC->psNLSF_CB->order == psEncC->predictLPCOrder ); for( i = 0; i < psEncC->psNLSF_CB->order; i++ ) { if( psIndices->NLSFIndices[ i+1 ] >= NLSF_QUANT_MAX_AMPLITUDE ) { ec_enc_icdf( psRangeEnc, 2 * NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); diff --git a/code/opus-1.2.1/silk/encode_pulses.c b/code/opus-1.5.2/silk/encode_pulses.c similarity index 99% rename from code/opus-1.2.1/silk/encode_pulses.c rename to code/opus-1.5.2/silk/encode_pulses.c index ab00264f..8a199913 100644 --- a/code/opus-1.2.1/silk/encode_pulses.c +++ b/code/opus-1.5.2/silk/encode_pulses.c @@ -86,7 +86,7 @@ void silk_encode_pulses( silk_assert( 1 << LOG2_SHELL_CODEC_FRAME_LENGTH == SHELL_CODEC_FRAME_LENGTH ); iter = silk_RSHIFT( frame_length, LOG2_SHELL_CODEC_FRAME_LENGTH ); if( iter * SHELL_CODEC_FRAME_LENGTH < frame_length ) { - silk_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ + celt_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ iter++; silk_memset( &pulses[ frame_length ], 0, SHELL_CODEC_FRAME_LENGTH * sizeof(opus_int8)); } diff --git a/code/opus-1.2.1/silk/errors.h b/code/opus-1.5.2/silk/errors.h similarity index 100% rename from code/opus-1.2.1/silk/errors.h rename to code/opus-1.5.2/silk/errors.h diff --git a/code/opus-1.2.1/silk/fixed/LTP_analysis_filter_FIX.c b/code/opus-1.5.2/silk/fixed/LTP_analysis_filter_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/LTP_analysis_filter_FIX.c rename to code/opus-1.5.2/silk/fixed/LTP_analysis_filter_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/LTP_scale_ctrl_FIX.c b/code/opus-1.5.2/silk/fixed/LTP_scale_ctrl_FIX.c similarity index 78% rename from code/opus-1.2.1/silk/fixed/LTP_scale_ctrl_FIX.c rename to code/opus-1.5.2/silk/fixed/LTP_scale_ctrl_FIX.c index 3dcedef8..db1016e0 100644 --- a/code/opus-1.2.1/silk/fixed/LTP_scale_ctrl_FIX.c +++ b/code/opus-1.5.2/silk/fixed/LTP_scale_ctrl_FIX.c @@ -42,9 +42,14 @@ void silk_LTP_scale_ctrl_FIX( if( condCoding == CODE_INDEPENDENTLY ) { /* Only scale if first frame in packet */ - round_loss = psEnc->sCmn.PacketLoss_perc + psEnc->sCmn.nFramesPerPacket; - psEnc->sCmn.indices.LTP_scaleIndex = (opus_int8)silk_LIMIT( - silk_SMULWB( silk_SMULBB( round_loss, psEncCtrl->LTPredCodGain_Q7 ), SILK_FIX_CONST( 0.1, 9 ) ), 0, 2 ); + round_loss = psEnc->sCmn.PacketLoss_perc * psEnc->sCmn.nFramesPerPacket; + if ( psEnc->sCmn.LBRR_flag ) { + /* LBRR reduces the effective loss. In practice, it does not square the loss because + losses aren't independent, but that still seems to work best. We also never go below 2%. */ + round_loss = 2 + silk_SMULBB( round_loss, round_loss ) / 100; + } + psEnc->sCmn.indices.LTP_scaleIndex = silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 2900-psEnc->sCmn.SNR_dB_Q7 ); + psEnc->sCmn.indices.LTP_scaleIndex += silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 3900-psEnc->sCmn.SNR_dB_Q7 ); } else { /* Default is minimum scaling */ psEnc->sCmn.indices.LTP_scaleIndex = 0; diff --git a/code/opus-1.2.1/silk/fixed/apply_sine_window_FIX.c b/code/opus-1.5.2/silk/fixed/apply_sine_window_FIX.c similarity index 96% rename from code/opus-1.2.1/silk/fixed/apply_sine_window_FIX.c rename to code/opus-1.5.2/silk/fixed/apply_sine_window_FIX.c index 4502b713..03e088a6 100644 --- a/code/opus-1.2.1/silk/fixed/apply_sine_window_FIX.c +++ b/code/opus-1.5.2/silk/fixed/apply_sine_window_FIX.c @@ -57,15 +57,15 @@ void silk_apply_sine_window( opus_int k, f_Q16, c_Q16; opus_int32 S0_Q16, S1_Q16; - silk_assert( win_type == 1 || win_type == 2 ); + celt_assert( win_type == 1 || win_type == 2 ); /* Length must be in a range from 16 to 120 and a multiple of 4 */ - silk_assert( length >= 16 && length <= 120 ); - silk_assert( ( length & 3 ) == 0 ); + celt_assert( length >= 16 && length <= 120 ); + celt_assert( ( length & 3 ) == 0 ); /* Frequency */ k = ( length >> 2 ) - 4; - silk_assert( k >= 0 && k <= 26 ); + celt_assert( k >= 0 && k <= 26 ); f_Q16 = (opus_int)freq_table_Q16[ k ]; /* Factor used for cosine approximation */ diff --git a/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_arm.h b/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_arm.h new file mode 100644 index 00000000..1992e432 --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_arm.h @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_WARPED_AUTOCORRELATION_FIX_ARM_H +# define SILK_WARPED_AUTOCORRELATION_FIX_ARM_H + +# include "celt/arm/armcpu.h" + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), PRESUME_NEON(silk_warped_autocorrelation_FIX)(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# if !defined(OVERRIDE_silk_warped_autocorrelation_FIX) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const SILK_WARPED_AUTOCORRELATION_FIX_IMPL[OPUS_ARCHMASK+1])(opus_int32*, opus_int*, const opus_int16*, const opus_int, const opus_int, const opus_int); +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((*SILK_WARPED_AUTOCORRELATION_FIX_IMPL[(arch)&OPUS_ARCHMASK])(corr, scale, input, warping_Q16, length, order)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), silk_warped_autocorrelation_FIX_neon(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# endif /* end FIXED_POINT */ + +#endif /* end SILK_WARPED_AUTOCORRELATION_FIX_ARM_H */ diff --git a/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c b/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c new file mode 100644 index 00000000..6f3be025 --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c @@ -0,0 +1,265 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc., Jean-Marc Valin +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#ifdef OPUS_CHECK_ASM +# include +#endif +#include "stack_alloc.h" +#include "main_FIX.h" + +static OPUS_INLINE void calc_corr( const opus_int32 *const input_QS, opus_int64 *const corr_QC, const opus_int offset, const int32x4_t state_QS_s32x4 ) +{ + int64x2_t corr_QC_s64x2[ 2 ], t_s64x2[ 2 ]; + const int32x4_t input_QS_s32x4 = vld1q_s32( input_QS + offset ); + corr_QC_s64x2[ 0 ] = vld1q_s64( corr_QC + offset + 0 ); + corr_QC_s64x2[ 1 ] = vld1q_s64( corr_QC + offset + 2 ); + t_s64x2[ 0 ] = vmull_s32( vget_low_s32( state_QS_s32x4 ), vget_low_s32( input_QS_s32x4 ) ); + t_s64x2[ 1 ] = vmull_s32( vget_high_s32( state_QS_s32x4 ), vget_high_s32( input_QS_s32x4 ) ); + corr_QC_s64x2[ 0 ] = vsraq_n_s64( corr_QC_s64x2[ 0 ], t_s64x2[ 0 ], 2 * QS - QC ); + corr_QC_s64x2[ 1 ] = vsraq_n_s64( corr_QC_s64x2[ 1 ], t_s64x2[ 1 ], 2 * QS - QC ); + vst1q_s64( corr_QC + offset + 0, corr_QC_s64x2[ 0 ] ); + vst1q_s64( corr_QC + offset + 2, corr_QC_s64x2[ 1 ] ); +} + +static OPUS_INLINE int32x4_t calc_state( const int32x4_t state_QS0_s32x4, const int32x4_t state_QS0_1_s32x4, const int32x4_t state_QS1_1_s32x4, const int32x4_t warping_Q16_s32x4 ) +{ + int32x4_t t_s32x4 = vsubq_s32( state_QS0_s32x4, state_QS0_1_s32x4 ); + t_s32x4 = vqdmulhq_s32( t_s32x4, warping_Q16_s32x4 ); + return vaddq_s32( state_QS1_1_s32x4, t_s32x4 ); +} + +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + if( ( MAX_SHAPE_LPC_ORDER > 24 ) || ( order < 6 ) ) { + silk_warped_autocorrelation_FIX_c( corr, scale, input, warping_Q16, length, order ); + } else { + opus_int n, i, lsh; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; /* In reverse order */ + opus_int64 corr_QC_orderT; + int64x2_t lsh_s64x2; + const opus_int orderT = ( order + 3 ) & ~3; + opus_int64 *corr_QCT; + opus_int32 *input_QS; + VARDECL( opus_int32, input_QST ); + VARDECL( opus_int32, state ); + SAVE_STACK; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + /* Strictly, only +3 is needed but +4 simplifies initialization using the 4x32 neon load. */ + ALLOC( input_QST, length + 2 * MAX_SHAPE_LPC_ORDER + 4, opus_int32 ); + + input_QS = input_QST; + /* input_QS has zero paddings in the beginning and end. */ + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + + /* Loop over samples */ + for( n = 0; n < length - 7; n += 8, input_QS += 8 ) { + const int16x8_t t0_s16x4 = vld1q_s16( input + n ); + vst1q_s32( input_QS + 0, vshll_n_s16( vget_low_s16( t0_s16x4 ), QS ) ); + vst1q_s32( input_QS + 4, vshll_n_s16( vget_high_s16( t0_s16x4 ), QS ) ); + } + for( ; n < length; n++, input_QS++ ) { + input_QS[ 0 ] = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + } + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS = input_QST + MAX_SHAPE_LPC_ORDER - orderT; + + /* The following loop runs ( length + order ) times, with ( order ) extra epilogues. */ + /* The zero paddings in input_QS guarantee corr_QC's correctness even with the extra epilogues. */ + /* The values of state_QS will be polluted by the extra epilogues, however they are temporary values. */ + + /* Keep the C code here to help understand the intrinsics optimization. */ + /* + { + opus_int32 state_QS[ 2 ][ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int32 *state_QST[ 3 ]; + state_QST[ 0 ] = state_QS[ 0 ]; + state_QST[ 1 ] = state_QS[ 1 ]; + for( n = 0; n < length + order; n++, input_QS++ ) { + state_QST[ 0 ][ orderT ] = input_QS[ orderT ]; + for( i = 0; i < orderT; i++ ) { + corr_QC[ i ] += silk_RSHIFT64( silk_SMULL( state_QST[ 0 ][ i ], input_QS[ i ] ), 2 * QS - QC ); + state_QST[ 1 ][ i ] = silk_SMLAWB( state_QST[ 1 ][ i + 1 ], state_QST[ 0 ][ i ] - state_QST[ 0 ][ i + 1 ], warping_Q16 ); + } + state_QST[ 2 ] = state_QST[ 0 ]; + state_QST[ 0 ] = state_QST[ 1 ]; + state_QST[ 1 ] = state_QST[ 2 ]; + } + } + */ + + { + const int32x4_t warping_Q16_s32x4 = vdupq_n_s32( warping_Q16 << 15 ); + const opus_int32 *in = input_QS + orderT; + opus_int o = orderT; + int32x4_t state_QS_s32x4[ 3 ][ 2 ]; + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + ALLOC( state, length + order + 4, opus_int32 ); + state_QS_s32x4[ 2 ][ 1 ] = vdupq_n_s32( 0 ); + + /* Calculate 8 taps of all inputs in each loop. */ + do { + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 0 ][ 1 ] = + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 1 ][ 1 ] = vdupq_n_s32( 0 ); + n = 0; + do { + calc_corr( input_QS + n, corr_QC, o - 8, state_QS_s32x4[ 0 ][ 0 ] ); + calc_corr( input_QS + n, corr_QC, o - 4, state_QS_s32x4[ 0 ][ 1 ] ); + state_QS_s32x4[ 2 ][ 1 ] = vld1q_s32( in + n ); + vst1q_lane_s32( state + n, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 0 ][ 1 ], 1 ); + state_QS_s32x4[ 2 ][ 1 ] = vextq_s32( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 0 ][ 1 ] = calc_state( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], state_QS_s32x4[ 1 ][ 1 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + state_QS_s32x4[ 1 ][ 1 ] = state_QS_s32x4[ 2 ][ 1 ]; + } while( ++n < ( length + order ) ); + in = state; + o -= 8; + } while( o > 4 ); + + if( o ) { + /* Calculate the last 4 taps of all inputs. */ + opus_int32 *stateT = state; + silk_assert( o == 4 ); + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 1 ][ 0 ] = vdupq_n_s32( 0 ); + n = length + order; + do { + calc_corr( input_QS, corr_QC, 0, state_QS_s32x4[ 0 ][ 0 ] ); + state_QS_s32x4[ 2 ][ 0 ] = vld1q_s32( stateT ); + vst1q_lane_s32( stateT, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + input_QS++; + stateT++; + } while( --n ); + } + } + + { + const opus_int16 *inputT = input; + int32x4_t t_s32x4; + int64x1_t t_s64x1; + int64x2_t t_s64x2 = vdupq_n_s64( 0 ); + for( n = 0; n <= length - 8; n += 8 ) { + int16x8_t input_s16x8 = vld1q_s16( inputT ); + t_s32x4 = vmull_s16( vget_low_s16( input_s16x8 ), vget_low_s16( input_s16x8 ) ); + t_s32x4 = vmlal_s16( t_s32x4, vget_high_s16( input_s16x8 ), vget_high_s16( input_s16x8 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_low_s32( t_s32x4 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_high_s32( t_s32x4 ) ); + inputT += 8; + } + t_s64x1 = vadd_s64( vget_low_s64( t_s64x2 ), vget_high_s64( t_s64x2 ) ); + corr_QC_orderT = vget_lane_s64( t_s64x1, 0 ); + for( ; n < length; n++ ) { + corr_QC_orderT += silk_SMULL( input[ n ], input[ n ] ); + } + corr_QC_orderT = silk_LSHIFT64( corr_QC_orderT, QC ); + corr_QC[ orderT ] = corr_QC_orderT; + } + + corr_QCT = corr_QC + orderT - order; + lsh = silk_CLZ64( corr_QC_orderT ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + lsh_s64x2 = vdupq_n_s64( lsh ); + for( i = 0; i <= order - 3; i += 4 ) { + int32x4_t corr_s32x4; + int64x2_t corr_QC0_s64x2, corr_QC1_s64x2; + corr_QC0_s64x2 = vld1q_s64( corr_QCT + i ); + corr_QC1_s64x2 = vld1q_s64( corr_QCT + i + 2 ); + corr_QC0_s64x2 = vshlq_s64( corr_QC0_s64x2, lsh_s64x2 ); + corr_QC1_s64x2 = vshlq_s64( corr_QC1_s64x2, lsh_s64x2 ); + corr_s32x4 = vcombine_s32( vmovn_s64( corr_QC1_s64x2 ), vmovn_s64( corr_QC0_s64x2 ) ); + corr_s32x4 = vrev64q_s32( corr_s32x4 ); + vst1q_s32( corr + order - i - 3, corr_s32x4 ); + } + if( lsh >= 0 ) { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_LSHIFT64( corr_QCT[ i ], lsh ) ); + } + } else { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr_QCT[ i ], -lsh ) ); + } + } + silk_assert( corr_QCT[ order ] >= 0 ); /* If breaking, decrease QC*/ + RESTORE_STACK; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 corr_c[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int scale_c; + silk_warped_autocorrelation_FIX_c( corr_c, &scale_c, input, warping_Q16, length, order ); + silk_assert( !memcmp( corr_c, corr, sizeof( corr_c[ 0 ] ) * ( order + 1 ) ) ); + silk_assert( scale_c == *scale ); + } +#endif +} diff --git a/code/opus-1.2.1/silk/fixed/autocorr_FIX.c b/code/opus-1.5.2/silk/fixed/autocorr_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/autocorr_FIX.c rename to code/opus-1.5.2/silk/fixed/autocorr_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/burg_modified_FIX.c b/code/opus-1.5.2/silk/fixed/burg_modified_FIX.c similarity index 97% rename from code/opus-1.2.1/silk/fixed/burg_modified_FIX.c rename to code/opus-1.5.2/silk/fixed/burg_modified_FIX.c index 80edbdfa..185a12b1 100644 --- a/code/opus-1.2.1/silk/fixed/burg_modified_FIX.c +++ b/code/opus-1.5.2/silk/fixed/burg_modified_FIX.c @@ -65,10 +65,10 @@ void silk_burg_modified_c( opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; opus_int64 C0_64; - silk_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); /* Compute autocorrelations, added over subframes */ - C0_64 = silk_inner_prod16_aligned_64( x, x, subfr_length*nb_subfr, arch ); + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); lz = silk_CLZ64(C0_64); rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; @@ -87,7 +87,7 @@ void silk_burg_modified_c( x_ptr = x + s * subfr_length; for( n = 1; n < D + 1; n++ ) { C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( - silk_inner_prod16_aligned_64( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); } } } else { @@ -150,7 +150,7 @@ void silk_burg_modified_c( C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ - /* We sometimes have get overflows in the multiplications (even beyond +/- 2^32), + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), but they cancel each other and the real result seems to always fit in a 32-bit signed integer. This was determined experimentally, not theoretically (unfortunately). */ tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ @@ -253,7 +253,7 @@ void silk_burg_modified_c( if( rshifts > 0 ) { for( s = 0; s < nb_subfr; s++ ) { x_ptr = x + s * subfr_length; - C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16_aligned_64( x_ptr, x_ptr, D, arch ), rshifts ); + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); } } else { for( s = 0; s < nb_subfr; s++ ) { diff --git a/code/opus-1.2.1/silk/fixed/corrMatrix_FIX.c b/code/opus-1.5.2/silk/fixed/corrMatrix_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/corrMatrix_FIX.c rename to code/opus-1.5.2/silk/fixed/corrMatrix_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/encode_frame_FIX.c b/code/opus-1.5.2/silk/fixed/encode_frame_FIX.c similarity index 94% rename from code/opus-1.2.1/silk/fixed/encode_frame_FIX.c rename to code/opus-1.5.2/silk/fixed/encode_frame_FIX.c index 4f9e0861..7c83360b 100644 --- a/code/opus-1.2.1/silk/fixed/encode_frame_FIX.c +++ b/code/opus-1.5.2/silk/fixed/encode_frame_FIX.c @@ -43,21 +43,28 @@ static OPUS_INLINE void silk_LBRR_encode_FIX( ); void silk_encode_do_VAD_FIX( - silk_encoder_state_FIX *psEnc /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ ) { + const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); + /****************************/ /* Voice Activity Detection */ /****************************/ silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); + /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ + if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { + psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; + } /**************************************************/ /* Convert speech activity into VAD and DTX flags */ /**************************************************/ - if( psEnc->sCmn.speech_activity_Q8 < SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ) ) { + if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; psEnc->sCmn.noSpeechCounter++; - if( psEnc->sCmn.noSpeechCounter < NB_SPEECH_FRAMES_BEFORE_DTX ) { + if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { psEnc->sCmn.inDTX = 0; } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; @@ -98,8 +105,11 @@ opus_int silk_encode_frame_FIX( opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; opus_int best_sum[ MAX_NB_SUBFR ]; + opus_int bits_margin; SAVE_STACK; + /* For CBR, 5 bits below budget is close enough. For VBR, allow up to 25% below the cap if we initially busted the budget. */ + bits_margin = useCBR ? 5 : maxBits/4; /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; @@ -255,7 +265,7 @@ opus_int silk_encode_frame_FIX( if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { /* Restore output state from earlier iteration that did meet the bitrate budget */ silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); - silk_assert( sRangeEnc_copy2.offs <= 1275 ); + celt_assert( sRangeEnc_copy2.offs <= 1275 ); silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) ); psEnc->sShape.LastGainIndex = LastGainIndex_copy2; @@ -275,7 +285,7 @@ opus_int silk_encode_frame_FIX( gainMult_upper = gainMult_Q8; gainsID_upper = gainsID; } - } else if( nBits < maxBits - 5 ) { + } else if( nBits < maxBits - bits_margin ) { found_lower = 1; nBits_lower = nBits; gainMult_lower = gainMult_Q8; @@ -283,13 +293,13 @@ opus_int silk_encode_frame_FIX( gainsID_lower = gainsID; /* Copy part of the output state */ silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); - silk_assert( psRangeEnc->offs <= 1275 ); + celt_assert( psRangeEnc->offs <= 1275 ); silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; } } else { - /* Within 5 bits of budget: close enough */ + /* Close enough */ break; } @@ -311,17 +321,10 @@ opus_int silk_encode_frame_FIX( if( ( found_lower & found_upper ) == 0 ) { /* Adjust gain according to high-rate rate/distortion curve */ if( nBits > maxBits ) { - if (gainMult_Q8 < 16384) { - gainMult_Q8 *= 2; - } else { - gainMult_Q8 = 32767; - } + gainMult_Q8 = silk_min_32( 1024, gainMult_Q8*3/2 ); } else { - opus_int32 gain_factor_Q16; - gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7 ) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) ); - gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 ); + gainMult_Q8 = silk_max_32( 64, gainMult_Q8*4/5 ); } - } else { /* Adjust gain by interpolating */ gainMult_Q8 = gainMult_lower + silk_DIV32_16( silk_MUL( gainMult_upper - gainMult_lower, maxBits - nBits_lower ), nBits_upper - nBits_lower ); diff --git a/code/opus-1.2.1/silk/fixed/find_LPC_FIX.c b/code/opus-1.5.2/silk/fixed/find_LPC_FIX.c similarity index 99% rename from code/opus-1.2.1/silk/fixed/find_LPC_FIX.c rename to code/opus-1.5.2/silk/fixed/find_LPC_FIX.c index e55b63ac..c762a0f2 100644 --- a/code/opus-1.2.1/silk/fixed/find_LPC_FIX.c +++ b/code/opus-1.5.2/silk/fixed/find_LPC_FIX.c @@ -146,6 +146,6 @@ void silk_find_LPC_FIX( silk_A2NLSF( NLSF_Q15, a_Q16, psEncC->predictLPCOrder ); } - silk_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); + celt_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); RESTORE_STACK; } diff --git a/code/opus-1.2.1/silk/fixed/find_LTP_FIX.c b/code/opus-1.5.2/silk/fixed/find_LTP_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/find_LTP_FIX.c rename to code/opus-1.5.2/silk/fixed/find_LTP_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/find_pitch_lags_FIX.c b/code/opus-1.5.2/silk/fixed/find_pitch_lags_FIX.c similarity index 99% rename from code/opus-1.2.1/silk/fixed/find_pitch_lags_FIX.c rename to code/opus-1.5.2/silk/fixed/find_pitch_lags_FIX.c index 9303e9db..6c3379f2 100644 --- a/code/opus-1.2.1/silk/fixed/find_pitch_lags_FIX.c +++ b/code/opus-1.5.2/silk/fixed/find_pitch_lags_FIX.c @@ -59,7 +59,7 @@ void silk_find_pitch_lags_FIX( buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; /* Safety check */ - silk_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); + celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); /*************************************/ /* Estimate LPC AR coefficients */ diff --git a/code/opus-1.2.1/silk/fixed/find_pred_coefs_FIX.c b/code/opus-1.5.2/silk/fixed/find_pred_coefs_FIX.c similarity index 97% rename from code/opus-1.2.1/silk/fixed/find_pred_coefs_FIX.c rename to code/opus-1.5.2/silk/fixed/find_pred_coefs_FIX.c index 24c6aab3..ad363fb7 100644 --- a/code/opus-1.2.1/silk/fixed/find_pred_coefs_FIX.c +++ b/code/opus-1.5.2/silk/fixed/find_pred_coefs_FIX.c @@ -42,7 +42,8 @@ void silk_find_pred_coefs_FIX( { opus_int i; opus_int32 invGains_Q16[ MAX_NB_SUBFR ], local_gains[ MAX_NB_SUBFR ]; - opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]; + /* Set to NLSF_Q15 to zero so we don't copy junk to the state. */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]={0}; const opus_int16 *x_ptr; opus_int16 *x_pre_ptr; VARDECL( opus_int16, LPC_in_pre ); @@ -80,7 +81,7 @@ void silk_find_pred_coefs_FIX( /**********/ /* VOICED */ /**********/ - silk_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + celt_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); ALLOC( xXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER, opus_int32 ); ALLOC( XXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER * LTP_ORDER, opus_int32 ); diff --git a/code/opus-1.2.1/silk/fixed/k2a_FIX.c b/code/opus-1.5.2/silk/fixed/k2a_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/k2a_FIX.c rename to code/opus-1.5.2/silk/fixed/k2a_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/k2a_Q16_FIX.c b/code/opus-1.5.2/silk/fixed/k2a_Q16_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/k2a_Q16_FIX.c rename to code/opus-1.5.2/silk/fixed/k2a_Q16_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/main_FIX.h b/code/opus-1.5.2/silk/fixed/main_FIX.h similarity index 98% rename from code/opus-1.2.1/silk/fixed/main_FIX.h rename to code/opus-1.5.2/silk/fixed/main_FIX.h index 780afa39..6d2112e5 100644 --- a/code/opus-1.2.1/silk/fixed/main_FIX.h +++ b/code/opus-1.5.2/silk/fixed/main_FIX.h @@ -66,7 +66,8 @@ void silk_HP_variable_cutoff( /* Encoder main function */ void silk_encode_do_VAD_FIX( - silk_encoder_state_FIX *psEnc /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ ); /* Encoder main function */ diff --git a/code/opus-1.5.2/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h b/code/opus-1.5.2/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h new file mode 100644 index 00000000..3999b5bd --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h @@ -0,0 +1,336 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + + +/**************************************************************/ +/* Compute noise shaping coefficients and initial gain values */ +/**************************************************************/ +#define OVERRIDE_silk_noise_shape_analysis_FIX + +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k, i, nSamples, Qnrg, b_Q14, warping_Q16, scale = 0; + opus_int32 SNR_adj_dB_Q7, HarmBoost_Q16, HarmShapeGain_Q16, Tilt_Q16, tmp32; + opus_int32 nrg, pre_nrg_Q30, log_energy_Q7, log_energy_prev_Q7, energy_variation_Q7; + opus_int32 delta_Q16, BWExp1_Q16, BWExp2_Q16, gain_mult_Q16, gain_add_Q16, strength_Q16, b_Q8; + opus_int32 auto_corr[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int32 refl_coef_Q16[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR1_Q24[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR2_Q24[ MAX_SHAPE_LPC_ORDER ]; + VARDECL( opus_int16, x_windowed ); + const opus_int16 *x_ptr, *pitch_res_ptr; + SAVE_STACK; + + /* Point to start of first LPC analysis block */ + x_ptr = x - psEnc->sCmn.la_shape; + + /****************/ + /* GAIN CONTROL */ + /****************/ + SNR_adj_dB_Q7 = psEnc->sCmn.SNR_dB_Q7; + + /* Input quality is the average of the quality in the lowest two VAD bands */ + psEncCtrl->input_quality_Q14 = ( opus_int )silk_RSHIFT( (opus_int32)psEnc->sCmn.input_quality_bands_Q15[ 0 ] + + psEnc->sCmn.input_quality_bands_Q15[ 1 ], 2 ); + + /* Coding quality level, between 0.0_Q0 and 1.0_Q0, but in Q14 */ + psEncCtrl->coding_quality_Q14 = silk_RSHIFT( silk_sigm_Q15( silk_RSHIFT_ROUND( SNR_adj_dB_Q7 - + SILK_FIX_CONST( 20.0, 7 ), 4 ) ), 1 ); + + /* Reduce coding SNR during low speech activity */ + if( psEnc->sCmn.useCBR == 0 ) { + b_Q8 = SILK_FIX_CONST( 1.0, 8 ) - psEnc->sCmn.speech_activity_Q8; + b_Q8 = silk_SMULWB( silk_LSHIFT( b_Q8, 8 ), b_Q8 ); + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMULBB( SILK_FIX_CONST( -BG_SNR_DECR_dB, 7 ) >> ( 4 + 1 ), b_Q8 ), /* Q11*/ + silk_SMULWB( SILK_FIX_CONST( 1.0, 14 ) + psEncCtrl->input_quality_Q14, psEncCtrl->coding_quality_Q14 ) ); /* Q12*/ + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce gains for periodic signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( HARM_SNR_INCR_dB, 8 ), psEnc->LTPCorr_Q15 ); + } else { + /* For unvoiced signals and low-quality input, adjust the quality slower than SNR_dB setting */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMLAWB( SILK_FIX_CONST( 6.0, 9 ), -SILK_FIX_CONST( 0.4, 18 ), psEnc->sCmn.SNR_dB_Q7 ), + SILK_FIX_CONST( 1.0, 14 ) - psEncCtrl->input_quality_Q14 ); + } + + /*************************/ + /* SPARSENESS PROCESSING */ + /*************************/ + /* Set quantizer offset */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Initially set to 0; may be overruled in process_gains(..) */ + psEnc->sCmn.indices.quantOffsetType = 0; + psEncCtrl->sparseness_Q8 = 0; + } else { + /* Sparseness measure, based on relative fluctuations of energy per 2 milliseconds */ + nSamples = silk_LSHIFT( psEnc->sCmn.fs_kHz, 1 ); + energy_variation_Q7 = 0; + log_energy_prev_Q7 = 0; + pitch_res_ptr = pitch_res; + for( k = 0; k < silk_SMULBB( SUB_FRAME_LENGTH_MS, psEnc->sCmn.nb_subfr ) / 2; k++ ) { + silk_sum_sqr_shift( &nrg, &scale, pitch_res_ptr, nSamples ); + nrg += silk_RSHIFT( nSamples, scale ); /* Q(-scale)*/ + + log_energy_Q7 = silk_lin2log( nrg ); + if( k > 0 ) { + energy_variation_Q7 += silk_abs( log_energy_Q7 - log_energy_prev_Q7 ); + } + log_energy_prev_Q7 = log_energy_Q7; + pitch_res_ptr += nSamples; + } + + psEncCtrl->sparseness_Q8 = silk_RSHIFT( silk_sigm_Q15( silk_SMULWB( energy_variation_Q7 - + SILK_FIX_CONST( 5.0, 7 ), SILK_FIX_CONST( 0.1, 16 ) ) ), 7 ); + + /* Set quantization offset depending on sparseness measure */ + if( psEncCtrl->sparseness_Q8 > SILK_FIX_CONST( SPARSENESS_THRESHOLD_QNT_OFFSET, 8 ) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + + /* Increase coding SNR for sparse signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( SPARSE_SNR_INCR_dB, 15 ), psEncCtrl->sparseness_Q8 - SILK_FIX_CONST( 0.5, 8 ) ); + } + + /*******************************/ + /* Control bandwidth expansion */ + /*******************************/ + /* More BWE for signals with high prediction gain */ + strength_Q16 = silk_SMULWB( psEncCtrl->predGain_Q16, SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ); + BWExp1_Q16 = BWExp2_Q16 = silk_DIV32_varQ( SILK_FIX_CONST( BANDWIDTH_EXPANSION, 16 ), + silk_SMLAWW( SILK_FIX_CONST( 1.0, 16 ), strength_Q16, strength_Q16 ), 16 ); + delta_Q16 = silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - silk_SMULBB( 3, psEncCtrl->coding_quality_Q14 ), + SILK_FIX_CONST( LOW_RATE_BANDWIDTH_EXPANSION_DELTA, 16 ) ); + BWExp1_Q16 = silk_SUB32( BWExp1_Q16, delta_Q16 ); + BWExp2_Q16 = silk_ADD32( BWExp2_Q16, delta_Q16 ); + /* BWExp1 will be applied after BWExp2, so make it relative */ + BWExp1_Q16 = silk_DIV32_16( silk_LSHIFT( BWExp1_Q16, 14 ), silk_RSHIFT( BWExp2_Q16, 2 ) ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Slightly more warping in analysis will move quantization noise up in frequency, where it's better masked */ + warping_Q16 = silk_SMLAWB( psEnc->sCmn.warping_Q16, (opus_int32)psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( 0.01, 18 ) ); + } else { + warping_Q16 = 0; + } + + /********************************************/ + /* Compute noise shaping AR coefs and gains */ + /********************************************/ + ALLOC( x_windowed, psEnc->sCmn.shapeWinLength, opus_int16 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Apply window: sine slope followed by flat part followed by cosine slope */ + opus_int shift, slope_part, flat_part; + flat_part = psEnc->sCmn.fs_kHz * 3; + slope_part = silk_RSHIFT( psEnc->sCmn.shapeWinLength - flat_part, 1 ); + + silk_apply_sine_window( x_windowed, x_ptr, 1, slope_part ); + shift = slope_part; + silk_memcpy( x_windowed + shift, x_ptr + shift, flat_part * sizeof(opus_int16) ); + shift += flat_part; + silk_apply_sine_window( x_windowed + shift, x_ptr + shift, 2, slope_part ); + + /* Update pointer: next LPC analysis block */ + x_ptr += psEnc->sCmn.subfr_length; + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Calculate warped auto correlation */ + silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder, arch ); + } else { + /* Calculate regular auto correlation */ + silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch ); + } + + /* Add white noise, as a fraction of energy */ + auto_corr[0] = silk_ADD32( auto_corr[0], silk_max_32( silk_SMULWB( silk_RSHIFT( auto_corr[ 0 ], 4 ), + SILK_FIX_CONST( SHAPE_WHITE_NOISE_FRACTION, 20 ) ), 1 ) ); + + /* Calculate the reflection coefficients using schur */ + nrg = silk_schur64( refl_coef_Q16, auto_corr, psEnc->sCmn.shapingLPCOrder ); + silk_assert( nrg >= 0 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a_Q16( AR2_Q24, refl_coef_Q16, psEnc->sCmn.shapingLPCOrder ); + + Qnrg = -scale; /* range: -12...30*/ + silk_assert( Qnrg >= -12 ); + silk_assert( Qnrg <= 30 ); + + /* Make sure that Qnrg is an even number */ + if( Qnrg & 1 ) { + Qnrg -= 1; + nrg >>= 1; + } + + tmp32 = silk_SQRT_APPROX( nrg ); + Qnrg >>= 1; /* range: -6...15*/ + + psEncCtrl->Gains_Q16[ k ] = (silk_LSHIFT32( silk_LIMIT( (tmp32), silk_RSHIFT32( silk_int32_MIN, (16 - Qnrg) ), \ + silk_RSHIFT32( silk_int32_MAX, (16 - Qnrg) ) ), (16 - Qnrg) )); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Adjust gain for warping */ + gain_mult_Q16 = warped_gain( AR2_Q24, warping_Q16, psEnc->sCmn.shapingLPCOrder ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + if ( silk_SMULWW( silk_RSHIFT_ROUND( psEncCtrl->Gains_Q16[ k ], 1 ), gain_mult_Q16 ) >= ( silk_int32_MAX >> 1 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_int32_MAX; + } else { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + } + } + + /* Bandwidth expansion for synthesis filter shaping */ + silk_bwexpander_32( AR2_Q24, psEnc->sCmn.shapingLPCOrder, BWExp2_Q16 ); + + /* Compute noise shaping filter coefficients */ + silk_memcpy( AR1_Q24, AR2_Q24, psEnc->sCmn.shapingLPCOrder * sizeof( opus_int32 ) ); + + /* Bandwidth expansion for analysis filter shaping */ + silk_assert( BWExp1_Q16 <= SILK_FIX_CONST( 1.0, 16 ) ); + silk_bwexpander_32( AR1_Q24, psEnc->sCmn.shapingLPCOrder, BWExp1_Q16 ); + + /* Ratio of prediction gains, in energy domain */ + pre_nrg_Q30 = silk_LPC_inverse_pred_gain_Q24( AR2_Q24, psEnc->sCmn.shapingLPCOrder, arch ); + nrg = silk_LPC_inverse_pred_gain_Q24( AR1_Q24, psEnc->sCmn.shapingLPCOrder, arch ); + + /*psEncCtrl->GainsPre[ k ] = 1.0f - 0.7f * ( 1.0f - pre_nrg / nrg ) = 0.3f + 0.7f * pre_nrg / nrg;*/ + pre_nrg_Q30 = silk_LSHIFT32( silk_SMULWB( pre_nrg_Q30, SILK_FIX_CONST( 0.7, 15 ) ), 1 ); + psEncCtrl->GainsPre_Q14[ k ] = ( opus_int ) SILK_FIX_CONST( 0.3, 14 ) + silk_DIV32_varQ( pre_nrg_Q30, nrg, 14 ); + + /* Convert to monic warped prediction coefficients and limit absolute values */ + limit_warped_coefs( AR2_Q24, AR1_Q24, warping_Q16, SILK_FIX_CONST( 3.999, 24 ), psEnc->sCmn.shapingLPCOrder ); + + /* Convert from Q24 to Q13 and store in int16 */ + for( i = 0; i < psEnc->sCmn.shapingLPCOrder; i++ ) { + psEncCtrl->AR1_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR1_Q24[ i ], 11 ) ); + psEncCtrl->AR2_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR2_Q24[ i ], 11 ) ); + } + } + + /*****************/ + /* Gain tweaking */ + /*****************/ + /* Increase gains during low speech activity and put lower limit on gains */ + gain_mult_Q16 = silk_log2lin( -silk_SMLAWB( -SILK_FIX_CONST( 16.0, 7 ), SNR_adj_dB_Q7, SILK_FIX_CONST( 0.16, 16 ) ) ); + gain_add_Q16 = silk_log2lin( silk_SMLAWB( SILK_FIX_CONST( 16.0, 7 ), SILK_FIX_CONST( MIN_QGAIN_DB, 7 ), SILK_FIX_CONST( 0.16, 16 ) ) ); + silk_assert( gain_mult_Q16 > 0 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + psEncCtrl->Gains_Q16[ k ] = silk_ADD_POS_SAT32( psEncCtrl->Gains_Q16[ k ], gain_add_Q16 ); + } + + gain_mult_Q16 = SILK_FIX_CONST( 1.0, 16 ) + silk_RSHIFT_ROUND( silk_MLA( SILK_FIX_CONST( INPUT_TILT, 26 ), + psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( HIGH_RATE_INPUT_TILT, 12 ) ), 10 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->GainsPre_Q14[ k ] = silk_SMULWB( gain_mult_Q16, psEncCtrl->GainsPre_Q14[ k ] ); + } + + /************************************************/ + /* Control low-frequency shaping and noise tilt */ + /************************************************/ + /* Less low frequency shaping for noisy inputs */ + strength_Q16 = silk_MUL( SILK_FIX_CONST( LOW_FREQ_SHAPING, 4 ), silk_SMLAWB( SILK_FIX_CONST( 1.0, 12 ), + SILK_FIX_CONST( LOW_QUALITY_LOW_FREQ_SHAPING_DECR, 13 ), psEnc->sCmn.input_quality_bands_Q15[ 0 ] - SILK_FIX_CONST( 1.0, 15 ) ) ); + strength_Q16 = silk_RSHIFT( silk_MUL( strength_Q16, psEnc->sCmn.speech_activity_Q8 ), 8 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce low frequencies quantization noise for periodic signals, depending on pitch lag */ + /*f = 400; freqz([1, -0.98 + 2e-4 * f], [1, -0.97 + 7e-4 * f], 2^12, Fs); axis([0, 1000, -10, 1])*/ + opus_int fs_kHz_inv = silk_DIV32_16( SILK_FIX_CONST( 0.2, 14 ), psEnc->sCmn.fs_kHz ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + b_Q14 = fs_kHz_inv + silk_DIV32_16( SILK_FIX_CONST( 3.0, 14 ), psEncCtrl->pitchL[ k ] ); + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ k ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - silk_SMULWB( strength_Q16, b_Q14 ), 16 ); + psEncCtrl->LF_shp_Q14[ k ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + } + silk_assert( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ) < SILK_FIX_CONST( 0.5, 24 ) ); /* Guarantees that second argument to SMULWB() is within range of an opus_int16*/ + Tilt_Q16 = - SILK_FIX_CONST( HP_NOISE_COEF, 16 ) - + silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - SILK_FIX_CONST( HP_NOISE_COEF, 16 ), + silk_SMULWB( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ), psEnc->sCmn.speech_activity_Q8 ) ); + } else { + b_Q14 = silk_DIV32_16( 21299, psEnc->sCmn.fs_kHz ); /* 1.3_Q0 = 21299_Q14*/ + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ 0 ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - + silk_SMULWB( strength_Q16, silk_SMULWB( SILK_FIX_CONST( 0.6, 16 ), b_Q14 ) ), 16 ); + psEncCtrl->LF_shp_Q14[ 0 ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + for( k = 1; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->LF_shp_Q14[ k ] = psEncCtrl->LF_shp_Q14[ 0 ]; + } + Tilt_Q16 = -SILK_FIX_CONST( HP_NOISE_COEF, 16 ); + } + + /****************************/ + /* HARMONIC SHAPING CONTROL */ + /****************************/ + /* Control boosting of harmonic frequencies */ + HarmBoost_Q16 = silk_SMULWB( silk_SMULWB( SILK_FIX_CONST( 1.0, 17 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 3 ), + psEnc->LTPCorr_Q15 ), SILK_FIX_CONST( LOW_RATE_HARMONIC_BOOST, 16 ) ); + + /* More harmonic boost for noisy input signals */ + HarmBoost_Q16 = silk_SMLAWB( HarmBoost_Q16, + SILK_FIX_CONST( 1.0, 16 ) - silk_LSHIFT( psEncCtrl->input_quality_Q14, 2 ), SILK_FIX_CONST( LOW_INPUT_QUALITY_HARMONIC_BOOST, 16 ) ); + + if( USE_HARM_SHAPING && psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* More harmonic noise shaping for high bitrates or noisy input */ + HarmShapeGain_Q16 = silk_SMLAWB( SILK_FIX_CONST( HARMONIC_SHAPING, 16 ), + SILK_FIX_CONST( 1.0, 16 ) - silk_SMULWB( SILK_FIX_CONST( 1.0, 18 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 4 ), + psEncCtrl->input_quality_Q14 ), SILK_FIX_CONST( HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING, 16 ) ); + + /* Less harmonic noise shaping for less periodic signals */ + HarmShapeGain_Q16 = silk_SMULWB( silk_LSHIFT( HarmShapeGain_Q16, 1 ), + silk_SQRT_APPROX( silk_LSHIFT( psEnc->LTPCorr_Q15, 15 ) ) ); + } else { + HarmShapeGain_Q16 = 0; + } + + /*************************/ + /* Smooth over subframes */ + /*************************/ + for( k = 0; k < MAX_NB_SUBFR; k++ ) { + psShapeSt->HarmBoost_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmBoost_smth_Q16, HarmBoost_Q16 - psShapeSt->HarmBoost_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->HarmShapeGain_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmShapeGain_smth_Q16, HarmShapeGain_Q16 - psShapeSt->HarmShapeGain_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->Tilt_smth_Q16 = + silk_SMLAWB( psShapeSt->Tilt_smth_Q16, Tilt_Q16 - psShapeSt->Tilt_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + + psEncCtrl->HarmBoost_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmBoost_smth_Q16, 2 ); + psEncCtrl->HarmShapeGain_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmShapeGain_smth_Q16, 2 ); + psEncCtrl->Tilt_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->Tilt_smth_Q16, 2 ); + } + RESTORE_STACK; +} diff --git a/code/opus-1.5.2/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h b/code/opus-1.5.2/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h new file mode 100644 index 00000000..66eb2ed2 --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h @@ -0,0 +1,165 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ +#define __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +#undef QC +#define QC 10 + +#undef QS +#define QS 14 + +/* Autocorrelations for a warped frequency axis */ +#define OVERRIDE_silk_warped_autocorrelation_FIX_c +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i, lsh; + opus_int32 tmp1_QS=0, tmp2_QS=0, tmp3_QS=0, tmp4_QS=0, tmp5_QS=0, tmp6_QS=0, tmp7_QS=0, tmp8_QS=0, start_1=0, start_2=0, start_3=0; + opus_int32 state_QS[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 temp64; + + opus_int32 val; + val = 2 * QS - QC; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n=n+4 ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + start_1 = tmp1_QS; + tmp3_QS = silk_LSHIFT32( (opus_int32)input[ n+1], QS ); + start_2 = tmp3_QS; + tmp5_QS = silk_LSHIFT32( (opus_int32)input[ n+2], QS ); + start_3 = tmp5_QS; + tmp7_QS = silk_LSHIFT32( (opus_int32)input[ n+3], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp1_QS, start_1); + + tmp4_QS = silk_SMLAWB( tmp1_QS, tmp2_QS - tmp3_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp3_QS, start_2); + + tmp6_QS = silk_SMLAWB( tmp3_QS, tmp4_QS - tmp5_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp5_QS, start_3); + + tmp8_QS = silk_SMLAWB( tmp5_QS, tmp6_QS - tmp7_QS, warping_Q16 ); + state_QS[ i ] = tmp7_QS; + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp7_QS, state_QS[0]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp2_QS, start_1); + + tmp3_QS = silk_SMLAWB( tmp2_QS, tmp1_QS - tmp4_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp4_QS, start_2); + + tmp5_QS = silk_SMLAWB( tmp4_QS, tmp3_QS - tmp6_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp6_QS, start_3); + + tmp7_QS = silk_SMLAWB( tmp6_QS, tmp5_QS - tmp8_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp8_QS; + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp8_QS, state_QS[ 0 ]); + + } + state_QS[ order ] = tmp7_QS; + + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp1_QS, start_1); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp3_QS, start_2); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp5_QS, start_3); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp7_QS, state_QS[ 0 ]); + } + + for(;n< length; n++ ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + state_QS[ i ] = tmp1_QS; + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp1_QS, state_QS[ 0 ]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp2_QS; + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp2_QS, state_QS[ 0 ]); + } + state_QS[ order ] = tmp1_QS; + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp1_QS, state_QS[ 0 ]); + } + + temp64 = corr_QC[ 0 ]; + temp64 = __builtin_mips_shilo(temp64, val); + + lsh = silk_CLZ64( temp64 ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + if( lsh >= 0 ) { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = __builtin_mips_shilo(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( __builtin_mips_shilo( temp64, -lsh ) ); + } + } else { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = __builtin_mips_shilo(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( __builtin_mips_shilo( temp64, -lsh ) ); + } + } + + corr_QC[ 0 ] = __builtin_mips_shilo(corr_QC[ 0 ], val); + + silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ +} +#endif /* __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/silk/fixed/noise_shape_analysis_FIX.c b/code/opus-1.5.2/silk/fixed/noise_shape_analysis_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/noise_shape_analysis_FIX.c rename to code/opus-1.5.2/silk/fixed/noise_shape_analysis_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/pitch_analysis_core_FIX.c b/code/opus-1.5.2/silk/fixed/pitch_analysis_core_FIX.c similarity index 96% rename from code/opus-1.2.1/silk/fixed/pitch_analysis_core_FIX.c rename to code/opus-1.5.2/silk/fixed/pitch_analysis_core_FIX.c index 8df109e6..14729046 100644 --- a/code/opus-1.2.1/silk/fixed/pitch_analysis_core_FIX.c +++ b/code/opus-1.5.2/silk/fixed/pitch_analysis_core_FIX.c @@ -122,11 +122,11 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 SAVE_STACK; /* Check for valid sampling frequency */ - silk_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); + celt_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); /* Check for valid complexity setting */ - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); silk_assert( search_thres1_Q16 >= 0 && search_thres1_Q16 <= (1<<16) ); silk_assert( search_thres2_Q13 >= 0 && search_thres2_Q13 <= (1<<13) ); @@ -164,7 +164,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 silk_resampler_down2_3( filt_state, frame_8kHz_buf, frame, frame_length ); frame_8kHz = frame_8kHz_buf; } else { - silk_assert( Fs_kHz == 8 ); + celt_assert( Fs_kHz == 8 ); frame_8kHz = frame; } @@ -188,14 +188,14 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 target_ptr = &frame_4kHz[ silk_LSHIFT( SF_LENGTH_4KHZ, 2 ) ]; for( k = 0; k < nb_subfr >> 1; k++ ) { /* Check that we are within range of the array */ - silk_assert( target_ptr >= frame_4kHz ); - silk_assert( target_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + celt_assert( target_ptr >= frame_4kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); basis_ptr = target_ptr - MIN_LAG_4KHZ; /* Check that we are within range of the array */ - silk_assert( basis_ptr >= frame_4kHz ); - silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + celt_assert( basis_ptr >= frame_4kHz ); + celt_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1, arch ); @@ -249,7 +249,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 /* Sort */ length_d_srch = silk_ADD_LSHIFT32( 4, complexity, 1 ); - silk_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); + celt_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); silk_insertion_sort_decreasing_int16( C, d_srch, CSTRIDE_4KHZ, length_d_srch ); @@ -274,7 +274,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 break; } } - silk_assert( length_d_srch > 0 ); + celt_assert( length_d_srch > 0 ); ALLOC( d_comp, D_COMP_STRIDE, opus_int16 ); for( i = D_COMP_MIN; i < D_COMP_MAX; i++ ) { @@ -325,8 +325,8 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 for( k = 0; k < nb_subfr; k++ ) { /* Check that we are within range of the array */ - silk_assert( target_ptr >= frame_8kHz ); - silk_assert( target_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); + celt_assert( target_ptr >= frame_8kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ, arch ), 1 ); for( j = 0; j < length_d_comp; j++ ) { @@ -550,7 +550,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 *lagIndex = (opus_int16)( lag - MIN_LAG_8KHZ ); *contourIndex = (opus_int8)CBimax; } - silk_assert( *lagIndex >= 0 ); + celt_assert( *lagIndex >= 0 ); /* return as voiced */ RESTORE_STACK; return 0; @@ -587,8 +587,8 @@ static void silk_P_Ana_calc_corr_st3( const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; SAVE_STACK; - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); if( nb_subfr == PE_MAX_NB_SUBFR ) { Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; @@ -596,7 +596,7 @@ static void silk_P_Ana_calc_corr_st3( nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; cbk_size = PE_NB_CBKS_STAGE3_MAX; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; @@ -612,7 +612,7 @@ static void silk_P_Ana_calc_corr_st3( /* Calculate the correlations for each subframe */ lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); - silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1, arch ); for( j = lag_low; j <= lag_high; j++ ) { silk_assert( lag_counter < SCRATCH_SIZE ); @@ -659,8 +659,8 @@ static void silk_P_Ana_calc_energy_st3( const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; SAVE_STACK; - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); if( nb_subfr == PE_MAX_NB_SUBFR ) { Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; @@ -668,7 +668,7 @@ static void silk_P_Ana_calc_energy_st3( nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; cbk_size = PE_NB_CBKS_STAGE3_MAX; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; diff --git a/code/opus-1.2.1/silk/fixed/process_gains_FIX.c b/code/opus-1.5.2/silk/fixed/process_gains_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/process_gains_FIX.c rename to code/opus-1.5.2/silk/fixed/process_gains_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/regularize_correlations_FIX.c b/code/opus-1.5.2/silk/fixed/regularize_correlations_FIX.c similarity index 100% rename from code/opus-1.2.1/silk/fixed/regularize_correlations_FIX.c rename to code/opus-1.5.2/silk/fixed/regularize_correlations_FIX.c diff --git a/code/opus-1.2.1/silk/fixed/residual_energy16_FIX.c b/code/opus-1.5.2/silk/fixed/residual_energy16_FIX.c similarity index 97% rename from code/opus-1.2.1/silk/fixed/residual_energy16_FIX.c rename to code/opus-1.5.2/silk/fixed/residual_energy16_FIX.c index ebffb2a6..7f130f3d 100644 --- a/code/opus-1.2.1/silk/fixed/residual_energy16_FIX.c +++ b/code/opus-1.5.2/silk/fixed/residual_energy16_FIX.c @@ -47,10 +47,10 @@ opus_int32 silk_residual_energy16_covar_FIX( const opus_int32 *pRow; /* Safety checks */ - silk_assert( D >= 0 ); - silk_assert( D <= 16 ); - silk_assert( cQ > 0 ); - silk_assert( cQ < 16 ); + celt_assert( D >= 0 ); + celt_assert( D <= 16 ); + celt_assert( cQ > 0 ); + celt_assert( cQ < 16 ); lshifts = 16 - cQ; Qxtra = lshifts; diff --git a/code/opus-1.2.1/silk/fixed/residual_energy_FIX.c b/code/opus-1.5.2/silk/fixed/residual_energy_FIX.c similarity index 98% rename from code/opus-1.2.1/silk/fixed/residual_energy_FIX.c rename to code/opus-1.5.2/silk/fixed/residual_energy_FIX.c index 41f74778..6c7cade9 100644 --- a/code/opus-1.2.1/silk/fixed/residual_energy_FIX.c +++ b/code/opus-1.5.2/silk/fixed/residual_energy_FIX.c @@ -58,7 +58,7 @@ void silk_residual_energy_FIX( /* Filter input to create the LPC residual for each frame half, and measure subframe energies */ ALLOC( LPC_res, ( MAX_NB_SUBFR >> 1 ) * offset, opus_int16 ); - silk_assert( ( nb_subfr >> 1 ) * ( MAX_NB_SUBFR >> 1 ) == nb_subfr ); + celt_assert( ( nb_subfr >> 1 ) * ( MAX_NB_SUBFR >> 1 ) == nb_subfr ); for( i = 0; i < nb_subfr >> 1; i++ ) { /* Calculate half frame LPC residual signal including preceding samples */ silk_LPC_analysis_filter( LPC_res, x_ptr, a_Q12[ i ], ( MAX_NB_SUBFR >> 1 ) * offset, LPC_order, arch ); diff --git a/code/opus-1.2.1/silk/fixed/schur64_FIX.c b/code/opus-1.5.2/silk/fixed/schur64_FIX.c similarity index 98% rename from code/opus-1.2.1/silk/fixed/schur64_FIX.c rename to code/opus-1.5.2/silk/fixed/schur64_FIX.c index b2cb12d9..4b7e19ea 100644 --- a/code/opus-1.2.1/silk/fixed/schur64_FIX.c +++ b/code/opus-1.5.2/silk/fixed/schur64_FIX.c @@ -43,7 +43,7 @@ opus_int32 silk_schur64( /* O returns residual ene opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; opus_int32 Ctmp1_Q30, Ctmp2_Q30, rc_tmp_Q31; - silk_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); /* Check for invalid input */ if( c[ 0 ] <= 0 ) { diff --git a/code/opus-1.2.1/silk/fixed/schur_FIX.c b/code/opus-1.5.2/silk/fixed/schur_FIX.c similarity index 98% rename from code/opus-1.2.1/silk/fixed/schur_FIX.c rename to code/opus-1.5.2/silk/fixed/schur_FIX.c index 59d44a6f..2840f6b1 100644 --- a/code/opus-1.2.1/silk/fixed/schur_FIX.c +++ b/code/opus-1.5.2/silk/fixed/schur_FIX.c @@ -43,7 +43,7 @@ opus_int32 silk_schur( /* O Returns residual ene opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; opus_int32 Ctmp1, Ctmp2, rc_tmp_Q15; - silk_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); /* Get number of leading zeros */ lz = silk_CLZ32( c[ 0 ] ); diff --git a/code/opus-1.2.1/silk/fixed/structs_FIX.h b/code/opus-1.5.2/silk/fixed/structs_FIX.h similarity index 100% rename from code/opus-1.2.1/silk/fixed/structs_FIX.h rename to code/opus-1.5.2/silk/fixed/structs_FIX.h diff --git a/code/opus-1.2.1/silk/fixed/vector_ops_FIX.c b/code/opus-1.5.2/silk/fixed/vector_ops_FIX.c similarity index 99% rename from code/opus-1.2.1/silk/fixed/vector_ops_FIX.c rename to code/opus-1.5.2/silk/fixed/vector_ops_FIX.c index d9498001..dcf84070 100644 --- a/code/opus-1.2.1/silk/fixed/vector_ops_FIX.c +++ b/code/opus-1.5.2/silk/fixed/vector_ops_FIX.c @@ -87,7 +87,7 @@ opus_int32 silk_inner_prod_aligned( #endif } -opus_int64 silk_inner_prod16_aligned_64_c( +opus_int64 silk_inner_prod16_c( const opus_int16 *inVec1, /* I input vector 1 */ const opus_int16 *inVec2, /* I input vector 2 */ const opus_int len /* I vector lengths */ diff --git a/code/opus-1.2.1/silk/fixed/warped_autocorrelation_FIX.c b/code/opus-1.5.2/silk/fixed/warped_autocorrelation_FIX.c similarity index 96% rename from code/opus-1.2.1/silk/fixed/warped_autocorrelation_FIX.c rename to code/opus-1.5.2/silk/fixed/warped_autocorrelation_FIX.c index 994c299a..5c79553b 100644 --- a/code/opus-1.2.1/silk/fixed/warped_autocorrelation_FIX.c +++ b/code/opus-1.5.2/silk/fixed/warped_autocorrelation_FIX.c @@ -37,6 +37,7 @@ POSSIBILITY OF SUCH DAMAGE. /* Autocorrelations for a warped frequency axis */ +#ifndef OVERRIDE_silk_warped_autocorrelation_FIX_c void silk_warped_autocorrelation_FIX_c( opus_int32 *corr, /* O Result [order + 1] */ opus_int *scale, /* O Scaling of the correlation vector */ @@ -52,7 +53,7 @@ void silk_warped_autocorrelation_FIX_c( opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; /* Order must be even */ - silk_assert( ( order & 1 ) == 0 ); + celt_assert( ( order & 1 ) == 0 ); silk_assert( 2 * QS - QC >= 0 ); /* Loop over samples */ @@ -88,3 +89,4 @@ void silk_warped_autocorrelation_FIX_c( } silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ } +#endif /* OVERRIDE_silk_warped_autocorrelation_FIX_c */ diff --git a/code/opus-1.5.2/silk/fixed/x86/burg_modified_FIX_sse4_1.c b/code/opus-1.5.2/silk/fixed/x86/burg_modified_FIX_sse4_1.c new file mode 100644 index 00000000..e58bf079 --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/x86/burg_modified_FIX_sse4_1.c @@ -0,0 +1,400 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include + +#include "SigProc_FIX.h" +#include "define.h" +#include "tuning_parameters.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ + +#define QA 25 +#define N_BITS_HEAD_ROOM 3 +#define MIN_RSHIFTS -16 +#define MAX_RSHIFTS (32 - QA) + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_sse4_1( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) +{ + opus_int k, n, s, lz, rshifts, reached_max_gain; + opus_int32 C0, num, nrg, rc_Q31, invGain_Q30, Atmp_QA, Atmp1, tmp1, tmp2, x1, x2; + const opus_int16 *x_ptr; + opus_int32 C_first_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 C_last_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; + opus_int64 C0_64; + + __m128i FIRST_3210, LAST_3210, ATMP_3210, TMP1_3210, TMP2_3210, T1_3210, T2_3210, PTR_3210, SUBFR_3210, X1_3210, X2_3210; + __m128i CONST1 = _mm_set1_epi32(1); + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); + lz = silk_CLZ64(C0_64); + rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; + if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; + if (rshifts < MIN_RSHIFTS) rshifts = MIN_RSHIFTS; + + if (rshifts > 0) { + C0 = (opus_int32)silk_RSHIFT64(C0_64, rshifts ); + } else { + C0 = silk_LSHIFT32((opus_int32)C0_64, -rshifts ); + } + + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; + x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); + for( n = 1; n < D + 1; n++ ) { + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); + } + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + + invGain_Q30 = (opus_int32)1 << 30; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + if( rshifts > -2 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], 16 - rshifts ); /* Q(16-rshifts) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 16 - rshifts ); /* Q(16-rshifts) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], QA - 16 ); /* Q(QA-16) */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], QA - 16 ); /* Q(QA-16) */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_SMLAWB( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_SMLAWB( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp_QA = Af_QA[ k ]; + tmp1 = silk_SMLAWB( tmp1, Atmp_QA, x_ptr[ n - k - 1 ] ); /* Q(QA-16) */ + tmp2 = silk_SMLAWB( tmp2, Atmp_QA, x_ptr[ subfr_length - n + k ] ); /* Q(QA-16) */ + } + tmp1 = silk_LSHIFT32( -tmp1, 32 - QA - rshifts ); /* Q(16-rshifts) */ + tmp2 = silk_LSHIFT32( -tmp2, 32 - QA - rshifts ); /* Q(16-rshifts) */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWB( CAf[ k ], tmp1, x_ptr[ n - k ] ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWB( CAb[ k ], tmp2, x_ptr[ subfr_length - n + k - 1 ] ); /* Q( -rshift ) */ + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], -rshifts ); /* Q( -rshifts ) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], -rshifts ); /* Q( -rshifts ) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], 17 ); /* Q17 */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 17 ); /* Q17 */ + + X1_3210 = _mm_set1_epi32( x1 ); + X2_3210 = _mm_set1_epi32( x2 ); + TMP1_3210 = _mm_setzero_si128(); + TMP2_3210 = _mm_setzero_si128(); + for( k = 0; k < n - 3; k += 4 ) { + PTR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 1 - 3 ] ); + SUBFR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k ] ); + FIRST_3210 = _mm_loadu_si128( (__m128i *)&C_first_row[ k ] ); + PTR_3210 = _mm_shuffle_epi32( PTR_3210, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + LAST_3210 = _mm_loadu_si128( (__m128i *)&C_last_row[ k ] ); + ATMP_3210 = _mm_loadu_si128( (__m128i *)&Af_QA[ k ] ); + + T1_3210 = _mm_mullo_epi32( PTR_3210, X1_3210 ); + T2_3210 = _mm_mullo_epi32( SUBFR_3210, X2_3210 ); + + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 7 ); + ATMP_3210 = _mm_add_epi32( ATMP_3210, CONST1 ); + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 1 ); + + FIRST_3210 = _mm_add_epi32( FIRST_3210, T1_3210 ); + LAST_3210 = _mm_add_epi32( LAST_3210, T2_3210 ); + + PTR_3210 = _mm_mullo_epi32( ATMP_3210, PTR_3210 ); + SUBFR_3210 = _mm_mullo_epi32( ATMP_3210, SUBFR_3210 ); + + _mm_storeu_si128( (__m128i *)&C_first_row[ k ], FIRST_3210 ); + _mm_storeu_si128( (__m128i *)&C_last_row[ k ], LAST_3210 ); + + TMP1_3210 = _mm_add_epi32( TMP1_3210, PTR_3210 ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, SUBFR_3210 ); + } + + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_unpackhi_epi64(TMP1_3210, TMP1_3210 ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_unpackhi_epi64(TMP2_3210, TMP2_3210 ) ); + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_shufflelo_epi16(TMP1_3210, 0x0E ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_shufflelo_epi16(TMP2_3210, 0x0E ) ); + + tmp1 += _mm_cvtsi128_si32( TMP1_3210 ); + tmp2 += _mm_cvtsi128_si32( TMP2_3210 ); + + for( ; k < n; k++ ) { + C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + } + + tmp1 = -tmp1; /* Q17 */ + tmp2 = -tmp2; /* Q17 */ + + { + __m128i xmm_tmp1, xmm_tmp2; + __m128i xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1; + __m128i xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1; + + xmm_tmp1 = _mm_set1_epi32( tmp1 ); + xmm_tmp2 = _mm_set1_epi32( tmp2 ); + + for( k = 0; k <= n - 3; k += 4 ) { + xmm_x_ptr_n_k_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 3 ] ); + xmm_x_ptr_sub_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k - 1 ] ); + + xmm_x_ptr_n_k_x2x0 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_slli_epi32( xmm_x_ptr_n_k_x2x0, -rshifts - 1 ); + xmm_x_ptr_sub_x2x0 = _mm_slli_epi32( xmm_x_ptr_sub_x2x0, -rshifts - 1 ); + + /* equal shift right 4 bytes, xmm_x_ptr_n_k_x3x1 = _mm_srli_si128(xmm_x_ptr_n_k_x2x0, 4)*/ + xmm_x_ptr_n_k_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_x_ptr_sub_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_sub_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_mul_epi32( xmm_x_ptr_n_k_x2x0, xmm_tmp1 ); + xmm_x_ptr_n_k_x3x1 = _mm_mul_epi32( xmm_x_ptr_n_k_x3x1, xmm_tmp1 ); + xmm_x_ptr_sub_x2x0 = _mm_mul_epi32( xmm_x_ptr_sub_x2x0, xmm_tmp2 ); + xmm_x_ptr_sub_x3x1 = _mm_mul_epi32( xmm_x_ptr_sub_x3x1, xmm_tmp2 ); + + xmm_x_ptr_n_k_x2x0 = _mm_srli_epi64( xmm_x_ptr_n_k_x2x0, 16 ); + xmm_x_ptr_n_k_x3x1 = _mm_slli_epi64( xmm_x_ptr_n_k_x3x1, 16 ); + xmm_x_ptr_sub_x2x0 = _mm_srli_epi64( xmm_x_ptr_sub_x2x0, 16 ); + xmm_x_ptr_sub_x3x1 = _mm_slli_epi64( xmm_x_ptr_sub_x3x1, 16 ); + + xmm_x_ptr_n_k_x2x0 = _mm_blend_epi16( xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1, 0xCC ); + xmm_x_ptr_sub_x2x0 = _mm_blend_epi16( xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1, 0xCC ); + + X1_3210 = _mm_loadu_si128( (__m128i *)&CAf[ k ] ); + PTR_3210 = _mm_loadu_si128( (__m128i *)&CAb[ k ] ); + + X1_3210 = _mm_add_epi32( X1_3210, xmm_x_ptr_n_k_x2x0 ); + PTR_3210 = _mm_add_epi32( PTR_3210, xmm_x_ptr_sub_x2x0 ); + + _mm_storeu_si128( (__m128i *)&CAf[ k ], X1_3210 ); + _mm_storeu_si128( (__m128i *)&CAb[ k ], PTR_3210 ); + } + + for( ; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWW( CAf[ k ], tmp1, + silk_LSHIFT32( (opus_int32)x_ptr[ n - k ], -rshifts - 1 ) ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWW( CAb[ k ], tmp2, + silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n + k - 1 ], -rshifts - 1 ) ); /* Q( -rshift ) */ + } + } + } + } + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + tmp1 = C_first_row[ n ]; /* Q( -rshifts ) */ + tmp2 = C_last_row[ n ]; /* Q( -rshifts ) */ + num = 0; /* Q( -rshifts ) */ + nrg = silk_ADD32( CAb[ 0 ], CAf[ 0 ] ); /* Q( 1-rshifts ) */ + for( k = 0; k < n; k++ ) { + Atmp_QA = Af_QA[ k ]; + lz = silk_CLZ32( silk_abs( Atmp_QA ) ) - 1; + lz = silk_min( 32 - QA, lz ); + Atmp1 = silk_LSHIFT32( Atmp_QA, lz ); /* Q( QA + lz ) */ + + tmp1 = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( C_last_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + tmp2 = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( C_first_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + num = silk_ADD_LSHIFT32( num, silk_SMMUL( CAb[ n - k ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + nrg = silk_ADD_LSHIFT32( nrg, silk_SMMUL( silk_ADD32( CAb[ k + 1 ], CAf[ k + 1 ] ), + Atmp1 ), 32 - QA - lz ); /* Q( 1-rshifts ) */ + } + CAf[ n + 1 ] = tmp1; /* Q( -rshifts ) */ + CAb[ n + 1 ] = tmp2; /* Q( -rshifts ) */ + num = silk_ADD32( num, tmp2 ); /* Q( -rshifts ) */ + num = silk_LSHIFT32( -num, 1 ); /* Q( 1-rshifts ) */ + + /* Calculate the next order reflection (parcor) coefficient */ + if( silk_abs( num ) < nrg ) { + rc_Q31 = silk_DIV32_varQ( num, nrg, 31 ); + } else { + rc_Q31 = ( num > 0 ) ? silk_int32_MAX : silk_int32_MIN; + } + + /* Update inverse prediction gain */ + tmp1 = ( (opus_int32)1 << 30 ) - silk_SMMUL( rc_Q31, rc_Q31 ); + tmp1 = silk_LSHIFT( silk_SMMUL( invGain_Q30, tmp1 ), 2 ); + if( tmp1 <= minInvGain_Q30 ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ + rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } + } + invGain_Q30 = minInvGain_Q30; + reached_max_gain = 1; + } else { + invGain_Q30 = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af_QA[ k ]; /* QA */ + tmp2 = Af_QA[ n - k - 1 ]; /* QA */ + Af_QA[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* QA */ + Af_QA[ n - k - 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* QA */ + } + Af_QA[ n ] = silk_RSHIFT32( rc_Q31, 31 - QA ); /* QA */ + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af_QA[ k ] = 0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; /* Q( -rshifts ) */ + tmp2 = CAb[ n - k + 1 ]; /* Q( -rshifts ) */ + CAf[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + CAb[ n - k + 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + } + } + + if( reached_max_gain ) { + for( k = 0; k < D; k++ ) { + /* Scale coefficients */ + A_Q16[ k ] = -silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); + } + /* Subtract energy of preceding samples from C0 */ + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= silk_LSHIFT32( silk_inner_prod_aligned( x_ptr, x_ptr, D, arch ), -rshifts ); + } + } + /* Approximate residual energy */ + *res_nrg = silk_LSHIFT( silk_SMMUL( invGain_Q30, C0 ), 2 ); + *res_nrg_Q = -rshifts; + } else { + /* Return residual energy */ + nrg = CAf[ 0 ]; /* Q( -rshifts ) */ + tmp1 = (opus_int32)1 << 16; /* Q16 */ + for( k = 0; k < D; k++ ) { + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); /* Q16 */ + nrg = silk_SMLAWW( nrg, CAf[ k + 1 ], Atmp1 ); /* Q( -rshifts ) */ + tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ + A_Q16[ k ] = -Atmp1; + } + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ + *res_nrg_Q = -rshifts; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 res_nrg_c = 0; + opus_int res_nrg_Q_c = 0; + opus_int32 A_Q16_c[ MAX_LPC_ORDER ] = {0}; + + silk_burg_modified_c( + &res_nrg_c, + &res_nrg_Q_c, + A_Q16_c, + x, + minInvGain_Q30, + subfr_length, + nb_subfr, + D, + 0 + ); + + silk_assert( *res_nrg == res_nrg_c ); + silk_assert( *res_nrg_Q == res_nrg_Q_c ); + silk_assert( !memcmp( A_Q16, A_Q16_c, D * sizeof( *A_Q16 ) ) ); + } +#endif +} diff --git a/code/opus-1.5.2/silk/fixed/x86/vector_ops_FIX_sse4_1.c b/code/opus-1.5.2/silk/fixed/x86/vector_ops_FIX_sse4_1.c new file mode 100644 index 00000000..a46289bb --- /dev/null +++ b/code/opus-1.5.2/silk/fixed/x86/vector_ops_FIX_sse4_1.c @@ -0,0 +1,93 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include "main.h" + +#include "SigProc_FIX.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +opus_int64 silk_inner_prod16_sse4_1( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i, dataSize4; + opus_int64 sum; + + __m128i xmm_prod_20, xmm_prod_31; + __m128i inVec1_3210, acc1; + __m128i inVec2_3210, acc2; + + sum = 0; + dataSize4 = len & ~3; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for( i = 0; i < dataSize4; i += 4 ) { + inVec1_3210 = OP_CVTEPI16_EPI32_M64( &inVec1[i + 0] ); + inVec2_3210 = OP_CVTEPI16_EPI32_M64( &inVec2[i + 0] ); + xmm_prod_20 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + inVec1_3210 = _mm_shuffle_epi32( inVec1_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + inVec2_3210 = _mm_shuffle_epi32( inVec2_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_prod_31 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + acc1 = _mm_add_epi64( acc1, xmm_prod_20 ); + acc2 = _mm_add_epi64( acc2, xmm_prod_31 ); + } + + acc1 = _mm_add_epi64( acc1, acc2 ); + + /* equal shift right 8 bytes */ + acc2 = _mm_shuffle_epi32( acc1, _MM_SHUFFLE( 0, 0, 3, 2 ) ); + acc1 = _mm_add_epi64( acc1, acc2 ); + + _mm_storel_epi64( (__m128i *)&sum, acc1 ); + + for( ; i < len; i++ ) { + sum = silk_SMLALBB( sum, inVec1[ i ], inVec2[ i ] ); + } + +#ifdef OPUS_CHECK_ASM + { + opus_int64 sum_c = silk_inner_prod16_c( inVec1, inVec2, len ); + silk_assert( sum == sum_c ); + } +#endif + + return sum; +} diff --git a/code/opus-1.2.1/silk/float/LPC_analysis_filter_FLP.c b/code/opus-1.5.2/silk/float/LPC_analysis_filter_FLP.c similarity index 99% rename from code/opus-1.2.1/silk/float/LPC_analysis_filter_FLP.c rename to code/opus-1.5.2/silk/float/LPC_analysis_filter_FLP.c index cae89a0a..0e1a1fed 100644 --- a/code/opus-1.2.1/silk/float/LPC_analysis_filter_FLP.c +++ b/code/opus-1.5.2/silk/float/LPC_analysis_filter_FLP.c @@ -215,7 +215,7 @@ void silk_LPC_analysis_filter_FLP( const opus_int Order /* I LPC order */ ) { - silk_assert( Order <= length ); + celt_assert( Order <= length ); switch( Order ) { case 6: @@ -239,7 +239,7 @@ void silk_LPC_analysis_filter_FLP( break; default: - silk_assert( 0 ); + celt_assert( 0 ); break; } diff --git a/code/opus-1.2.1/silk/float/LPC_inv_pred_gain_FLP.c b/code/opus-1.5.2/silk/float/LPC_inv_pred_gain_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/LPC_inv_pred_gain_FLP.c rename to code/opus-1.5.2/silk/float/LPC_inv_pred_gain_FLP.c diff --git a/code/opus-1.2.1/silk/float/LTP_analysis_filter_FLP.c b/code/opus-1.5.2/silk/float/LTP_analysis_filter_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/LTP_analysis_filter_FLP.c rename to code/opus-1.5.2/silk/float/LTP_analysis_filter_FLP.c diff --git a/code/opus-1.2.1/silk/float/LTP_scale_ctrl_FLP.c b/code/opus-1.5.2/silk/float/LTP_scale_ctrl_FLP.c similarity index 78% rename from code/opus-1.2.1/silk/float/LTP_scale_ctrl_FLP.c rename to code/opus-1.5.2/silk/float/LTP_scale_ctrl_FLP.c index 8dbe29d0..6f30ff09 100644 --- a/code/opus-1.2.1/silk/float/LTP_scale_ctrl_FLP.c +++ b/code/opus-1.5.2/silk/float/LTP_scale_ctrl_FLP.c @@ -41,8 +41,14 @@ void silk_LTP_scale_ctrl_FLP( if( condCoding == CODE_INDEPENDENTLY ) { /* Only scale if first frame in packet */ - round_loss = psEnc->sCmn.PacketLoss_perc + psEnc->sCmn.nFramesPerPacket; - psEnc->sCmn.indices.LTP_scaleIndex = (opus_int8)silk_LIMIT( round_loss * psEncCtrl->LTPredCodGain * 0.1f, 0.0f, 2.0f ); + round_loss = psEnc->sCmn.PacketLoss_perc * psEnc->sCmn.nFramesPerPacket; + if ( psEnc->sCmn.LBRR_flag ) { + /* LBRR reduces the effective loss. In practice, it does not square the loss because + losses aren't independent, but that still seems to work best. We also never go below 2%. */ + round_loss = 2 + silk_SMULBB( round_loss, round_loss) / 100; + } + psEnc->sCmn.indices.LTP_scaleIndex = silk_SMULBB( psEncCtrl->LTPredCodGain, round_loss ) > silk_log2lin( 2900 - psEnc->sCmn.SNR_dB_Q7 ); + psEnc->sCmn.indices.LTP_scaleIndex += silk_SMULBB( psEncCtrl->LTPredCodGain, round_loss ) > silk_log2lin( 3900 - psEnc->sCmn.SNR_dB_Q7 ); } else { /* Default is minimum scaling */ psEnc->sCmn.indices.LTP_scaleIndex = 0; diff --git a/code/opus-1.2.1/silk/float/SigProc_FLP.h b/code/opus-1.5.2/silk/float/SigProc_FLP.h similarity index 95% rename from code/opus-1.2.1/silk/float/SigProc_FLP.h rename to code/opus-1.5.2/silk/float/SigProc_FLP.h index 953de8b0..ff9281b8 100644 --- a/code/opus-1.2.1/silk/float/SigProc_FLP.h +++ b/code/opus-1.5.2/silk/float/SigProc_FLP.h @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" #include "float_cast.h" +#include "main.h" #include #ifdef __cplusplus @@ -73,7 +74,8 @@ void silk_autocorrelation_FLP( silk_float *results, /* O result (length correlationCount) */ const silk_float *inputData, /* I input data to correlate */ opus_int inputDataSize, /* I length of input */ - opus_int correlationCount /* I number of correlation taps to compute */ + opus_int correlationCount, /* I number of correlation taps to compute */ + int arch ); opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, 1 unvoiced */ @@ -105,7 +107,8 @@ silk_float silk_burg_modified_FLP( /* O returns residual energy const silk_float minInvGain, /* I minimum inverse prediction gain */ const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */ const opus_int nb_subfr, /* I number of subframes stacked in x */ - const opus_int D /* I order */ + const opus_int D, /* I order */ + int arch ); /* multiply a vector by a constant */ @@ -124,12 +127,17 @@ void silk_scale_copy_vector_FLP( ); /* inner product of two silk_float arrays, with result as double */ -double silk_inner_product_FLP( +double silk_inner_product_FLP_c( const silk_float *data1, const silk_float *data2, opus_int dataSize ); +#ifndef OVERRIDE_inner_product_FLP +#define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,silk_inner_product_FLP_c(data1, data2, dataSize)) +#endif + + /* sum of squares of a silk_float array, with result as double */ double silk_energy_FLP( const silk_float *data, diff --git a/code/opus-1.2.1/silk/float/apply_sine_window_FLP.c b/code/opus-1.5.2/silk/float/apply_sine_window_FLP.c similarity index 97% rename from code/opus-1.2.1/silk/float/apply_sine_window_FLP.c rename to code/opus-1.5.2/silk/float/apply_sine_window_FLP.c index 6aae57c0..e49e7179 100644 --- a/code/opus-1.2.1/silk/float/apply_sine_window_FLP.c +++ b/code/opus-1.5.2/silk/float/apply_sine_window_FLP.c @@ -45,10 +45,10 @@ void silk_apply_sine_window_FLP( opus_int k; silk_float freq, c, S0, S1; - silk_assert( win_type == 1 || win_type == 2 ); + celt_assert( win_type == 1 || win_type == 2 ); /* Length must be multiple of 4 */ - silk_assert( ( length & 3 ) == 0 ); + celt_assert( ( length & 3 ) == 0 ); freq = PI / ( length + 1 ); diff --git a/code/opus-1.2.1/silk/float/autocorrelation_FLP.c b/code/opus-1.5.2/silk/float/autocorrelation_FLP.c similarity index 92% rename from code/opus-1.2.1/silk/float/autocorrelation_FLP.c rename to code/opus-1.5.2/silk/float/autocorrelation_FLP.c index 8b8a9e65..4253b26e 100644 --- a/code/opus-1.2.1/silk/float/autocorrelation_FLP.c +++ b/code/opus-1.5.2/silk/float/autocorrelation_FLP.c @@ -37,7 +37,8 @@ void silk_autocorrelation_FLP( silk_float *results, /* O result (length correlationCount) */ const silk_float *inputData, /* I input data to correlate */ opus_int inputDataSize, /* I length of input */ - opus_int correlationCount /* I number of correlation taps to compute */ + opus_int correlationCount, /* I number of correlation taps to compute */ + int arch ) { opus_int i; @@ -47,6 +48,6 @@ void silk_autocorrelation_FLP( } for( i = 0; i < correlationCount; i++ ) { - results[ i ] = (silk_float)silk_inner_product_FLP( inputData, inputData + i, inputDataSize - i ); + results[ i ] = (silk_float)silk_inner_product_FLP( inputData, inputData + i, inputDataSize - i, arch ); } } diff --git a/code/opus-1.2.1/silk/float/burg_modified_FLP.c b/code/opus-1.5.2/silk/float/burg_modified_FLP.c similarity index 97% rename from code/opus-1.2.1/silk/float/burg_modified_FLP.c rename to code/opus-1.5.2/silk/float/burg_modified_FLP.c index ea5dc25a..f5bef5dd 100644 --- a/code/opus-1.2.1/silk/float/burg_modified_FLP.c +++ b/code/opus-1.5.2/silk/float/burg_modified_FLP.c @@ -42,7 +42,8 @@ silk_float silk_burg_modified_FLP( /* O returns residual energy const silk_float minInvGain, /* I minimum inverse prediction gain */ const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */ const opus_int nb_subfr, /* I number of subframes stacked in x */ - const opus_int D /* I order */ + const opus_int D, /* I order */ + int arch ) { opus_int k, n, s, reached_max_gain; @@ -52,7 +53,7 @@ silk_float silk_burg_modified_FLP( /* O returns residual energy double CAf[ SILK_MAX_ORDER_LPC + 1 ], CAb[ SILK_MAX_ORDER_LPC + 1 ]; double Af[ SILK_MAX_ORDER_LPC ]; - silk_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); /* Compute autocorrelations, added over subframes */ C0 = silk_energy_FLP( x, nb_subfr * subfr_length ); @@ -60,7 +61,7 @@ silk_float silk_burg_modified_FLP( /* O returns residual energy for( s = 0; s < nb_subfr; s++ ) { x_ptr = x + s * subfr_length; for( n = 1; n < D + 1; n++ ) { - C_first_row[ n - 1 ] += silk_inner_product_FLP( x_ptr, x_ptr + n, subfr_length - n ); + C_first_row[ n - 1 ] += silk_inner_product_FLP( x_ptr, x_ptr + n, subfr_length - n, arch ); } } silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( double ) ); diff --git a/code/opus-1.2.1/silk/float/bwexpander_FLP.c b/code/opus-1.5.2/silk/float/bwexpander_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/bwexpander_FLP.c rename to code/opus-1.5.2/silk/float/bwexpander_FLP.c diff --git a/code/opus-1.2.1/silk/float/corrMatrix_FLP.c b/code/opus-1.5.2/silk/float/corrMatrix_FLP.c similarity index 94% rename from code/opus-1.2.1/silk/float/corrMatrix_FLP.c rename to code/opus-1.5.2/silk/float/corrMatrix_FLP.c index eae6a1cf..eef6e8aa 100644 --- a/code/opus-1.2.1/silk/float/corrMatrix_FLP.c +++ b/code/opus-1.5.2/silk/float/corrMatrix_FLP.c @@ -41,7 +41,8 @@ void silk_corrVector_FLP( const silk_float *t, /* I Target vector [L] */ const opus_int L, /* I Length of vecors */ const opus_int Order, /* I Max lag for correlation */ - silk_float *Xt /* O X'*t correlation vector [order] */ + silk_float *Xt, /* O X'*t correlation vector [order] */ + int arch ) { opus_int lag; @@ -50,7 +51,7 @@ void silk_corrVector_FLP( ptr1 = &x[ Order - 1 ]; /* Points to first sample of column 0 of X: X[:,0] */ for( lag = 0; lag < Order; lag++ ) { /* Calculate X[:,lag]'*t */ - Xt[ lag ] = (silk_float)silk_inner_product_FLP( ptr1, t, L ); + Xt[ lag ] = (silk_float)silk_inner_product_FLP( ptr1, t, L, arch ); ptr1--; /* Next column of X */ } } @@ -60,7 +61,8 @@ void silk_corrMatrix_FLP( const silk_float *x, /* I x vector [ L+order-1 ] used to create X */ const opus_int L, /* I Length of vectors */ const opus_int Order, /* I Max lag for correlation */ - silk_float *XX /* O X'*X correlation matrix [order x order] */ + silk_float *XX, /* O X'*X correlation matrix [order x order] */ + int arch ) { opus_int j, lag; @@ -79,7 +81,7 @@ void silk_corrMatrix_FLP( ptr2 = &x[ Order - 2 ]; /* First sample of column 1 of X */ for( lag = 1; lag < Order; lag++ ) { /* Calculate X[:,0]'*X[:,lag] */ - energy = silk_inner_product_FLP( ptr1, ptr2, L ); + energy = silk_inner_product_FLP( ptr1, ptr2, L, arch ); matrix_ptr( XX, lag, 0, Order ) = ( silk_float )energy; matrix_ptr( XX, 0, lag, Order ) = ( silk_float )energy; /* Calculate X[:,j]'*X[:,j + lag] */ diff --git a/code/opus-1.2.1/silk/float/encode_frame_FLP.c b/code/opus-1.5.2/silk/float/encode_frame_FLP.c similarity index 94% rename from code/opus-1.2.1/silk/float/encode_frame_FLP.c rename to code/opus-1.5.2/silk/float/encode_frame_FLP.c index c3ad50a9..8a327c56 100644 --- a/code/opus-1.2.1/silk/float/encode_frame_FLP.c +++ b/code/opus-1.5.2/silk/float/encode_frame_FLP.c @@ -42,21 +42,28 @@ static OPUS_INLINE void silk_LBRR_encode_FLP( ); void silk_encode_do_VAD_FLP( - silk_encoder_state_FLP *psEnc /* I/O Encoder state FLP */ + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int activity /* I Decision of Opus voice activity detector */ ) { + const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); + /****************************/ /* Voice Activity Detection */ /****************************/ silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); + /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ + if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { + psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; + } /**************************************************/ /* Convert speech activity into VAD and DTX flags */ /**************************************************/ - if( psEnc->sCmn.speech_activity_Q8 < SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ) ) { + if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; psEnc->sCmn.noSpeechCounter++; - if( psEnc->sCmn.noSpeechCounter < NB_SPEECH_FRAMES_BEFORE_DTX ) { + if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { psEnc->sCmn.inDTX = 0; } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; @@ -100,7 +107,10 @@ opus_int silk_encode_frame_FLP( opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; opus_int best_sum[ MAX_NB_SUBFR ]; + opus_int bits_margin; + /* For CBR, 5 bits below budget is close enough. For VBR, allow up to 25% below the cap if we initially busted the budget. */ + bits_margin = useCBR ? 5 : maxBits/4; /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; @@ -241,7 +251,7 @@ opus_int silk_encode_frame_FLP( if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { /* Restore output state from earlier iteration that did meet the bitrate budget */ silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); - silk_assert( sRangeEnc_copy2.offs <= 1275 ); + celt_assert( sRangeEnc_copy2.offs <= 1275 ); silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) ); psEnc->sShape.LastGainIndex = LastGainIndex_copy2; @@ -263,7 +273,7 @@ opus_int silk_encode_frame_FLP( gainMult_upper = gainMult_Q8; gainsID_upper = gainsID; } - } else if( nBits < maxBits - 5 ) { + } else if( nBits < maxBits - bits_margin ) { found_lower = 1; nBits_lower = nBits; gainMult_lower = gainMult_Q8; @@ -271,13 +281,13 @@ opus_int silk_encode_frame_FLP( gainsID_lower = gainsID; /* Copy part of the output state */ silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); - silk_assert( psRangeEnc->offs <= 1275 ); + celt_assert( psRangeEnc->offs <= 1275 ); silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; } } else { - /* Within 5 bits of budget: close enough */ + /* Close enough */ break; } @@ -299,15 +309,9 @@ opus_int silk_encode_frame_FLP( if( ( found_lower & found_upper ) == 0 ) { /* Adjust gain according to high-rate rate/distortion curve */ if( nBits > maxBits ) { - if (gainMult_Q8 < 16384) { - gainMult_Q8 *= 2; - } else { - gainMult_Q8 = 32767; - } + gainMult_Q8 = silk_min_32( 1024, gainMult_Q8*3/2 ); } else { - opus_int32 gain_factor_Q16; - gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7 ) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) ); - gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 ); + gainMult_Q8 = silk_max_32( 64, gainMult_Q8*4/5 ); } } else { /* Adjust gain by interpolating */ diff --git a/code/opus-1.2.1/silk/float/energy_FLP.c b/code/opus-1.5.2/silk/float/energy_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/energy_FLP.c rename to code/opus-1.5.2/silk/float/energy_FLP.c diff --git a/code/opus-1.2.1/silk/float/find_LPC_FLP.c b/code/opus-1.5.2/silk/float/find_LPC_FLP.c similarity index 95% rename from code/opus-1.2.1/silk/float/find_LPC_FLP.c rename to code/opus-1.5.2/silk/float/find_LPC_FLP.c index 4d63964f..6ccd711d 100644 --- a/code/opus-1.2.1/silk/float/find_LPC_FLP.c +++ b/code/opus-1.5.2/silk/float/find_LPC_FLP.c @@ -38,7 +38,8 @@ void silk_find_LPC_FLP( silk_encoder_state *psEncC, /* I/O Encoder state */ opus_int16 NLSF_Q15[], /* O NLSFs */ const silk_float x[], /* I Input signal */ - const silk_float minInvGain /* I Inverse of max prediction gain */ + const silk_float minInvGain, /* I Inverse of max prediction gain */ + int arch ) { opus_int k, subfr_length; @@ -56,12 +57,12 @@ void silk_find_LPC_FLP( psEncC->indices.NLSFInterpCoef_Q2 = 4; /* Burg AR analysis for the full frame */ - res_nrg = silk_burg_modified_FLP( a, x, minInvGain, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder ); + res_nrg = silk_burg_modified_FLP( a, x, minInvGain, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder, arch ); if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) { /* Optimal solution for last 10 ms; subtract residual energy here, as that's easier than */ /* adding it to the residual energy of the first 10 ms in each iteration of the search below */ - res_nrg -= silk_burg_modified_FLP( a_tmp, x + ( MAX_NB_SUBFR / 2 ) * subfr_length, minInvGain, subfr_length, MAX_NB_SUBFR / 2, psEncC->predictLPCOrder ); + res_nrg -= silk_burg_modified_FLP( a_tmp, x + ( MAX_NB_SUBFR / 2 ) * subfr_length, minInvGain, subfr_length, MAX_NB_SUBFR / 2, psEncC->predictLPCOrder, arch ); /* Convert to NLSFs */ silk_A2NLSF_FLP( NLSF_Q15, a_tmp, psEncC->predictLPCOrder ); @@ -99,6 +100,6 @@ void silk_find_LPC_FLP( silk_A2NLSF_FLP( NLSF_Q15, a, psEncC->predictLPCOrder ); } - silk_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || + celt_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); } diff --git a/code/opus-1.2.1/silk/float/find_LTP_FLP.c b/code/opus-1.5.2/silk/float/find_LTP_FLP.c similarity index 95% rename from code/opus-1.2.1/silk/float/find_LTP_FLP.c rename to code/opus-1.5.2/silk/float/find_LTP_FLP.c index f9706493..90aeeac0 100644 --- a/code/opus-1.2.1/silk/float/find_LTP_FLP.c +++ b/code/opus-1.5.2/silk/float/find_LTP_FLP.c @@ -38,7 +38,8 @@ void silk_find_LTP_FLP( const silk_float r_ptr[], /* I LPC residual */ const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ const opus_int subfr_length, /* I Subframe length */ - const opus_int nb_subfr /* I number of subframes */ + const opus_int nb_subfr, /* I number of subframes */ + int arch ) { opus_int k; @@ -50,8 +51,8 @@ void silk_find_LTP_FLP( XX_ptr = XX; for( k = 0; k < nb_subfr; k++ ) { lag_ptr = r_ptr - ( lag[ k ] + LTP_ORDER / 2 ); - silk_corrMatrix_FLP( lag_ptr, subfr_length, LTP_ORDER, XX_ptr ); - silk_corrVector_FLP( lag_ptr, r_ptr, subfr_length, LTP_ORDER, xX_ptr ); + silk_corrMatrix_FLP( lag_ptr, subfr_length, LTP_ORDER, XX_ptr, arch ); + silk_corrVector_FLP( lag_ptr, r_ptr, subfr_length, LTP_ORDER, xX_ptr, arch ); xx = ( silk_float )silk_energy_FLP( r_ptr, subfr_length + LTP_ORDER ); temp = 1.0f / silk_max( xx, LTP_CORR_INV_MAX * 0.5f * ( XX_ptr[ 0 ] + XX_ptr[ 24 ] ) + 1.0f ); silk_scale_vector_FLP( XX_ptr, temp, LTP_ORDER * LTP_ORDER ); diff --git a/code/opus-1.2.1/silk/float/find_pitch_lags_FLP.c b/code/opus-1.5.2/silk/float/find_pitch_lags_FLP.c similarity index 97% rename from code/opus-1.2.1/silk/float/find_pitch_lags_FLP.c rename to code/opus-1.5.2/silk/float/find_pitch_lags_FLP.c index f3b22d25..1f6bd599 100644 --- a/code/opus-1.2.1/silk/float/find_pitch_lags_FLP.c +++ b/code/opus-1.5.2/silk/float/find_pitch_lags_FLP.c @@ -56,7 +56,7 @@ void silk_find_pitch_lags_FLP( buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; /* Safety check */ - silk_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); + celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); x_buf = x - psEnc->sCmn.ltp_mem_length; @@ -82,7 +82,7 @@ void silk_find_pitch_lags_FLP( silk_apply_sine_window_FLP( Wsig_ptr, x_buf_ptr, 2, psEnc->sCmn.la_pitch ); /* Calculate autocorrelation sequence */ - silk_autocorrelation_FLP( auto_corr, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1 ); + silk_autocorrelation_FLP( auto_corr, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1, arch ); /* Add white noise, as a fraction of the energy */ auto_corr[ 0 ] += auto_corr[ 0 ] * FIND_PITCH_WHITE_NOISE_FRACTION + 1; diff --git a/code/opus-1.2.1/silk/float/find_pred_coefs_FLP.c b/code/opus-1.5.2/silk/float/find_pred_coefs_FLP.c similarity index 95% rename from code/opus-1.2.1/silk/float/find_pred_coefs_FLP.c rename to code/opus-1.5.2/silk/float/find_pred_coefs_FLP.c index cb2e763b..f3c54cf4 100644 --- a/code/opus-1.2.1/silk/float/find_pred_coefs_FLP.c +++ b/code/opus-1.5.2/silk/float/find_pred_coefs_FLP.c @@ -44,7 +44,8 @@ void silk_find_pred_coefs_FLP( silk_float XXLTP[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ]; silk_float xXLTP[ MAX_NB_SUBFR * LTP_ORDER ]; silk_float invGains[ MAX_NB_SUBFR ]; - opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]; + /* Set to NLSF_Q15 to zero so we don't copy junk to the state. */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]={0}; const silk_float *x_ptr; silk_float *x_pre_ptr, LPC_in_pre[ MAX_NB_SUBFR * MAX_LPC_ORDER + MAX_FRAME_LENGTH ]; silk_float minInvGain; @@ -59,10 +60,10 @@ void silk_find_pred_coefs_FLP( /**********/ /* VOICED */ /**********/ - silk_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + celt_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); /* LTP analysis */ - silk_find_LTP_FLP( XXLTP, xXLTP, res_pitch, psEncCtrl->pitchL, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr ); + silk_find_LTP_FLP( XXLTP, xXLTP, res_pitch, psEncCtrl->pitchL, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); /* Quantize LTP gain parameters */ silk_quant_LTP_gains_FLP( psEncCtrl->LTPCoef, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, @@ -101,7 +102,7 @@ void silk_find_pred_coefs_FLP( } /* LPC_in_pre contains the LTP-filtered input for voiced, and the unfiltered input for unvoiced */ - silk_find_LPC_FLP( &psEnc->sCmn, NLSF_Q15, LPC_in_pre, minInvGain ); + silk_find_LPC_FLP( &psEnc->sCmn, NLSF_Q15, LPC_in_pre, minInvGain, psEnc->sCmn.arch ); /* Quantize LSFs */ silk_process_NLSFs_FLP( &psEnc->sCmn, psEncCtrl->PredCoef, NLSF_Q15, psEnc->sCmn.prev_NLSFq_Q15 ); diff --git a/code/opus-1.2.1/silk/float/inner_product_FLP.c b/code/opus-1.5.2/silk/float/inner_product_FLP.c similarity index 98% rename from code/opus-1.2.1/silk/float/inner_product_FLP.c rename to code/opus-1.5.2/silk/float/inner_product_FLP.c index cdd39d24..88b160ab 100644 --- a/code/opus-1.2.1/silk/float/inner_product_FLP.c +++ b/code/opus-1.5.2/silk/float/inner_product_FLP.c @@ -32,7 +32,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FLP.h" /* inner product of two silk_float arrays, with result as double */ -double silk_inner_product_FLP( +double silk_inner_product_FLP_c( const silk_float *data1, const silk_float *data2, opus_int dataSize diff --git a/code/opus-1.2.1/silk/float/k2a_FLP.c b/code/opus-1.5.2/silk/float/k2a_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/k2a_FLP.c rename to code/opus-1.5.2/silk/float/k2a_FLP.c diff --git a/code/opus-1.2.1/silk/float/main_FLP.h b/code/opus-1.5.2/silk/float/main_FLP.h similarity index 96% rename from code/opus-1.2.1/silk/float/main_FLP.h rename to code/opus-1.5.2/silk/float/main_FLP.h index f47fc93b..2e4435cc 100644 --- a/code/opus-1.2.1/silk/float/main_FLP.h +++ b/code/opus-1.5.2/silk/float/main_FLP.h @@ -56,7 +56,8 @@ void silk_HP_variable_cutoff( /* Encoder main function */ void silk_encode_do_VAD_FLP( - silk_encoder_state_FLP *psEnc /* I/O Encoder state FLP */ + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int activity /* I Decision of Opus voice activity detector */ ); /* Encoder main function */ @@ -137,7 +138,8 @@ void silk_find_LPC_FLP( silk_encoder_state *psEncC, /* I/O Encoder state */ opus_int16 NLSF_Q15[], /* O NLSFs */ const silk_float x[], /* I Input signal */ - const silk_float minInvGain /* I Prediction gain from LTP (dB) */ + const silk_float minInvGain, /* I Prediction gain from LTP (dB) */ + int arch ); /* LTP analysis */ @@ -147,7 +149,8 @@ void silk_find_LTP_FLP( const silk_float r_ptr[], /* I LPC residual */ const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ const opus_int subfr_length, /* I Subframe length */ - const opus_int nb_subfr /* I number of subframes */ + const opus_int nb_subfr, /* I number of subframes */ + int arch ); void silk_LTP_analysis_filter_FLP( @@ -220,7 +223,8 @@ void silk_corrMatrix_FLP( const silk_float *x, /* I x vector [ L+order-1 ] used to create X */ const opus_int L, /* I Length of vectors */ const opus_int Order, /* I Max lag for correlation */ - silk_float *XX /* O X'*X correlation matrix [order x order] */ + silk_float *XX, /* O X'*X correlation matrix [order x order] */ + int arch ); /* Calculates correlation vector X'*t */ @@ -229,7 +233,8 @@ void silk_corrVector_FLP( const silk_float *t, /* I Target vector [L] */ const opus_int L, /* I Length of vecors */ const opus_int Order, /* I Max lag for correlation */ - silk_float *Xt /* O X'*t correlation vector [order] */ + silk_float *Xt, /* O X'*t correlation vector [order] */ + int arch ); /* Apply sine window to signal vector. */ diff --git a/code/opus-1.2.1/silk/float/noise_shape_analysis_FLP.c b/code/opus-1.5.2/silk/float/noise_shape_analysis_FLP.c similarity index 99% rename from code/opus-1.2.1/silk/float/noise_shape_analysis_FLP.c rename to code/opus-1.5.2/silk/float/noise_shape_analysis_FLP.c index cb3d8a50..0b5ea952 100644 --- a/code/opus-1.2.1/silk/float/noise_shape_analysis_FLP.c +++ b/code/opus-1.5.2/silk/float/noise_shape_analysis_FLP.c @@ -255,7 +255,7 @@ void silk_noise_shape_analysis_FLP( psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder ); } else { /* Calculate regular auto correlation */ - silk_autocorrelation_FLP( auto_corr, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1 ); + silk_autocorrelation_FLP( auto_corr, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, psEnc->sCmn.arch ); } /* Add white noise, as a fraction of energy */ diff --git a/code/opus-1.2.1/silk/float/pitch_analysis_core_FLP.c b/code/opus-1.5.2/silk/float/pitch_analysis_core_FLP.c similarity index 96% rename from code/opus-1.2.1/silk/float/pitch_analysis_core_FLP.c rename to code/opus-1.5.2/silk/float/pitch_analysis_core_FLP.c index b3716937..0530a883 100644 --- a/code/opus-1.2.1/silk/float/pitch_analysis_core_FLP.c +++ b/code/opus-1.5.2/silk/float/pitch_analysis_core_FLP.c @@ -109,11 +109,11 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, const opus_int8 *Lag_CB_ptr; /* Check for valid sampling frequency */ - silk_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); + celt_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); /* Check for valid complexity setting */ - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); silk_assert( search_thres1 >= 0.0f && search_thres1 <= 1.0f ); silk_assert( search_thres2 >= 0.0f && search_thres2 <= 1.0f ); @@ -148,7 +148,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, silk_resampler_down2_3( filt_state, frame_8_FIX, frame_12_FIX, frame_length ); silk_short2float_array( frame_8kHz, frame_8_FIX, frame_length_8kHz ); } else { - silk_assert( Fs_kHz == 8 ); + celt_assert( Fs_kHz == 8 ); silk_float2short_array( frame_8_FIX, frame, frame_length_8kHz ); } @@ -169,14 +169,14 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, target_ptr = &frame_4kHz[ silk_LSHIFT( sf_length_4kHz, 2 ) ]; for( k = 0; k < nb_subfr >> 1; k++ ) { /* Check that we are within range of the array */ - silk_assert( target_ptr >= frame_4kHz ); - silk_assert( target_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + celt_assert( target_ptr >= frame_4kHz ); + celt_assert( target_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); basis_ptr = target_ptr - min_lag_4kHz; /* Check that we are within range of the array */ - silk_assert( basis_ptr >= frame_4kHz ); - silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + celt_assert( basis_ptr >= frame_4kHz ); + celt_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); celt_pitch_xcorr( target_ptr, target_ptr-max_lag_4kHz, xcorr, sf_length_8kHz, max_lag_4kHz - min_lag_4kHz + 1, arch ); @@ -215,7 +215,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, /* Sort */ length_d_srch = 4 + 2 * complexity; - silk_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); + celt_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); silk_insertion_sort_decreasing_FLP( &C[ 0 ][ min_lag_4kHz ], d_srch, max_lag_4kHz - min_lag_4kHz + 1, length_d_srch ); /* Escape if correlation is very low already here */ @@ -238,7 +238,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, break; } } - silk_assert( length_d_srch > 0 ); + celt_assert( length_d_srch > 0 ); for( i = min_lag_8kHz - 5; i < max_lag_8kHz + 5; i++ ) { d_comp[ i ] = 0; @@ -291,7 +291,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, for( j = 0; j < length_d_comp; j++ ) { d = d_comp[ j ]; basis_ptr = target_ptr - d; - cross_corr = silk_inner_product_FLP( basis_ptr, target_ptr, sf_length_8kHz ); + cross_corr = silk_inner_product_FLP( basis_ptr, target_ptr, sf_length_8kHz, arch ); if( cross_corr > 0.0f ) { energy = silk_energy_FLP( basis_ptr, sf_length_8kHz ); C[ k ][ d ] = (silk_float)( 2 * cross_corr / ( energy + energy_tmp ) ); @@ -471,7 +471,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, *lagIndex = (opus_int16)( lag - min_lag_8kHz ); *contourIndex = (opus_int8)CBimax; } - silk_assert( *lagIndex >= 0 ); + celt_assert( *lagIndex >= 0 ); /* return as voiced */ return 0; } @@ -506,8 +506,8 @@ static void silk_P_Ana_calc_corr_st3( opus_val32 xcorr[ SCRATCH_SIZE ]; const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); if( nb_subfr == PE_MAX_NB_SUBFR ) { Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; @@ -515,7 +515,7 @@ static void silk_P_Ana_calc_corr_st3( nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; cbk_size = PE_NB_CBKS_STAGE3_MAX; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; @@ -572,8 +572,8 @@ static void silk_P_Ana_calc_energy_st3( silk_float scratch_mem[ SCRATCH_SIZE ]; const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; - silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); - silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); if( nb_subfr == PE_MAX_NB_SUBFR ) { Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; @@ -581,7 +581,7 @@ static void silk_P_Ana_calc_energy_st3( nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; cbk_size = PE_NB_CBKS_STAGE3_MAX; } else { - silk_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; diff --git a/code/opus-1.2.1/silk/float/process_gains_FLP.c b/code/opus-1.5.2/silk/float/process_gains_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/process_gains_FLP.c rename to code/opus-1.5.2/silk/float/process_gains_FLP.c diff --git a/code/opus-1.2.1/silk/float/regularize_correlations_FLP.c b/code/opus-1.5.2/silk/float/regularize_correlations_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/regularize_correlations_FLP.c rename to code/opus-1.5.2/silk/float/regularize_correlations_FLP.c diff --git a/code/opus-1.2.1/silk/float/residual_energy_FLP.c b/code/opus-1.5.2/silk/float/residual_energy_FLP.c similarity index 99% rename from code/opus-1.2.1/silk/float/residual_energy_FLP.c rename to code/opus-1.5.2/silk/float/residual_energy_FLP.c index b2e03a86..1bd07b33 100644 --- a/code/opus-1.2.1/silk/float/residual_energy_FLP.c +++ b/code/opus-1.5.2/silk/float/residual_energy_FLP.c @@ -47,7 +47,7 @@ silk_float silk_residual_energy_covar_FLP( /* O silk_float tmp, nrg = 0.0f, regularization; /* Safety checks */ - silk_assert( D >= 0 ); + celt_assert( D >= 0 ); regularization = REGULARIZATION_FACTOR * ( wXX[ 0 ] + wXX[ D * D - 1 ] ); for( k = 0; k < MAX_ITERATIONS_RESIDUAL_NRG; k++ ) { diff --git a/code/opus-1.2.1/silk/float/scale_copy_vector_FLP.c b/code/opus-1.5.2/silk/float/scale_copy_vector_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/scale_copy_vector_FLP.c rename to code/opus-1.5.2/silk/float/scale_copy_vector_FLP.c diff --git a/code/opus-1.2.1/silk/float/scale_vector_FLP.c b/code/opus-1.5.2/silk/float/scale_vector_FLP.c similarity index 100% rename from code/opus-1.2.1/silk/float/scale_vector_FLP.c rename to code/opus-1.5.2/silk/float/scale_vector_FLP.c diff --git a/code/opus-1.2.1/silk/float/schur_FLP.c b/code/opus-1.5.2/silk/float/schur_FLP.c similarity index 97% rename from code/opus-1.2.1/silk/float/schur_FLP.c rename to code/opus-1.5.2/silk/float/schur_FLP.c index c1e0bbb5..8526c748 100644 --- a/code/opus-1.2.1/silk/float/schur_FLP.c +++ b/code/opus-1.5.2/silk/float/schur_FLP.c @@ -41,7 +41,7 @@ silk_float silk_schur_FLP( /* O returns residual energy double C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; double Ctmp1, Ctmp2, rc_tmp; - silk_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); /* Copy correlations */ k = 0; diff --git a/code/opus-1.2.1/silk/float/sort_FLP.c b/code/opus-1.5.2/silk/float/sort_FLP.c similarity index 97% rename from code/opus-1.2.1/silk/float/sort_FLP.c rename to code/opus-1.5.2/silk/float/sort_FLP.c index f08d7592..0e18f319 100644 --- a/code/opus-1.2.1/silk/float/sort_FLP.c +++ b/code/opus-1.5.2/silk/float/sort_FLP.c @@ -47,9 +47,9 @@ void silk_insertion_sort_decreasing_FLP( opus_int i, j; /* Safety checks */ - silk_assert( K > 0 ); - silk_assert( L > 0 ); - silk_assert( L >= K ); + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); /* Write start indices in index vector */ for( i = 0; i < K; i++ ) { diff --git a/code/opus-1.2.1/silk/float/structs_FLP.h b/code/opus-1.5.2/silk/float/structs_FLP.h similarity index 100% rename from code/opus-1.2.1/silk/float/structs_FLP.h rename to code/opus-1.5.2/silk/float/structs_FLP.h diff --git a/code/opus-1.2.1/silk/float/warped_autocorrelation_FLP.c b/code/opus-1.5.2/silk/float/warped_autocorrelation_FLP.c similarity index 86% rename from code/opus-1.2.1/silk/float/warped_autocorrelation_FLP.c rename to code/opus-1.5.2/silk/float/warped_autocorrelation_FLP.c index 542414f4..116dab92 100644 --- a/code/opus-1.2.1/silk/float/warped_autocorrelation_FLP.c +++ b/code/opus-1.5.2/silk/float/warped_autocorrelation_FLP.c @@ -42,11 +42,11 @@ void silk_warped_autocorrelation_FLP( { opus_int n, i; double tmp1, tmp2; - double state[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 }; - double C[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 }; + double state[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + double C[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; /* Order must be even */ - silk_assert( ( order & 1 ) == 0 ); + celt_assert( ( order & 1 ) == 0 ); /* Loop over samples */ for( n = 0; n < length; n++ ) { @@ -54,11 +54,13 @@ void silk_warped_autocorrelation_FLP( /* Loop over allpass sections */ for( i = 0; i < order; i += 2 ) { /* Output of allpass section */ - tmp2 = state[ i ] + warping * ( state[ i + 1 ] - tmp1 ); + /* We voluntarily use two multiples instead of factoring the expression to + reduce the length of the dependency chain (tmp1->tmp2->tmp1... ). */ + tmp2 = state[ i ] + warping * state[ i + 1 ] - warping * tmp1; state[ i ] = tmp1; C[ i ] += state[ 0 ] * tmp1; /* Output of allpass section */ - tmp1 = state[ i + 1 ] + warping * ( state[ i + 2 ] - tmp2 ); + tmp1 = state[ i + 1 ] + warping * state[ i + 2 ] - warping * tmp2; state[ i + 1 ] = tmp2; C[ i + 1 ] += state[ 0 ] * tmp2; } diff --git a/code/opus-1.2.1/silk/float/wrappers_FLP.c b/code/opus-1.5.2/silk/float/wrappers_FLP.c similarity index 98% rename from code/opus-1.2.1/silk/float/wrappers_FLP.c rename to code/opus-1.5.2/silk/float/wrappers_FLP.c index ad90b874..c0c183e3 100644 --- a/code/opus-1.2.1/silk/float/wrappers_FLP.c +++ b/code/opus-1.5.2/silk/float/wrappers_FLP.c @@ -190,12 +190,14 @@ void silk_quant_LTP_gains_FLP( opus_int32 XX_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ]; opus_int32 xX_Q17[ MAX_NB_SUBFR * LTP_ORDER ]; - for( i = 0; i < nb_subfr * LTP_ORDER * LTP_ORDER; i++ ) { + i = 0; + do { XX_Q17[ i ] = (opus_int32)silk_float2int( XX[ i ] * 131072.0f ); - } - for( i = 0; i < nb_subfr * LTP_ORDER; i++ ) { + } while ( ++i < nb_subfr * LTP_ORDER * LTP_ORDER ); + i = 0; + do { xX_Q17[ i ] = (opus_int32)silk_float2int( xX[ i ] * 131072.0f ); - } + } while ( ++i < nb_subfr * LTP_ORDER ); silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, sum_log_gain_Q7, &pred_gain_dB_Q7, XX_Q17, xX_Q17, subfr_len, nb_subfr, arch ); diff --git a/code/opus-1.2.1/silk/control_SNR.c b/code/opus-1.5.2/silk/float/x86/inner_product_FLP_avx2.c similarity index 50% rename from code/opus-1.2.1/silk/control_SNR.c rename to code/opus-1.5.2/silk/float/x86/inner_product_FLP_avx2.c index 464c1acf..4a2daaf5 100644 --- a/code/opus-1.2.1/silk/control_SNR.c +++ b/code/opus-1.5.2/silk/float/x86/inner_product_FLP_avx2.c @@ -1,5 +1,6 @@ /*********************************************************************** Copyright (c) 2006-2011, Skype Limited. All rights reserved. + 2023 Amazon Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: @@ -29,47 +30,56 @@ POSSIBILITY OF SUCH DAMAGE. #include "config.h" #endif -#include "main.h" -#include "tuning_parameters.h" +#include "SigProc_FLP.h" +#include -/* Control SNR of redidual quantizer */ -opus_int silk_control_SNR( - silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ - opus_int32 TargetRate_bps /* I Target max bitrate (bps) */ + +/* inner product of two silk_float arrays, with result as double */ +double silk_inner_product_FLP_avx2( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize ) { - opus_int k, ret = SILK_NO_ERROR; - opus_int32 frac_Q6; - const opus_int32 *rateTable; + opus_int i; + __m256d accum1, accum2; + double result; - /* Set bitrate/coding quality */ - TargetRate_bps = silk_LIMIT( TargetRate_bps, MIN_TARGET_RATE_BPS, MAX_TARGET_RATE_BPS ); - if( TargetRate_bps != psEncC->TargetRate_bps ) { - psEncC->TargetRate_bps = TargetRate_bps; + /* 4x unrolled loop */ + result = 0.0; + accum1 = accum2 = _mm256_setzero_pd(); + for( i = 0; i < dataSize - 7; i += 8 ) { + __m128 x1f, x2f; + __m256d x1d, x2d; + x1f = _mm_loadu_ps( &data1[ i ] ); + x2f = _mm_loadu_ps( &data2[ i ] ); + x1d = _mm256_cvtps_pd( x1f ); + x2d = _mm256_cvtps_pd( x2f ); + accum1 = _mm256_fmadd_pd( x1d, x2d, accum1 ); + x1f = _mm_loadu_ps( &data1[ i + 4 ] ); + x2f = _mm_loadu_ps( &data2[ i + 4 ] ); + x1d = _mm256_cvtps_pd( x1f ); + x2d = _mm256_cvtps_pd( x2f ); + accum2 = _mm256_fmadd_pd( x1d, x2d, accum2 ); + } + for( ; i < dataSize - 3; i += 4 ) { + __m128 x1f, x2f; + __m256d x1d, x2d; + x1f = _mm_loadu_ps( &data1[ i ] ); + x2f = _mm_loadu_ps( &data2[ i ] ); + x1d = _mm256_cvtps_pd( x1f ); + x2d = _mm256_cvtps_pd( x2f ); + accum1 = _mm256_fmadd_pd( x1d, x2d, accum1 ); + } + accum1 = _mm256_add_pd(accum1, accum2); + accum1 = _mm256_add_pd(accum1, _mm256_permute2f128_pd(accum1, accum1, 1)); + accum1 = _mm256_hadd_pd(accum1,accum1); + result = _mm256_cvtsd_f64(accum1); - /* If new TargetRate_bps, translate to SNR_dB value */ - if( psEncC->fs_kHz == 8 ) { - rateTable = silk_TargetRate_table_NB; - } else if( psEncC->fs_kHz == 12 ) { - rateTable = silk_TargetRate_table_MB; - } else { - rateTable = silk_TargetRate_table_WB; - } - - /* Reduce bitrate for 10 ms modes in these calculations */ - if( psEncC->nb_subfr == 2 ) { - TargetRate_bps -= REDUCE_BITRATE_10_MS_BPS; - } - - /* Find bitrate interval in table and interpolate */ - for( k = 1; k < TARGET_RATE_TAB_SZ; k++ ) { - if( TargetRate_bps <= rateTable[ k ] ) { - frac_Q6 = silk_DIV32( silk_LSHIFT( TargetRate_bps - rateTable[ k - 1 ], 6 ), rateTable[ k ] - rateTable[ k - 1 ] ); - psEncC->SNR_dB_Q7 = silk_LSHIFT( silk_SNR_table_Q1[ k - 1 ], 6 ) + silk_MUL( frac_Q6, silk_SNR_table_Q1[ k ] - silk_SNR_table_Q1[ k - 1 ] ); - break; - } - } + /* add any remaining products */ + for( ; i < dataSize; i++ ) { + result += data1[ i ] * (double)data2[ i ]; } - return ret; + return result; } diff --git a/code/opus-1.2.1/silk/gain_quant.c b/code/opus-1.5.2/silk/gain_quant.c similarity index 100% rename from code/opus-1.2.1/silk/gain_quant.c rename to code/opus-1.5.2/silk/gain_quant.c diff --git a/code/opus-1.2.1/silk/init_decoder.c b/code/opus-1.5.2/silk/init_decoder.c similarity index 77% rename from code/opus-1.2.1/silk/init_decoder.c rename to code/opus-1.5.2/silk/init_decoder.c index 16c03dcd..01bc4b7a 100644 --- a/code/opus-1.2.1/silk/init_decoder.c +++ b/code/opus-1.5.2/silk/init_decoder.c @@ -31,15 +31,21 @@ POSSIBILITY OF SUCH DAMAGE. #include "main.h" +#ifdef ENABLE_OSCE +#include "osce.h" +#endif + +#include "structs.h" + /************************/ -/* Init Decoder State */ +/* Reset Decoder State */ /************************/ -opus_int silk_init_decoder( +opus_int silk_reset_decoder( silk_decoder_state *psDec /* I/O Decoder state pointer */ ) { /* Clear the entire encoder state, except anything copied */ - silk_memset( psDec, 0, sizeof( silk_decoder_state ) ); + silk_memset( &psDec->SILK_DECODER_STATE_RESET_START, 0, sizeof( silk_decoder_state ) - ((char*) &psDec->SILK_DECODER_STATE_RESET_START - (char*)psDec) ); /* Used to deactivate LSF interpolation */ psDec->first_frame_after_reset = 1; @@ -52,6 +58,27 @@ opus_int silk_init_decoder( /* Reset PLC state */ silk_PLC_Reset( psDec ); +#ifdef ENABLE_OSCE + /* Reset OSCE state and method */ + osce_reset(&psDec->osce, OSCE_DEFAULT_METHOD); +#endif + + return 0; +} + + +/************************/ +/* Init Decoder State */ +/************************/ +opus_int silk_init_decoder( + silk_decoder_state *psDec /* I/O Decoder state pointer */ +) +{ + /* Clear the entire encoder state, except anything copied */ + silk_memset( psDec, 0, sizeof( silk_decoder_state ) ); + + silk_reset_decoder( psDec ); + return(0); } diff --git a/code/opus-1.2.1/silk/init_encoder.c b/code/opus-1.5.2/silk/init_encoder.c similarity index 98% rename from code/opus-1.2.1/silk/init_encoder.c rename to code/opus-1.5.2/silk/init_encoder.c index 65995c33..10d41287 100644 --- a/code/opus-1.2.1/silk/init_encoder.c +++ b/code/opus-1.5.2/silk/init_encoder.c @@ -36,6 +36,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "tuning_parameters.h" #include "cpu_support.h" +#ifdef ENABLE_DRED +#include "dred_encoder.h" +#endif + /*********************************/ /* Initialize Silk Encoder state */ /*********************************/ diff --git a/code/opus-1.2.1/silk/inner_prod_aligned.c b/code/opus-1.5.2/silk/inner_prod_aligned.c similarity index 100% rename from code/opus-1.2.1/silk/inner_prod_aligned.c rename to code/opus-1.5.2/silk/inner_prod_aligned.c diff --git a/code/opus-1.2.1/silk/interpolate.c b/code/opus-1.5.2/silk/interpolate.c similarity index 97% rename from code/opus-1.2.1/silk/interpolate.c rename to code/opus-1.5.2/silk/interpolate.c index 1bd8ca4d..833c28ef 100644 --- a/code/opus-1.2.1/silk/interpolate.c +++ b/code/opus-1.5.2/silk/interpolate.c @@ -42,8 +42,8 @@ void silk_interpolate( { opus_int i; - silk_assert( ifact_Q2 >= 0 ); - silk_assert( ifact_Q2 <= 4 ); + celt_assert( ifact_Q2 >= 0 ); + celt_assert( ifact_Q2 <= 4 ); for( i = 0; i < d; i++ ) { xi[ i ] = (opus_int16)silk_ADD_RSHIFT( x0[ i ], silk_SMULBB( x1[ i ] - x0[ i ], ifact_Q2 ), 2 ); diff --git a/code/opus-1.2.1/silk/lin2log.c b/code/opus-1.5.2/silk/lin2log.c similarity index 100% rename from code/opus-1.2.1/silk/lin2log.c rename to code/opus-1.5.2/silk/lin2log.c diff --git a/code/opus-1.2.1/silk/log2lin.c b/code/opus-1.5.2/silk/log2lin.c similarity index 100% rename from code/opus-1.2.1/silk/log2lin.c rename to code/opus-1.5.2/silk/log2lin.c diff --git a/code/opus-1.2.1/silk/macros.h b/code/opus-1.5.2/silk/macros.h similarity index 100% rename from code/opus-1.2.1/silk/macros.h rename to code/opus-1.5.2/silk/macros.h diff --git a/code/opus-1.2.1/silk/main.h b/code/opus-1.5.2/silk/main.h similarity index 93% rename from code/opus-1.2.1/silk/main.h rename to code/opus-1.5.2/silk/main.h index 1a33eed5..cd576d8c 100644 --- a/code/opus-1.2.1/silk/main.h +++ b/code/opus-1.5.2/silk/main.h @@ -247,21 +247,21 @@ void silk_VQ_WMat_EC_c( /************************************/ void silk_NSQ_c( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int16 x16[], /* I Input */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ); #if !defined(OVERRIDE_silk_NSQ) @@ -273,21 +273,21 @@ void silk_NSQ_c( /* Noise shaping using delayed decision */ void silk_NSQ_del_dec_c( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int16 x16[], /* I Input */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ); #if !defined(OVERRIDE_silk_NSQ_del_dec) @@ -389,6 +389,10 @@ void silk_NLSF_decode( /****************************************************/ /* Decoder Functions */ /****************************************************/ +opus_int silk_reset_decoder( + silk_decoder_state *psDec /* I/O Decoder state pointer */ +); + opus_int silk_init_decoder( silk_decoder_state *psDec /* I/O Decoder state pointer */ ); @@ -410,6 +414,12 @@ opus_int silk_decode_frame( opus_int32 *pN, /* O Pointer to size of output frame */ opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ opus_int condCoding, /* I The type of conditional coding to use */ +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState *lpcnet, +#endif +#ifdef ENABLE_OSCE + OSCEModel *osce_model, +#endif int arch /* I Run-time architecture */ ); diff --git a/code/opus-1.2.1/silk/mips/NSQ_del_dec_mipsr1.h b/code/opus-1.5.2/silk/mips/NSQ_del_dec_mipsr1.h similarity index 99% rename from code/opus-1.2.1/silk/mips/NSQ_del_dec_mipsr1.h rename to code/opus-1.5.2/silk/mips/NSQ_del_dec_mipsr1.h index cd70713a..85bfb637 100644 --- a/code/opus-1.2.1/silk/mips/NSQ_del_dec_mipsr1.h +++ b/code/opus-1.5.2/silk/mips/NSQ_del_dec_mipsr1.h @@ -25,8 +25,8 @@ ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ***********************************************************************/ -#ifndef __NSQ_DEL_DEC_MIPSR1_H__ -#define __NSQ_DEL_DEC_MIPSR1_H__ +#ifndef NSQ_DEL_DEC_MIPSR1_H__ +#define NSQ_DEL_DEC_MIPSR1_H__ #ifdef HAVE_CONFIG_H #include "config.h" @@ -407,4 +407,4 @@ static inline void silk_noise_shape_quantizer_del_dec( } } -#endif /* __NSQ_DEL_DEC_MIPSR1_H__ */ +#endif /* NSQ_DEL_DEC_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/silk/mips/macros_mipsr1.h b/code/opus-1.5.2/silk/mips/macros_mipsr1.h similarity index 96% rename from code/opus-1.2.1/silk/mips/macros_mipsr1.h rename to code/opus-1.5.2/silk/mips/macros_mipsr1.h index 12ed981a..af408802 100644 --- a/code/opus-1.2.1/silk/mips/macros_mipsr1.h +++ b/code/opus-1.5.2/silk/mips/macros_mipsr1.h @@ -26,8 +26,8 @@ POSSIBILITY OF SUCH DAMAGE. ***********************************************************************/ -#ifndef __SILK_MACROS_MIPSR1_H__ -#define __SILK_MACROS_MIPSR1_H__ +#ifndef SILK_MACROS_MIPSR1_H__ +#define SILK_MACROS_MIPSR1_H__ #define mips_clz(x) __builtin_clz(x) @@ -89,4 +89,4 @@ static inline opus_int32 silk_CLZ32(opus_int32 in32) return re32; } -#endif /* __SILK_MACROS_MIPSR1_H__ */ +#endif /* SILK_MACROS_MIPSR1_H__ */ diff --git a/code/opus-1.2.1/silk/mips/sigproc_fix_mipsr1.h b/code/opus-1.5.2/silk/mips/sigproc_fix_mipsr1.h similarity index 100% rename from code/opus-1.2.1/silk/mips/sigproc_fix_mipsr1.h rename to code/opus-1.5.2/silk/mips/sigproc_fix_mipsr1.h diff --git a/code/opus-1.2.1/silk/pitch_est_defines.h b/code/opus-1.5.2/silk/pitch_est_defines.h similarity index 100% rename from code/opus-1.2.1/silk/pitch_est_defines.h rename to code/opus-1.5.2/silk/pitch_est_defines.h diff --git a/code/opus-1.2.1/silk/pitch_est_tables.c b/code/opus-1.5.2/silk/pitch_est_tables.c similarity index 100% rename from code/opus-1.2.1/silk/pitch_est_tables.c rename to code/opus-1.5.2/silk/pitch_est_tables.c diff --git a/code/opus-1.2.1/silk/process_NLSFs.c b/code/opus-1.5.2/silk/process_NLSFs.c similarity index 96% rename from code/opus-1.2.1/silk/process_NLSFs.c rename to code/opus-1.5.2/silk/process_NLSFs.c index 2f10f8df..d1308095 100644 --- a/code/opus-1.2.1/silk/process_NLSFs.c +++ b/code/opus-1.5.2/silk/process_NLSFs.c @@ -48,7 +48,7 @@ void silk_process_NLSFs( silk_assert( psEncC->speech_activity_Q8 >= 0 ); silk_assert( psEncC->speech_activity_Q8 <= SILK_FIX_CONST( 1.0, 8 ) ); - silk_assert( psEncC->useInterpolatedNLSFs == 1 || psEncC->indices.NLSFInterpCoef_Q2 == ( 1 << 2 ) ); + celt_assert( psEncC->useInterpolatedNLSFs == 1 || psEncC->indices.NLSFInterpCoef_Q2 == ( 1 << 2 ) ); /***********************/ /* Calculate mu values */ @@ -60,7 +60,7 @@ void silk_process_NLSFs( NLSF_mu_Q20 = silk_ADD_RSHIFT( NLSF_mu_Q20, NLSF_mu_Q20, 1 ); } - silk_assert( NLSF_mu_Q20 > 0 ); + celt_assert( NLSF_mu_Q20 > 0 ); silk_assert( NLSF_mu_Q20 <= SILK_FIX_CONST( 0.005, 20 ) ); /* Calculate NLSF weights */ @@ -101,7 +101,7 @@ void silk_process_NLSFs( } else { /* Copy LPC coefficients for first half from second half */ - silk_assert( psEncC->predictLPCOrder <= MAX_LPC_ORDER ); + celt_assert( psEncC->predictLPCOrder <= MAX_LPC_ORDER ); silk_memcpy( PredCoef_Q12[ 0 ], PredCoef_Q12[ 1 ], psEncC->predictLPCOrder * sizeof( opus_int16 ) ); } } diff --git a/code/opus-1.2.1/silk/quant_LTP_gains.c b/code/opus-1.5.2/silk/quant_LTP_gains.c similarity index 100% rename from code/opus-1.2.1/silk/quant_LTP_gains.c rename to code/opus-1.5.2/silk/quant_LTP_gains.c diff --git a/code/opus-1.2.1/silk/resampler.c b/code/opus-1.5.2/silk/resampler.c similarity index 98% rename from code/opus-1.2.1/silk/resampler.c rename to code/opus-1.5.2/silk/resampler.c index 374fbb37..1f11e508 100644 --- a/code/opus-1.2.1/silk/resampler.c +++ b/code/opus-1.5.2/silk/resampler.c @@ -91,14 +91,14 @@ opus_int silk_resampler_init( if( forEnc ) { if( ( Fs_Hz_in != 8000 && Fs_Hz_in != 12000 && Fs_Hz_in != 16000 && Fs_Hz_in != 24000 && Fs_Hz_in != 48000 ) || ( Fs_Hz_out != 8000 && Fs_Hz_out != 12000 && Fs_Hz_out != 16000 ) ) { - silk_assert( 0 ); + celt_assert( 0 ); return -1; } S->inputDelay = delay_matrix_enc[ rateID( Fs_Hz_in ) ][ rateID( Fs_Hz_out ) ]; } else { if( ( Fs_Hz_in != 8000 && Fs_Hz_in != 12000 && Fs_Hz_in != 16000 ) || ( Fs_Hz_out != 8000 && Fs_Hz_out != 12000 && Fs_Hz_out != 16000 && Fs_Hz_out != 24000 && Fs_Hz_out != 48000 ) ) { - silk_assert( 0 ); + celt_assert( 0 ); return -1; } S->inputDelay = delay_matrix_dec[ rateID( Fs_Hz_in ) ][ rateID( Fs_Hz_out ) ]; @@ -151,7 +151,7 @@ opus_int silk_resampler_init( S->Coefs = silk_Resampler_1_6_COEFS; } else { /* None available */ - silk_assert( 0 ); + celt_assert( 0 ); return -1; } } else { @@ -181,9 +181,9 @@ opus_int silk_resampler( opus_int nSamples; /* Need at least 1 ms of input data */ - silk_assert( inLen >= S->Fs_in_kHz ); + celt_assert( inLen >= S->Fs_in_kHz ); /* Delay can't exceed the 1 ms of buffering */ - silk_assert( S->inputDelay <= S->Fs_in_kHz ); + celt_assert( S->inputDelay <= S->Fs_in_kHz ); nSamples = S->Fs_in_kHz - S->inputDelay; diff --git a/code/opus-1.2.1/silk/resampler_down2.c b/code/opus-1.5.2/silk/resampler_down2.c similarity index 97% rename from code/opus-1.2.1/silk/resampler_down2.c rename to code/opus-1.5.2/silk/resampler_down2.c index cec36346..971d7bfd 100644 --- a/code/opus-1.2.1/silk/resampler_down2.c +++ b/code/opus-1.5.2/silk/resampler_down2.c @@ -43,8 +43,8 @@ void silk_resampler_down2( opus_int32 k, len2 = silk_RSHIFT32( inLen, 1 ); opus_int32 in32, out32, Y, X; - silk_assert( silk_resampler_down2_0 > 0 ); - silk_assert( silk_resampler_down2_1 < 0 ); + celt_assert( silk_resampler_down2_0 > 0 ); + celt_assert( silk_resampler_down2_1 < 0 ); /* Internal variables and state are in Q10 format */ for( k = 0; k < len2; k++ ) { diff --git a/code/opus-1.2.1/silk/resampler_down2_3.c b/code/opus-1.5.2/silk/resampler_down2_3.c similarity index 100% rename from code/opus-1.2.1/silk/resampler_down2_3.c rename to code/opus-1.5.2/silk/resampler_down2_3.c diff --git a/code/opus-1.2.1/silk/resampler_private.h b/code/opus-1.5.2/silk/resampler_private.h similarity index 100% rename from code/opus-1.2.1/silk/resampler_private.h rename to code/opus-1.5.2/silk/resampler_private.h diff --git a/code/opus-1.2.1/silk/resampler_private_AR2.c b/code/opus-1.5.2/silk/resampler_private_AR2.c similarity index 100% rename from code/opus-1.2.1/silk/resampler_private_AR2.c rename to code/opus-1.5.2/silk/resampler_private_AR2.c diff --git a/code/opus-1.2.1/silk/resampler_private_IIR_FIR.c b/code/opus-1.5.2/silk/resampler_private_IIR_FIR.c similarity index 100% rename from code/opus-1.2.1/silk/resampler_private_IIR_FIR.c rename to code/opus-1.5.2/silk/resampler_private_IIR_FIR.c diff --git a/code/opus-1.2.1/silk/resampler_private_down_FIR.c b/code/opus-1.5.2/silk/resampler_private_down_FIR.c similarity index 99% rename from code/opus-1.2.1/silk/resampler_private_down_FIR.c rename to code/opus-1.5.2/silk/resampler_private_down_FIR.c index 783e42b3..3e8735a3 100644 --- a/code/opus-1.2.1/silk/resampler_private_down_FIR.c +++ b/code/opus-1.5.2/silk/resampler_private_down_FIR.c @@ -136,7 +136,7 @@ static OPUS_INLINE opus_int16 *silk_resampler_private_down_FIR_INTERPOL( } break; default: - silk_assert( 0 ); + celt_assert( 0 ); } return out; } diff --git a/code/opus-1.2.1/silk/resampler_private_up2_HQ.c b/code/opus-1.5.2/silk/resampler_private_up2_HQ.c similarity index 100% rename from code/opus-1.2.1/silk/resampler_private_up2_HQ.c rename to code/opus-1.5.2/silk/resampler_private_up2_HQ.c diff --git a/code/opus-1.2.1/silk/resampler_rom.c b/code/opus-1.5.2/silk/resampler_rom.c similarity index 100% rename from code/opus-1.2.1/silk/resampler_rom.c rename to code/opus-1.5.2/silk/resampler_rom.c diff --git a/code/opus-1.2.1/silk/resampler_rom.h b/code/opus-1.5.2/silk/resampler_rom.h similarity index 100% rename from code/opus-1.2.1/silk/resampler_rom.h rename to code/opus-1.5.2/silk/resampler_rom.h diff --git a/code/opus-1.2.1/silk/resampler_structs.h b/code/opus-1.5.2/silk/resampler_structs.h similarity index 100% rename from code/opus-1.2.1/silk/resampler_structs.h rename to code/opus-1.5.2/silk/resampler_structs.h diff --git a/code/opus-1.2.1/silk/shell_coder.c b/code/opus-1.5.2/silk/shell_coder.c similarity index 100% rename from code/opus-1.2.1/silk/shell_coder.c rename to code/opus-1.5.2/silk/shell_coder.c diff --git a/code/opus-1.2.1/silk/sigm_Q15.c b/code/opus-1.5.2/silk/sigm_Q15.c similarity index 100% rename from code/opus-1.2.1/silk/sigm_Q15.c rename to code/opus-1.5.2/silk/sigm_Q15.c diff --git a/code/opus-1.2.1/silk/sort.c b/code/opus-1.5.2/silk/sort.c similarity index 96% rename from code/opus-1.2.1/silk/sort.c rename to code/opus-1.5.2/silk/sort.c index 7187c9ef..4fba16f8 100644 --- a/code/opus-1.2.1/silk/sort.c +++ b/code/opus-1.5.2/silk/sort.c @@ -48,9 +48,9 @@ void silk_insertion_sort_increasing( opus_int i, j; /* Safety checks */ - silk_assert( K > 0 ); - silk_assert( L > 0 ); - silk_assert( L >= K ); + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); /* Write start indices in index vector */ for( i = 0; i < K; i++ ) { @@ -96,9 +96,9 @@ void silk_insertion_sort_decreasing_int16( opus_int value; /* Safety checks */ - silk_assert( K > 0 ); - silk_assert( L > 0 ); - silk_assert( L >= K ); + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); /* Write start indices in index vector */ for( i = 0; i < K; i++ ) { @@ -141,7 +141,7 @@ void silk_insertion_sort_increasing_all_values_int16( opus_int i, j; /* Safety checks */ - silk_assert( L > 0 ); + celt_assert( L > 0 ); /* Sort vector elements by value, increasing order */ for( i = 1; i < L; i++ ) { diff --git a/code/opus-1.2.1/silk/stereo_LR_to_MS.c b/code/opus-1.5.2/silk/stereo_LR_to_MS.c similarity index 95% rename from code/opus-1.2.1/silk/stereo_LR_to_MS.c rename to code/opus-1.5.2/silk/stereo_LR_to_MS.c index dda0298d..751452cb 100644 --- a/code/opus-1.2.1/silk/stereo_LR_to_MS.c +++ b/code/opus-1.5.2/silk/stereo_LR_to_MS.c @@ -77,7 +77,7 @@ void silk_stereo_LR_to_MS( ALLOC( LP_mid, frame_length, opus_int16 ); ALLOC( HP_mid, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { - sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); LP_mid[ n ] = sum; HP_mid[ n ] = mid[ n + 1 ] - sum; } @@ -86,7 +86,7 @@ void silk_stereo_LR_to_MS( ALLOC( LP_side, frame_length, opus_int16 ); ALLOC( HP_side, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { - sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); LP_side[ n ] = sum; HP_side[ n ] = side[ n + 1 ] - sum; } @@ -109,7 +109,7 @@ void silk_stereo_LR_to_MS( if( total_rate_bps < 1 ) { total_rate_bps = 1; } - min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 900 ); + min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 600 ); silk_assert( min_mid_rate_bps < 32767 ); /* Default bitrate distribution: 8 parts for Mid and (5+3*frac) parts for Side. so: mid_rate = ( 8 / ( 13 + 3 * frac ) ) * total_ rate */ frac_3_Q16 = silk_MUL( 3, frac_Q16 ); @@ -207,7 +207,7 @@ void silk_stereo_LR_to_MS( pred0_Q13 += delta0_Q13; pred1_Q13 += delta1_Q13; w_Q24 += deltaw_Q24; - sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); @@ -217,7 +217,7 @@ void silk_stereo_LR_to_MS( pred1_Q13 = -pred_Q13[ 1 ]; w_Q24 = silk_LSHIFT( width_Q14, 10 ); for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { - sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); diff --git a/code/opus-1.2.1/silk/stereo_MS_to_LR.c b/code/opus-1.5.2/silk/stereo_MS_to_LR.c similarity index 95% rename from code/opus-1.2.1/silk/stereo_MS_to_LR.c rename to code/opus-1.5.2/silk/stereo_MS_to_LR.c index 62521a4f..1e01bb6e 100644 --- a/code/opus-1.2.1/silk/stereo_MS_to_LR.c +++ b/code/opus-1.5.2/silk/stereo_MS_to_LR.c @@ -59,7 +59,7 @@ void silk_stereo_MS_to_LR( for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) { pred0_Q13 += delta0_Q13; pred1_Q13 += delta1_Q13; - sum = silk_LSHIFT( silk_ADD_LSHIFT( x1[ n ] + x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT32( x1[ n ] + (opus_int32)x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_LSHIFT( (opus_int32)x2[ n + 1 ], 8 ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)x1[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); @@ -67,7 +67,7 @@ void silk_stereo_MS_to_LR( pred0_Q13 = pred_Q13[ 0 ]; pred1_Q13 = pred_Q13[ 1 ]; for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { - sum = silk_LSHIFT( silk_ADD_LSHIFT( x1[ n ] + x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT32( x1[ n ] + (opus_int32)x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_LSHIFT( (opus_int32)x2[ n + 1 ], 8 ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)x1[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); diff --git a/code/opus-1.2.1/silk/stereo_decode_pred.c b/code/opus-1.5.2/silk/stereo_decode_pred.c similarity index 100% rename from code/opus-1.2.1/silk/stereo_decode_pred.c rename to code/opus-1.5.2/silk/stereo_decode_pred.c diff --git a/code/opus-1.2.1/silk/stereo_encode_pred.c b/code/opus-1.5.2/silk/stereo_encode_pred.c similarity index 95% rename from code/opus-1.2.1/silk/stereo_encode_pred.c rename to code/opus-1.5.2/silk/stereo_encode_pred.c index e6dd1950..03becb67 100644 --- a/code/opus-1.2.1/silk/stereo_encode_pred.c +++ b/code/opus-1.5.2/silk/stereo_encode_pred.c @@ -41,11 +41,11 @@ void silk_stereo_encode_pred( /* Entropy coding */ n = 5 * ix[ 0 ][ 2 ] + ix[ 1 ][ 2 ]; - silk_assert( n < 25 ); + celt_assert( n < 25 ); ec_enc_icdf( psRangeEnc, n, silk_stereo_pred_joint_iCDF, 8 ); for( n = 0; n < 2; n++ ) { - silk_assert( ix[ n ][ 0 ] < 3 ); - silk_assert( ix[ n ][ 1 ] < STEREO_QUANT_SUB_STEPS ); + celt_assert( ix[ n ][ 0 ] < 3 ); + celt_assert( ix[ n ][ 1 ] < STEREO_QUANT_SUB_STEPS ); ec_enc_icdf( psRangeEnc, ix[ n ][ 0 ], silk_uniform3_iCDF, 8 ); ec_enc_icdf( psRangeEnc, ix[ n ][ 1 ], silk_uniform5_iCDF, 8 ); } diff --git a/code/opus-1.2.1/silk/stereo_find_predictor.c b/code/opus-1.5.2/silk/stereo_find_predictor.c similarity index 100% rename from code/opus-1.2.1/silk/stereo_find_predictor.c rename to code/opus-1.5.2/silk/stereo_find_predictor.c diff --git a/code/opus-1.2.1/silk/stereo_quant_pred.c b/code/opus-1.5.2/silk/stereo_quant_pred.c similarity index 100% rename from code/opus-1.2.1/silk/stereo_quant_pred.c rename to code/opus-1.5.2/silk/stereo_quant_pred.c diff --git a/code/opus-1.2.1/silk/structs.h b/code/opus-1.5.2/silk/structs.h similarity index 96% rename from code/opus-1.2.1/silk/structs.h rename to code/opus-1.5.2/silk/structs.h index 4ff590b0..38243be1 100644 --- a/code/opus-1.2.1/silk/structs.h +++ b/code/opus-1.5.2/silk/structs.h @@ -34,6 +34,21 @@ POSSIBILITY OF SUCH DAMAGE. #include "entenc.h" #include "entdec.h" +#ifdef ENABLE_DEEP_PLC +#include "lpcnet.h" +#include "lpcnet_private.h" +#endif + +#ifdef ENABLE_DRED +#include "dred_encoder.h" +#include "dred_decoder.h" +#endif + +#ifdef ENABLE_OSCE +#include "osce_config.h" +#include "osce_structs.h" +#endif + #ifdef __cplusplus extern "C" { @@ -78,6 +93,7 @@ typedef struct { opus_int32 In_LP_State[ 2 ]; /* Low pass filter state */ opus_int32 transition_frame_no; /* Counter which is mapped to a cut-off frequency */ opus_int mode; /* Operating mode, <0: switch down, >0: switch up; 0: do nothing */ + opus_int32 saved_fs_kHz; /* If non-zero, holds the last sampling rate before a bandwidth switching reset. */ } silk_LP_state; /* Structure containing NLSF codebook */ @@ -227,6 +243,14 @@ typedef struct { } silk_encoder_state; +#ifdef ENABLE_OSCE +typedef struct { + OSCEFeatureState features; + OSCEState state; + int method; +} silk_OSCE_struct; +#endif + /* Struct for Packet Loss Concealment */ typedef struct { opus_int32 pitchL_Q8; /* Pitch lag to use for voiced concealment */ @@ -242,6 +266,7 @@ typedef struct { opus_int fs_kHz; opus_int nb_subfr; opus_int subfr_length; + opus_int enable_deep_plc; } silk_PLC_struct; /* Struct for CNG */ @@ -258,6 +283,10 @@ typedef struct { /* Decoder state */ /********************************/ typedef struct { +#ifdef ENABLE_OSCE + silk_OSCE_struct osce; +#endif +#define SILK_DECODER_STATE_RESET_START prev_gain_Q16 opus_int32 prev_gain_Q16; opus_int32 exc_Q14[ MAX_FRAME_LENGTH ]; opus_int32 sLPC_Q14_buf[ MAX_LPC_ORDER ]; diff --git a/code/opus-1.2.1/silk/sum_sqr_shift.c b/code/opus-1.5.2/silk/sum_sqr_shift.c similarity index 100% rename from code/opus-1.2.1/silk/sum_sqr_shift.c rename to code/opus-1.5.2/silk/sum_sqr_shift.c diff --git a/code/opus-1.2.1/silk/table_LSF_cos.c b/code/opus-1.5.2/silk/table_LSF_cos.c similarity index 100% rename from code/opus-1.2.1/silk/table_LSF_cos.c rename to code/opus-1.5.2/silk/table_LSF_cos.c diff --git a/code/opus-1.2.1/silk/tables.h b/code/opus-1.5.2/silk/tables.h similarity index 93% rename from code/opus-1.2.1/silk/tables.h rename to code/opus-1.5.2/silk/tables.h index 8b0380ee..95230c45 100644 --- a/code/opus-1.2.1/silk/tables.h +++ b/code/opus-1.5.2/silk/tables.h @@ -97,12 +97,6 @@ extern const opus_uint8 silk_NLSF_interpolation_factor_iCDF[ 5 ]; extern const silk_NLSF_CB_struct silk_NLSF_CB_WB; /* 1040 */ extern const silk_NLSF_CB_struct silk_NLSF_CB_NB_MB; /* 728 */ -/* Piece-wise linear mapping from bitrate in kbps to coding quality in dB SNR */ -extern const opus_int32 silk_TargetRate_table_NB[ TARGET_RATE_TAB_SZ ]; /* 32 */ -extern const opus_int32 silk_TargetRate_table_MB[ TARGET_RATE_TAB_SZ ]; /* 32 */ -extern const opus_int32 silk_TargetRate_table_WB[ TARGET_RATE_TAB_SZ ]; /* 32 */ -extern const opus_int16 silk_SNR_table_Q1[ TARGET_RATE_TAB_SZ ]; /* 32 */ - /* Quantization offsets */ extern const opus_int16 silk_Quantization_Offsets_Q10[ 2 ][ 2 ]; /* 8 */ diff --git a/code/opus-1.2.1/silk/tables_LTP.c b/code/opus-1.5.2/silk/tables_LTP.c similarity index 100% rename from code/opus-1.2.1/silk/tables_LTP.c rename to code/opus-1.5.2/silk/tables_LTP.c diff --git a/code/opus-1.2.1/silk/tables_NLSF_CB_NB_MB.c b/code/opus-1.5.2/silk/tables_NLSF_CB_NB_MB.c similarity index 100% rename from code/opus-1.2.1/silk/tables_NLSF_CB_NB_MB.c rename to code/opus-1.5.2/silk/tables_NLSF_CB_NB_MB.c diff --git a/code/opus-1.2.1/silk/tables_NLSF_CB_WB.c b/code/opus-1.5.2/silk/tables_NLSF_CB_WB.c similarity index 100% rename from code/opus-1.2.1/silk/tables_NLSF_CB_WB.c rename to code/opus-1.5.2/silk/tables_NLSF_CB_WB.c diff --git a/code/opus-1.2.1/silk/tables_gain.c b/code/opus-1.5.2/silk/tables_gain.c similarity index 100% rename from code/opus-1.2.1/silk/tables_gain.c rename to code/opus-1.5.2/silk/tables_gain.c diff --git a/code/opus-1.2.1/silk/tables_other.c b/code/opus-1.5.2/silk/tables_other.c similarity index 87% rename from code/opus-1.2.1/silk/tables_other.c rename to code/opus-1.5.2/silk/tables_other.c index 398686bf..e34d9077 100644 --- a/code/opus-1.2.1/silk/tables_other.c +++ b/code/opus-1.5.2/silk/tables_other.c @@ -38,20 +38,6 @@ extern "C" { #endif -/* Piece-wise linear mapping from bitrate in kbps to coding quality in dB SNR */ -const opus_int32 silk_TargetRate_table_NB[ TARGET_RATE_TAB_SZ ] = { - 0, 8000, 9400, 11500, 13500, 17500, 25000, MAX_TARGET_RATE_BPS -}; -const opus_int32 silk_TargetRate_table_MB[ TARGET_RATE_TAB_SZ ] = { - 0, 9000, 12000, 14500, 18500, 24500, 35500, MAX_TARGET_RATE_BPS -}; -const opus_int32 silk_TargetRate_table_WB[ TARGET_RATE_TAB_SZ ] = { - 0, 10500, 14000, 17000, 21500, 28500, 42000, MAX_TARGET_RATE_BPS -}; -const opus_int16 silk_SNR_table_Q1[ TARGET_RATE_TAB_SZ ] = { - 18, 29, 38, 40, 46, 52, 62, 84 -}; - /* Tables for stereo predictor coding */ const opus_int16 silk_stereo_pred_quant_Q13[ STEREO_QUANT_TAB_SIZE ] = { -13732, -10050, -8266, -7526, -6500, -5000, -2950, -820, diff --git a/code/opus-1.2.1/silk/tables_pitch_lag.c b/code/opus-1.5.2/silk/tables_pitch_lag.c similarity index 100% rename from code/opus-1.2.1/silk/tables_pitch_lag.c rename to code/opus-1.5.2/silk/tables_pitch_lag.c diff --git a/code/opus-1.2.1/silk/tables_pulses_per_block.c b/code/opus-1.5.2/silk/tables_pulses_per_block.c similarity index 100% rename from code/opus-1.2.1/silk/tables_pulses_per_block.c rename to code/opus-1.5.2/silk/tables_pulses_per_block.c diff --git a/code/opus-1.5.2/silk/tests/test_unit_LPC_inv_pred_gain.c b/code/opus-1.5.2/silk/tests/test_unit_LPC_inv_pred_gain.c new file mode 100644 index 00000000..acdd31af --- /dev/null +++ b/code/opus-1.5.2/silk/tests/test_unit_LPC_inv_pred_gain.c @@ -0,0 +1,130 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc., Jean-Marc Valin +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include "celt/stack_alloc.h" +#include "cpu_support.h" +#include "SigProc_FIX.h" + +/* Computes the impulse response of the filter so we + can catch filters that are definitely unstable. Some + unstable filters may be classified as stable, but not + the other way around. */ +int check_stability(opus_int16 *A_Q12, int order) { + int i; + int j; + int sum_a, sum_abs_a; + double y[SILK_MAX_ORDER_LPC] = {0}; + sum_a = sum_abs_a = 0; + for( j = 0; j < order; j++ ) { + sum_a += A_Q12[ j ]; + sum_abs_a += silk_abs( A_Q12[ j ] ); + } + /* Check DC stability. */ + if( sum_a >= 4096 ) { + return 0; + } + /* If the sum of absolute values is less than 1, the filter + has to be stable. */ + if( sum_abs_a < 4096 ) { + return 1; + } + y[0] = 1; + for( i = 0; i < 10000; i++ ) { + double sum = 0; + for( j = 0; j < order; j++ ) { + sum += y[ j ]*A_Q12[ j ]; + } + for( j = order - 1; j > 0; j-- ) { + y[ j ] = y[ j - 1 ]; + } + y[ 0 ] = sum*(1./4096); + /* If impulse response reaches +/- 10000, the filter + is definitely unstable. */ + if( !(y[ 0 ] < 10000 && y[ 0 ] > -10000) ) { + return 0; + } + /* Test every 8 sample for low amplitude. */ + if( ( i & 0x7 ) == 0 ) { + double amp = 0; + for( j = 0; j < order; j++ ) { + amp += fabs(y[j]); + } + if( amp < 0.00001 ) { + return 1; + } + } + } + return 1; +} + +int main(void) { + const int arch = opus_select_arch(); + /* Set to 10000 so all branches in C function are triggered */ + const int loop_num = 10000; + int count = 0; + ALLOC_STACK; + + /* FIXME: Make the seed random (with option to set it explicitly) + so we get wider coverage. */ + srand(0); + + printf("Testing silk_LPC_inverse_pred_gain() optimization ...\n"); + for( count = 0; count < loop_num; count++ ) { + unsigned int i; + opus_int order; + unsigned int shift; + opus_int16 A_Q12[ SILK_MAX_ORDER_LPC ]; + opus_int32 gain; + + for( order = 2; order <= SILK_MAX_ORDER_LPC; order += 2 ) { /* order must be even. */ + for( shift = 0; shift < 16; shift++ ) { /* Different dynamic range. */ + for( i = 0; i < SILK_MAX_ORDER_LPC; i++ ) { + A_Q12[i] = ((opus_int16)rand()) >> shift; + } + gain = silk_LPC_inverse_pred_gain(A_Q12, order, arch); + /* Look for filters that silk_LPC_inverse_pred_gain() thinks are + stable but definitely aren't. */ + if( gain != 0 && !check_stability(A_Q12, order) ) { + fprintf(stderr, "**Loop %4d failed!**\n", count); + return 1; + } + } + } + if( !(count % 500) ) { + printf("Loop %4d passed\n", count); + } + } + printf("silk_LPC_inverse_pred_gain() optimization passed\n"); + RESTORE_STACK; + return 0; +} diff --git a/code/opus-1.2.1/silk/tuning_parameters.h b/code/opus-1.5.2/silk/tuning_parameters.h similarity index 100% rename from code/opus-1.2.1/silk/tuning_parameters.h rename to code/opus-1.5.2/silk/tuning_parameters.h diff --git a/code/opus-1.2.1/silk/typedef.h b/code/opus-1.5.2/silk/typedef.h similarity index 97% rename from code/opus-1.2.1/silk/typedef.h rename to code/opus-1.5.2/silk/typedef.h index 97b7e709..793d2c0c 100644 --- a/code/opus-1.2.1/silk/typedef.h +++ b/code/opus-1.5.2/silk/typedef.h @@ -67,6 +67,9 @@ __attribute__((noreturn)) static OPUS_INLINE void _silk_fatal(const char *str, const char *file, int line) { fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); +#if defined(_MSC_VER) + _set_abort_behavior( 0, _WRITE_ABORT_MSG); +#endif abort(); } # define silk_assert(COND) {if (!(COND)) {silk_fatal("assertion failed: " #COND);}} diff --git a/code/opus-1.5.2/silk/x86/NSQ_del_dec_avx2.c b/code/opus-1.5.2/silk/x86/NSQ_del_dec_avx2.c new file mode 100644 index 00000000..21f00c2d --- /dev/null +++ b/code/opus-1.5.2/silk/x86/NSQ_del_dec_avx2.c @@ -0,0 +1,1074 @@ +/*********************************************************************** +Copyright (c) 2021 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef OPUS_CHECK_ASM +#include +#endif + +#include "opus_defines.h" +#include + +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" +#include "celt/x86/x86cpu.h" + +/* Returns TRUE if all assumptions met */ +static OPUS_INLINE int verify_assumptions(const silk_encoder_state *psEncC) +{ + /* This optimization is based on these assumptions */ + /* These assumptions are fundamental and hence assert are */ + /* used. Should any assert triggers, we have to re-visit */ + /* all related code to make sure it still functions the */ + /* same as the C implementation. */ + silk_assert(MAX_DEL_DEC_STATES <= 4 && + MAX_FRAME_LENGTH % 4 == 0 && + MAX_SUB_FRAME_LENGTH % 4 == 0 && + LTP_MEM_LENGTH_MS % 4 == 0 ); + silk_assert(psEncC->fs_kHz == 8 || + psEncC->fs_kHz == 12 || + psEncC->fs_kHz == 16 ); + silk_assert(psEncC->nb_subfr <= MAX_NB_SUBFR && + psEncC->nb_subfr > 0 ); + silk_assert(psEncC->nStatesDelayedDecision <= MAX_DEL_DEC_STATES && + psEncC->nStatesDelayedDecision > 0 ); + silk_assert(psEncC->ltp_mem_length == psEncC->fs_kHz * LTP_MEM_LENGTH_MS); + + /* Regressions were observed on certain AMD Zen CPUs when */ + /* nStatesDelayedDecision is 1 or 2. Ideally we should detect */ + /* these CPUs and enable this optimization on others; however, */ + /* there is no good way to do so under current OPUS framework. */ + return psEncC->nStatesDelayedDecision == 3 || + psEncC->nStatesDelayedDecision == 4; +} + +/* Intrinsics not defined on MSVC */ +#ifdef _MSC_VER +#include +static inline int __builtin_sadd_overflow(opus_int32 a, opus_int32 b, opus_int32* res) +{ + *res = a+b; + return (*res ^ a) & (*res ^ b) & 0x80000000; +} +static inline int __builtin_ctz(unsigned int x) +{ + DWORD res = 0; + return _BitScanForward(&res, x) ? res : 32; +} +#endif + +static OPUS_INLINE __m128i silk_cvtepi64_epi32_high(__m256i num) +{ + return _mm256_castsi256_si128(_mm256_permutevar8x32_epi32(num, _mm256_set_epi32(0, 0, 0, 0, 7, 5, 3, 1))); +} + +static OPUS_INLINE opus_int16 silk_sat16(opus_int32 num) +{ + num = num > silk_int16_MAX ? silk_int16_MAX : num; + num = num < silk_int16_MIN ? silk_int16_MIN : num; + return num; +} + +static OPUS_INLINE opus_int32 silk_sar_round_32(opus_int32 a, int bits) +{ + silk_assert(bits > 0 && bits < 31); + a += 1 << (bits-1); + return a >> bits; +} + +static OPUS_INLINE opus_int64 silk_sar_round_smulww(opus_int32 a, opus_int32 b, int bits) +{ + silk_assert(bits > 0 && bits < 63); +#ifdef OPUS_CHECK_ASM + return silk_RSHIFT_ROUND(silk_SMULWW(a, b), bits); +#else + /* This code is more correct, but it won't overflow like the C code in some rare cases. */ + silk_assert(bits > 0 && bits < 63); + opus_int64 t = ((opus_int64)a) * ((opus_int64)b); + bits += 16; + t += 1ull << (bits-1); + return t >> bits; +#endif +} + +static OPUS_INLINE opus_int32 silk_add_sat32(opus_int32 a, opus_int32 b) +{ + opus_int32 sum; + if (__builtin_sadd_overflow(a, b, &sum)) + { + return a >= 0 ? silk_int32_MAX : silk_int32_MIN; + } + return sum; +} + +static OPUS_INLINE __m128i silk_mm_srai_round_epi32(__m128i a, int bits) +{ + silk_assert(bits > 0 && bits < 31); + return _mm_srai_epi32(_mm_add_epi32(a, _mm_set1_epi32(1 << (bits - 1))), bits); +} + +/* add/subtract with output saturated */ +static OPUS_INLINE __m128i silk_mm_add_sat_epi32(__m128i a, __m128i b) +{ + __m128i r = _mm_add_epi32(a, b); + __m128i OF = _mm_and_si128(_mm_xor_si128(a, r), _mm_xor_si128(b, r)); /* OF = (sum ^ a) & (sum ^ b) */ + __m128i SAT = _mm_add_epi32(_mm_srli_epi32(a, 31), _mm_set1_epi32(0x7FFFFFFF)); /* SAT = (a >> 31) + 0x7FFFFFFF */ + return _mm_blendv_epi8(r, SAT, _mm_srai_epi32(OF, 31)); +} +static OPUS_INLINE __m128i silk_mm_sub_sat_epi32(__m128i a, __m128i b) +{ + __m128i r = _mm_sub_epi32(a, b); + __m128i OF = _mm_andnot_si128(_mm_xor_si128(b, r), _mm_xor_si128(a, r)); /* OF = (sum ^ a) & (sum ^ ~b) = (sum ^ a) & ~(sum ^ b) */ + __m128i SAT = _mm_add_epi32(_mm_srli_epi32(a, 31), _mm_set1_epi32(0x7FFFFFFF)); /* SAT = (a >> 31) + 0x7FFFFFFF */ + return _mm_blendv_epi8(r, SAT, _mm_srai_epi32(OF, 31)); +} +static OPUS_INLINE __m256i silk_mm256_sub_sat_epi32(__m256i a, __m256i b) +{ + __m256i r = _mm256_sub_epi32(a, b); + __m256i OF = _mm256_andnot_si256(_mm256_xor_si256(b, r), _mm256_xor_si256(a, r)); /* OF = (sum ^ a) & (sum ^ ~b) = (sum ^ a) & ~(sum ^ b) */ + __m256i SAT = _mm256_add_epi32(_mm256_srli_epi32(a, 31), _mm256_set1_epi32(0x7FFFFFFF)); /* SAT = (a >> 31) + 0x7FFFFFFF */ + return _mm256_blendv_epi8(r, SAT, _mm256_srai_epi32(OF, 31)); +} + +static OPUS_INLINE __m128i silk_mm_limit_epi32(__m128i num, opus_int32 limit1, opus_int32 limit2) +{ + opus_int32 lo = limit1 < limit2 ? limit1 : limit2; + opus_int32 hi = limit1 > limit2 ? limit1 : limit2; + + num = _mm_min_epi32(num, _mm_set1_epi32(hi)); + num = _mm_max_epi32(num, _mm_set1_epi32(lo)); + return num; +} + +/* cond < 0 ? -num : num */ +static OPUS_INLINE __m128i silk_mm_sign_epi32(__m128i num, __m128i cond) +{ + return _mm_sign_epi32(num, _mm_or_si128(cond, _mm_set1_epi32(1))); +} +static OPUS_INLINE __m256i silk_mm256_sign_epi32(__m256i num, __m256i cond) +{ + return _mm256_sign_epi32(num, _mm256_or_si256(cond, _mm256_set1_epi32(1))); +} + +/* (a32 * b32) >> 16 */ +static OPUS_INLINE __m128i silk_mm_smulww_epi32(__m128i a, opus_int32 b) +{ + return silk_cvtepi64_epi32_high(_mm256_slli_epi64(_mm256_mul_epi32(_mm256_cvtepi32_epi64(a), _mm256_set1_epi32(b)), 16)); +} + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +static OPUS_INLINE __m128i silk_mm_smulwb_epi32(__m128i a, opus_int32 b) +{ + return silk_cvtepi64_epi32_high(_mm256_mul_epi32(_mm256_cvtepi32_epi64(a), _mm256_set1_epi32(silk_LSHIFT(b, 16)))); +} + +/* (opus_int32)((opus_int16)(a3))) * (opus_int32)((opus_int16)(b32)) output have to be 32bit int */ +static OPUS_INLINE __m256i silk_mm256_smulbb_epi32(__m256i a, __m256i b) +{ + const char FF = (char)0xFF; + __m256i msk = _mm256_set_epi8( + FF, FF, FF, FF, FF, FF, FF, FF, 13, 12, 9, 8, 5, 4, 1, 0, + FF, FF, FF, FF, FF, FF, FF, FF, 13, 12, 9, 8, 5, 4, 1, 0); + __m256i lo = _mm256_mullo_epi16(a, b); + __m256i hi = _mm256_mulhi_epi16(a, b); + lo = _mm256_shuffle_epi8(lo, msk); + hi = _mm256_shuffle_epi8(hi, msk); + return _mm256_unpacklo_epi16(lo, hi); +} + +static OPUS_INLINE __m256i silk_mm256_reverse_epi32(__m256i v) +{ + v = _mm256_shuffle_epi32(v, 0x1B); + v = _mm256_permute4x64_epi64(v, 0x4E); + return v; +} + +static OPUS_INLINE opus_int32 silk_mm256_hsum_epi32(__m256i v) +{ + __m128i sum = _mm_add_epi32(_mm256_extracti128_si256(v, 1), _mm256_extracti128_si256(v, 0)); + sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0x4E)); + sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0xB1)); + return _mm_cvtsi128_si32(sum); +} + +static OPUS_INLINE __m128i silk_mm_hmin_epi32(__m128i num) +{ + num = _mm_min_epi32(num, _mm_shuffle_epi32(num, 0x4E)); /* 0123 -> 2301 */ + num = _mm_min_epi32(num, _mm_shuffle_epi32(num, 0xB1)); /* 0123 -> 1032 */ + return num; +} + +static OPUS_INLINE __m128i silk_mm_hmax_epi32(__m128i num) +{ + num = _mm_max_epi32(num, _mm_shuffle_epi32(num, 0x4E)); /* 0123 -> 2310 */ + num = _mm_max_epi32(num, _mm_shuffle_epi32(num, 0xB1)); /* 0123 -> 1032 */ + return num; +} + +static OPUS_INLINE __m128i silk_mm_mask_hmin_epi32(__m128i num, __m128i mask) +{ + num = _mm_blendv_epi8(num, _mm_set1_epi32(silk_int32_MAX), mask); + return silk_mm_hmin_epi32(num); +} + +static OPUS_INLINE __m128i silk_mm_mask_hmax_epi32(__m128i num, __m128i mask) +{ + num = _mm_blendv_epi8(num, _mm_set1_epi32(silk_int32_MIN), mask); + return silk_mm_hmax_epi32(num); +} + +static OPUS_INLINE __m128i silk_mm256_rand_epi32(__m128i seed) +{ + seed = _mm_mullo_epi32(seed, _mm_set1_epi32(RAND_MULTIPLIER)); + seed = _mm_add_epi32(seed, _mm_set1_epi32(RAND_INCREMENT)); + return seed; +} + +static OPUS_INLINE opus_int32 silk_index_of_first_equal_epi32(__m128i a, __m128i b) +{ + unsigned int mask = _mm_movemask_epi8(_mm_cmpeq_epi32(a, b)) & 0x1111; + silk_assert(mask != 0); + return __builtin_ctz(mask) >> 2; +} + +static __m128i silk_index_to_selector(opus_int32 index) +{ + silk_assert(index < 4); + index <<= 2; + return _mm_set_epi8( + index + 3, index + 2, index + 1, index + 0, + index + 3, index + 2, index + 1, index + 0, + index + 3, index + 2, index + 1, index + 0, + index + 3, index + 2, index + 1, index + 0); +} + +static opus_int32 silk_select_winner(__m128i num, __m128i selector) +{ + return _mm_cvtsi128_si32(_mm_shuffle_epi8(num, selector)); +} + +typedef struct +{ + __m128i RandState; + __m128i Q_Q10; + __m128i Xq_Q14; + __m128i Pred_Q15; + __m128i Shape_Q14; +} NSQ_del_dec_sample_struct; + +typedef struct +{ + __m128i sLPC_Q14[MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH]; + __m128i LF_AR_Q14; + __m128i Seed; + __m128i SeedInit; + __m128i RD_Q10; + __m128i Diff_Q14; + __m128i sAR2_Q14[MAX_SHAPE_LPC_ORDER]; + NSQ_del_dec_sample_struct Samples[DECISION_DELAY]; +} NSQ_del_dec_struct; + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_avx2( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct *psDelDec, /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[MAX_SUB_FRAME_LENGTH], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[MAX_NB_SUBFR], /* I */ + const opus_int pitchL[MAX_NB_SUBFR], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +); + +/*******************************************/ +/* LPC analysis filter */ +/* NB! State is kept internally and the */ +/* filter always starts with zero state */ +/* first d output samples are set to zero */ +/*******************************************/ +static OPUS_INLINE void silk_LPC_analysis_filter_avx2( + opus_int16 *out, /* O Output signal */ + const opus_int16 *in, /* I Input signal */ + const opus_int16 *B, /* I MA prediction coefficients, Q12 [order] */ + const opus_int32 len, /* I Signal length */ + const opus_int32 order /* I Filter order */ +); + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_avx2( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[DECISION_DELAY], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + __m128i MaskDelDec, /* I Mask of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +); + +void silk_NSQ_del_dec_avx2( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[MAX_NB_SUBFR], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[MAX_NB_SUBFR], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[MAX_NB_SUBFR], /* I Quantization step sizes */ + const opus_int32 pitchL[MAX_NB_SUBFR], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[MAX_FRAME_LENGTH]; + const opus_int8 *const pulses_a = pulses; + + silk_memcpy(&NSQ_c, NSQ, sizeof(NSQ_c)); + silk_memcpy(&psIndices_c, psIndices, sizeof(psIndices_c)); + silk_memcpy(pulses_c, pulses, sizeof(pulses_c)); + silk_NSQ_del_dec_c(psEncC, &NSQ_c, &psIndices_c, x16, pulses_c, PredCoef_Q12, LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, + pitchL, Lambda_Q10, LTP_scale_Q14); +#endif + + if (!verify_assumptions(psEncC)) + { + silk_NSQ_del_dec_c(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14); + return; + } + + opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; + opus_int last_smple_idx, smpl_buf_idx, decisionDelay; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL(opus_int32, sLTP_Q15); + VARDECL(opus_int16, sLTP); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + opus_int32 Gain_Q10; + opus_int32 x_sc_Q10[MAX_SUB_FRAME_LENGTH]; + opus_int32 delayedGain_Q10[DECISION_DELAY]; + NSQ_del_dec_struct psDelDec = {0}; + NSQ_del_dec_sample_struct *psSample; + __m128i RDmin_Q10, MaskDelDec, Winner_selector; + SAVE_STACK; + + MaskDelDec = _mm_cvtepi8_epi32(_mm_cvtsi32_si128(0xFFFFFF00ul << ((psEncC->nStatesDelayedDecision - 1) << 3))); + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert(NSQ->prev_gain_Q16 != 0); + psDelDec.Seed = _mm_and_si128( + _mm_add_epi32(_mm_set_epi32(3, 2, 1, 0), _mm_set1_epi32(psIndices->Seed)), + _mm_set1_epi32(3)); + psDelDec.SeedInit = psDelDec.Seed; + psDelDec.RD_Q10 = _mm_setzero_si128(); + psDelDec.LF_AR_Q14 = _mm_set1_epi32(NSQ->sLF_AR_shp_Q14); + psDelDec.Diff_Q14 = _mm_set1_epi32(NSQ->sDiff_shp_Q14); + psDelDec.Samples[0].Shape_Q14 = _mm_set1_epi32(NSQ->sLTP_shp_Q14[psEncC->ltp_mem_length - 1]); + for (i = 0; i < NSQ_LPC_BUF_LENGTH; i++) + { + psDelDec.sLPC_Q14[i] = _mm_set1_epi32(NSQ->sLPC_Q14[i]); + } + for (i = 0; i < MAX_SHAPE_LPC_ORDER; i++) + { + psDelDec.sAR2_Q14[i] = _mm_set1_epi32(NSQ->sAR2_Q14[i]); + } + + offset_Q10 = silk_Quantization_Offsets_Q10[psIndices->signalType >> 1][psIndices->quantOffsetType]; + smpl_buf_idx = 0; /* index of oldest samples */ + + decisionDelay = silk_min_int(DECISION_DELAY, psEncC->subfr_length); + + /* For voiced frames limit the decision delay to lower than the pitch lag */ + if (psIndices->signalType == TYPE_VOICED) + { + for (k = 0; k < psEncC->nb_subfr; k++) + { + decisionDelay = silk_min_int(decisionDelay, pitchL[k] - LTP_ORDER / 2 - 1); + } + } + else + { + if (lag > 0) + { + decisionDelay = silk_min_int(decisionDelay, lag - LTP_ORDER / 2 - 1); + } + } + + if (psIndices->NLSFInterpCoef_Q2 == 4) + { + LSF_interpolation_flag = 0; + } + else + { + LSF_interpolation_flag = 1; + } + + ALLOC(sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32); + ALLOC(sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16); + /* Set up pointers to start of sub frame */ + pxq = &NSQ->xq[psEncC->ltp_mem_length]; + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + subfr = 0; + for (k = 0; k < psEncC->nb_subfr; k++) + { + A_Q12 = &PredCoef_Q12[((k >> 1) | (1 ^ LSF_interpolation_flag)) * MAX_LPC_ORDER]; + B_Q14 = <PCoef_Q14[k * LTP_ORDER]; + AR_shp_Q13 = &AR_Q13[k * MAX_SHAPE_LPC_ORDER]; + + /* Noise shape parameters */ + silk_assert(HarmShapeGain_Q14[k] >= 0); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if (psIndices->signalType == TYPE_VOICED) + { + /* Voiced */ + lag = pitchL[k]; + + /* Re-whitening */ + if ((k & (3 ^ (LSF_interpolation_flag << 1))) == 0) + { + if (k == 2) + { + /* RESET DELAYED DECISIONS */ + /* Find winner */ + RDmin_Q10 = silk_mm_mask_hmin_epi32(psDelDec.RD_Q10, MaskDelDec); + Winner_ind = silk_index_of_first_equal_epi32(RDmin_Q10, psDelDec.RD_Q10); + Winner_selector = silk_index_to_selector(Winner_ind); + psDelDec.RD_Q10 = _mm_add_epi32( + psDelDec.RD_Q10, + _mm_blendv_epi8( + _mm_set1_epi32(silk_int32_MAX >> 4), + _mm_setzero_si128(), + _mm_cvtepi8_epi32(_mm_cvtsi32_si128(0xFFU << (unsigned)(Winner_ind << 3))))); + + /* Copy final part of signals from winner state to output and long-term filter states */ + last_smple_idx = smpl_buf_idx + decisionDelay; + for (i = 0; i < decisionDelay; i++) + { + last_smple_idx = (last_smple_idx + DECISION_DELAY - 1) % DECISION_DELAY; + psSample = &psDelDec.Samples[last_smple_idx]; + pulses[i - decisionDelay] = + (opus_int8)silk_sar_round_32(silk_select_winner(psSample->Q_Q10, Winner_selector), 10); + pxq[i - decisionDelay] = + silk_sat16((opus_int32)silk_sar_round_smulww(silk_select_winner(psSample->Xq_Q14, Winner_selector), Gains_Q16[1], 14)); + NSQ->sLTP_shp_Q14[NSQ->sLTP_shp_buf_idx - decisionDelay + i] = + silk_select_winner(psSample->Shape_Q14, Winner_selector); + } + + subfr = 0; + } + + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + silk_assert(start_idx > 0); + + silk_LPC_analysis_filter_avx2(&sLTP[start_idx], &NSQ->xq[start_idx + k * psEncC->subfr_length], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder); + + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + NSQ->rewhite_flag = 1; + } + } + + silk_nsq_del_dec_scale_states_avx2(psEncC, NSQ, &psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, + LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay); + + silk_noise_shape_quantizer_del_dec_avx2(NSQ, &psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, + delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[k], LF_shp_Q14[k], + Gains_Q16[k], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, + psEncC->predictLPCOrder, psEncC->warping_Q16, MaskDelDec, &smpl_buf_idx, decisionDelay); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Find winner */ + RDmin_Q10 = silk_mm_mask_hmin_epi32(psDelDec.RD_Q10, MaskDelDec); + Winner_selector = silk_index_to_selector(silk_index_of_first_equal_epi32(RDmin_Q10, psDelDec.RD_Q10)); + + /* Copy final part of signals from winner state to output and long-term filter states */ + psIndices->Seed = silk_select_winner(psDelDec.SeedInit, Winner_selector); + last_smple_idx = smpl_buf_idx + decisionDelay; + Gain_Q10 = Gains_Q16[psEncC->nb_subfr - 1] >> 6; + for (i = 0; i < decisionDelay; i++) + { + last_smple_idx = (last_smple_idx + DECISION_DELAY - 1) % DECISION_DELAY; + psSample = &psDelDec.Samples[last_smple_idx]; + + pulses[i - decisionDelay] = + (opus_int8)silk_sar_round_32(silk_select_winner(psSample->Q_Q10, Winner_selector), 10); + pxq[i - decisionDelay] = + silk_sat16((opus_int32)silk_sar_round_smulww(silk_select_winner(psSample->Xq_Q14, Winner_selector), Gain_Q10, 8)); + NSQ->sLTP_shp_Q14[NSQ->sLTP_shp_buf_idx - decisionDelay + i] = + silk_select_winner(psSample->Shape_Q14, Winner_selector); + } + for (i = 0; i < NSQ_LPC_BUF_LENGTH; i++) + { + NSQ->sLPC_Q14[i] = silk_select_winner(psDelDec.sLPC_Q14[i], Winner_selector); + } + for (i = 0; i < MAX_SHAPE_LPC_ORDER; i++) + { + NSQ->sAR2_Q14[i] = silk_select_winner(psDelDec.sAR2_Q14[i], Winner_selector); + } + + /* Update states */ + NSQ->sLF_AR_shp_Q14 = silk_select_winner(psDelDec.LF_AR_Q14, Winner_selector); + NSQ->sDiff_shp_Q14 = silk_select_winner(psDelDec.Diff_Q14, Winner_selector); + NSQ->lagPrev = pitchL[psEncC->nb_subfr - 1]; + + /* Save quantized speech signal */ + silk_memmove(NSQ->xq, &NSQ->xq[psEncC->frame_length], psEncC->ltp_mem_length * sizeof(opus_int16)); + silk_memmove(NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[psEncC->frame_length], psEncC->ltp_mem_length * sizeof(opus_int32)); + +#ifdef OPUS_CHECK_ASM + silk_assert(!memcmp(&NSQ_c, NSQ, sizeof(NSQ_c))); + silk_assert(!memcmp(&psIndices_c, psIndices, sizeof(psIndices_c))); + silk_assert(!memcmp(pulses_c, pulses_a, sizeof(pulses_c))); +#endif + + RESTORE_STACK; +} + +static OPUS_INLINE __m128i silk_noise_shape_quantizer_short_prediction_x4(const __m128i *buf32, const opus_int16 *coef16, opus_int order) +{ + __m256i out; + silk_assert(order == 10 || order == 16); + + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + out = _mm256_set1_epi32(order >> 1); + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-0]), _mm256_set1_epi32(silk_LSHIFT(coef16[0], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-1]), _mm256_set1_epi32(silk_LSHIFT(coef16[1], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-2]), _mm256_set1_epi32(silk_LSHIFT(coef16[2], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-3]), _mm256_set1_epi32(silk_LSHIFT(coef16[3], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-4]), _mm256_set1_epi32(silk_LSHIFT(coef16[4], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-5]), _mm256_set1_epi32(silk_LSHIFT(coef16[5], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-6]), _mm256_set1_epi32(silk_LSHIFT(coef16[6], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-7]), _mm256_set1_epi32(silk_LSHIFT(coef16[7], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-8]), _mm256_set1_epi32(silk_LSHIFT(coef16[8], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-9]), _mm256_set1_epi32(silk_LSHIFT(coef16[9], 16)))); /* High DWORD */ + + if (order == 16) + { + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-10]), _mm256_set1_epi32(silk_LSHIFT(coef16[10], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-11]), _mm256_set1_epi32(silk_LSHIFT(coef16[11], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-12]), _mm256_set1_epi32(silk_LSHIFT(coef16[12], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-13]), _mm256_set1_epi32(silk_LSHIFT(coef16[13], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-14]), _mm256_set1_epi32(silk_LSHIFT(coef16[14], 16)))); /* High DWORD */ + out = _mm256_add_epi32(out, _mm256_mul_epi32(_mm256_cvtepi32_epi64(buf32[-15]), _mm256_set1_epi32(silk_LSHIFT(coef16[15], 16)))); /* High DWORD */ + } + return silk_cvtepi64_epi32_high(out); +} + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_avx2( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct *psDelDec, /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[DECISION_DELAY], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + __m128i MaskDelDec, /* I Mask of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +) +{ + int i; + opus_int32 *shp_lag_ptr = &NSQ->sLTP_shp_Q14[NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2]; + opus_int32 *pred_lag_ptr = &sLTP_Q15[NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2]; + opus_int32 Gain_Q10 = Gain_Q16 >> 6; + + for (i = 0; i < length; i++) + { + /* Perform common calculations used in all states */ + /* NSQ_sample_struct */ + /* Low 128 bits => 1st set */ + /* High 128 bits => 2nd set */ + int j; + __m256i SS_Q_Q10; + __m256i SS_RD_Q10; + __m256i SS_xq_Q14; + __m256i SS_LF_AR_Q14; + __m256i SS_Diff_Q14; + __m256i SS_sLTP_shp_Q14; + __m256i SS_LPC_exc_Q14; + __m256i exc_Q14; + __m256i q_Q10, rr_Q10, rd_Q10; + __m256i mask; + __m128i LPC_pred_Q14, n_AR_Q14; + __m128i RDmin_Q10, RDmax_Q10; + __m128i n_LF_Q14; + __m128i r_Q10, q1_Q0, q1_Q10, q2_Q10; + __m128i Winner_rand_state, Winner_selector; + __m128i tmp0, tmp1; + NSQ_del_dec_sample_struct *psLastSample, *psSample; + opus_int32 RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 LTP_pred_Q14, n_LTP_Q14; + + /* Long-term prediction */ + if (signalType == TYPE_VOICED) + { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q14 = 2; + LTP_pred_Q14 += silk_SMULWB(pred_lag_ptr[-0], b_Q14[0]); + LTP_pred_Q14 += silk_SMULWB(pred_lag_ptr[-1], b_Q14[1]); + LTP_pred_Q14 += silk_SMULWB(pred_lag_ptr[-2], b_Q14[2]); + LTP_pred_Q14 += silk_SMULWB(pred_lag_ptr[-3], b_Q14[3]); + LTP_pred_Q14 += silk_SMULWB(pred_lag_ptr[-4], b_Q14[4]); + LTP_pred_Q14 = silk_LSHIFT(LTP_pred_Q14, 1); /* Q13 -> Q14 */ + pred_lag_ptr++; + } + else + { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if (lag > 0) + { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_add_sat32(shp_lag_ptr[0], shp_lag_ptr[-2]); + n_LTP_Q14 = silk_SMULWB(n_LTP_Q14, HarmShapeFIRPacked_Q14); + n_LTP_Q14 = n_LTP_Q14 + silk_SMULWT(shp_lag_ptr[-1], HarmShapeFIRPacked_Q14); + n_LTP_Q14 = LTP_pred_Q14 - (silk_LSHIFT(n_LTP_Q14, 2)); /* Q12 -> Q14 */ + shp_lag_ptr++; + } + else + { + n_LTP_Q14 = 0; + } + + /* BEGIN Updating Delayed Decision States */ + + /* Generate dither */ + psDelDec->Seed = silk_mm256_rand_epi32(psDelDec->Seed); + + /* Short-term prediction */ + LPC_pred_Q14 = silk_noise_shape_quantizer_short_prediction_x4(&psDelDec->sLPC_Q14[NSQ_LPC_BUF_LENGTH - 1 + i], a_Q12, predictLPCOrder); + LPC_pred_Q14 = _mm_slli_epi32(LPC_pred_Q14, 4); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + silk_assert(shapingLPCOrder > 0); + silk_assert((shapingLPCOrder & 1) == 0); /* check that order is even */ + /* Output of lowpass section */ + tmp0 = _mm_add_epi32(psDelDec->Diff_Q14, silk_mm_smulwb_epi32(psDelDec->sAR2_Q14[0], warping_Q16)); + n_AR_Q14 = _mm_set1_epi32(shapingLPCOrder >> 1); + for (j = 0; j < shapingLPCOrder - 1; j++) + { + /* Output of allpass section */ + tmp1 = psDelDec->sAR2_Q14[j]; + psDelDec->sAR2_Q14[j] = tmp0; + n_AR_Q14 = _mm_add_epi32(n_AR_Q14, silk_mm_smulwb_epi32(tmp0, AR_shp_Q13[j])); + tmp0 = _mm_add_epi32(tmp1, silk_mm_smulwb_epi32(_mm_sub_epi32(psDelDec->sAR2_Q14[j + 1], tmp0), warping_Q16)); + } + psDelDec->sAR2_Q14[shapingLPCOrder - 1] = tmp0; + n_AR_Q14 = _mm_add_epi32(n_AR_Q14, silk_mm_smulwb_epi32(tmp0, AR_shp_Q13[shapingLPCOrder - 1])); + + n_AR_Q14 = _mm_slli_epi32(n_AR_Q14, 1); /* Q11 -> Q12 */ + n_AR_Q14 = _mm_add_epi32(n_AR_Q14, silk_mm_smulwb_epi32(psDelDec->LF_AR_Q14, Tilt_Q14)); /* Q12 */ + n_AR_Q14 = _mm_slli_epi32(n_AR_Q14, 2); /* Q12 -> Q14 */ + + tmp0 = silk_mm_smulwb_epi32(psDelDec->Samples[*smpl_buf_idx].Shape_Q14, LF_shp_Q14); /* Q12 */ + tmp1 = silk_mm_smulwb_epi32(psDelDec->LF_AR_Q14, LF_shp_Q14 >> 16); /* Q12 */ + n_LF_Q14 = _mm_add_epi32(tmp0, tmp1); /* Q12 */ + n_LF_Q14 = _mm_slli_epi32(n_LF_Q14, 2); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp0 = silk_mm_add_sat_epi32(n_AR_Q14, n_LF_Q14); /* Q14 */ + tmp1 = _mm_add_epi32(_mm_set1_epi32(n_LTP_Q14), LPC_pred_Q14); /* Q13 */ + tmp0 = silk_mm_sub_sat_epi32(tmp1, tmp0); /* Q13 */ + tmp0 = silk_mm_srai_round_epi32(tmp0, 4); /* Q10 */ + + r_Q10 = _mm_sub_epi32(_mm_set1_epi32(x_Q10[i]), tmp0); /* residual error Q10 */ + + /* Flip sign depending on dither */ + r_Q10 = silk_mm_sign_epi32(r_Q10, psDelDec->Seed); + r_Q10 = silk_mm_limit_epi32(r_Q10, -(31 << 10), 30 << 10); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = _mm_sub_epi32(r_Q10, _mm_set1_epi32(offset_Q10)); + q1_Q0 = _mm_srai_epi32(q1_Q10, 10); + if (Lambda_Q10 > 2048) + { + /* For aggressive RDO, the bias becomes more than one pulse. */ + tmp0 = _mm_sub_epi32(_mm_abs_epi32(q1_Q10), _mm_set1_epi32(Lambda_Q10 / 2 - 512)); /* rdo_offset */ + q1_Q0 = _mm_srai_epi32(q1_Q10, 31); + tmp1 = _mm_cmpgt_epi32(tmp0, _mm_setzero_si128()); + tmp0 = _mm_srai_epi32(silk_mm_sign_epi32(tmp0, q1_Q10), 10); + q1_Q0 = _mm_blendv_epi8(q1_Q0, tmp0, tmp1); + } + + tmp0 = _mm_sign_epi32(_mm_set1_epi32(QUANT_LEVEL_ADJUST_Q10), q1_Q0); + q1_Q10 = _mm_sub_epi32(_mm_slli_epi32(q1_Q0, 10), tmp0); + q1_Q10 = _mm_add_epi32(q1_Q10, _mm_set1_epi32(offset_Q10)); + + /* check if q1_Q0 is 0 or -1 */ + tmp0 = _mm_add_epi32(_mm_srli_epi32(q1_Q0, 31), q1_Q0); + tmp1 = _mm_cmpeq_epi32(tmp0, _mm_setzero_si128()); + tmp0 = _mm_blendv_epi8(_mm_set1_epi32(1024), _mm_set1_epi32(1024 - QUANT_LEVEL_ADJUST_Q10), tmp1); + q2_Q10 = _mm_add_epi32(q1_Q10, tmp0); + q_Q10 = _mm256_set_m128i(q2_Q10, q1_Q10); + + rr_Q10 = _mm256_sub_epi32(_mm256_broadcastsi128_si256(r_Q10), q_Q10); + rd_Q10 = _mm256_abs_epi32(q_Q10); + rr_Q10 = silk_mm256_smulbb_epi32(rr_Q10, rr_Q10); + rd_Q10 = silk_mm256_smulbb_epi32(rd_Q10, _mm256_set1_epi32(Lambda_Q10)); + rd_Q10 = _mm256_add_epi32(rd_Q10, rr_Q10); + rd_Q10 = _mm256_srai_epi32(rd_Q10, 10); + + mask = _mm256_broadcastsi128_si256(_mm_cmplt_epi32(_mm256_extracti128_si256(rd_Q10, 0), _mm256_extracti128_si256(rd_Q10, 1))); + SS_RD_Q10 = _mm256_add_epi32( + _mm256_broadcastsi128_si256(psDelDec->RD_Q10), + _mm256_blendv_epi8( + _mm256_permute2x128_si256(rd_Q10, rd_Q10, 0x1), + rd_Q10, + mask)); + SS_Q_Q10 = _mm256_blendv_epi8( + _mm256_permute2x128_si256(q_Q10, q_Q10, 0x1), + q_Q10, + mask); + + /* Update states for best and second best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_mm256_sign_epi32(_mm256_slli_epi32(SS_Q_Q10, 4), _mm256_broadcastsi128_si256(psDelDec->Seed)); + + /* Add predictions */ + exc_Q14 = _mm256_add_epi32(exc_Q14, _mm256_set1_epi32(LTP_pred_Q14)); + SS_LPC_exc_Q14 = _mm256_slli_epi32(exc_Q14, 1); + SS_xq_Q14 = _mm256_add_epi32(exc_Q14, _mm256_broadcastsi128_si256(LPC_pred_Q14)); + + /* Update states */ + SS_Diff_Q14 = _mm256_sub_epi32(SS_xq_Q14, _mm256_set1_epi32(silk_LSHIFT(x_Q10[i], 4))); + SS_LF_AR_Q14 = _mm256_sub_epi32(SS_Diff_Q14, _mm256_broadcastsi128_si256(n_AR_Q14)); + SS_sLTP_shp_Q14 = silk_mm256_sub_sat_epi32(SS_LF_AR_Q14, _mm256_broadcastsi128_si256(n_LF_Q14)); + + /* END Updating Delayed Decision States */ + + *smpl_buf_idx = (*smpl_buf_idx + DECISION_DELAY - 1) % DECISION_DELAY; + last_smple_idx = (*smpl_buf_idx + decisionDelay) % DECISION_DELAY; + psLastSample = &psDelDec->Samples[last_smple_idx]; + + /* Find winner */ + RDmin_Q10 = silk_mm_mask_hmin_epi32(_mm256_castsi256_si128(SS_RD_Q10), MaskDelDec); + Winner_selector = silk_index_to_selector(silk_index_of_first_equal_epi32(RDmin_Q10, _mm256_castsi256_si128(SS_RD_Q10))); + + /* Increase RD values of expired states */ + Winner_rand_state = _mm_shuffle_epi8(psLastSample->RandState, Winner_selector); + + SS_RD_Q10 = _mm256_blendv_epi8( + _mm256_add_epi32(SS_RD_Q10, _mm256_set1_epi32(silk_int32_MAX >> 4)), + SS_RD_Q10, + _mm256_broadcastsi128_si256(_mm_cmpeq_epi32(psLastSample->RandState, Winner_rand_state))); + + /* find worst in first set */ + RDmax_Q10 = silk_mm_mask_hmax_epi32(_mm256_extracti128_si256(SS_RD_Q10, 0), MaskDelDec); + /* find best in second set */ + RDmin_Q10 = silk_mm_mask_hmin_epi32(_mm256_extracti128_si256(SS_RD_Q10, 1), MaskDelDec); + + /* Replace a state if best from second set outperforms worst in first set */ + tmp0 = _mm_cmplt_epi32(RDmin_Q10, RDmax_Q10); + if (!_mm_test_all_zeros(tmp0, tmp0)) + { + int t; + RDmax_ind = silk_index_of_first_equal_epi32(RDmax_Q10, _mm256_extracti128_si256(SS_RD_Q10, 0)); + RDmin_ind = silk_index_of_first_equal_epi32(RDmin_Q10, _mm256_extracti128_si256(SS_RD_Q10, 1)); + tmp1 = _mm_cvtepi8_epi32(_mm_cvtsi32_si128(0xFFU << (unsigned)(RDmax_ind << 3))); + tmp0 = _mm_blendv_epi8( + _mm_set_epi8(0xF, 0xE, 0xD, 0xC, 0xB, 0xA, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0), + silk_index_to_selector(RDmin_ind), + tmp1); + for (t = i; t < MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH; t++) + { + psDelDec->sLPC_Q14[t] = _mm_shuffle_epi8(psDelDec->sLPC_Q14[t], tmp0); + } + psDelDec->Seed = _mm_shuffle_epi8(psDelDec->Seed, tmp0); + psDelDec->SeedInit = _mm_shuffle_epi8(psDelDec->SeedInit, tmp0); + for (t = 0; t < MAX_SHAPE_LPC_ORDER; t++) + { + psDelDec->sAR2_Q14[t] = _mm_shuffle_epi8(psDelDec->sAR2_Q14[t], tmp0); + } + for (t = 0; t < DECISION_DELAY; t++) + { + psDelDec->Samples[t].RandState = _mm_shuffle_epi8(psDelDec->Samples[t].RandState, tmp0); + psDelDec->Samples[t].Q_Q10 = _mm_shuffle_epi8(psDelDec->Samples[t].Q_Q10, tmp0); + psDelDec->Samples[t].Xq_Q14 = _mm_shuffle_epi8(psDelDec->Samples[t].Xq_Q14, tmp0); + psDelDec->Samples[t].Pred_Q15 = _mm_shuffle_epi8(psDelDec->Samples[t].Pred_Q15, tmp0); + psDelDec->Samples[t].Shape_Q14 = _mm_shuffle_epi8(psDelDec->Samples[t].Shape_Q14, tmp0); + } + mask = _mm256_castsi128_si256(_mm_blendv_epi8(_mm_set_epi32(0x3, 0x2, 0x1, 0x0), _mm_set1_epi32(RDmin_ind + 4), tmp1)); + SS_Q_Q10 = _mm256_permutevar8x32_epi32(SS_Q_Q10, mask); + SS_RD_Q10 = _mm256_permutevar8x32_epi32(SS_RD_Q10, mask); + SS_xq_Q14 = _mm256_permutevar8x32_epi32(SS_xq_Q14, mask); + SS_LF_AR_Q14 = _mm256_permutevar8x32_epi32(SS_LF_AR_Q14, mask); + SS_Diff_Q14 = _mm256_permutevar8x32_epi32(SS_Diff_Q14, mask); + SS_sLTP_shp_Q14 = _mm256_permutevar8x32_epi32(SS_sLTP_shp_Q14, mask); + SS_LPC_exc_Q14 = _mm256_permutevar8x32_epi32(SS_LPC_exc_Q14, mask); + } + + /* Write samples from winner to output and long-term filter states */ + if (subfr > 0 || i >= decisionDelay) + { + pulses[i - decisionDelay] = + (opus_int8)silk_sar_round_32(silk_select_winner(psLastSample->Q_Q10, Winner_selector), 10); + xq[i - decisionDelay] = + silk_sat16((opus_int32)silk_sar_round_smulww(silk_select_winner(psLastSample->Xq_Q14, Winner_selector), delayedGain_Q10[last_smple_idx], 8)); + NSQ->sLTP_shp_Q14[NSQ->sLTP_shp_buf_idx - decisionDelay] = + silk_select_winner(psLastSample->Shape_Q14, Winner_selector); + sLTP_Q15[NSQ->sLTP_buf_idx - decisionDelay] = + silk_select_winner(psLastSample->Pred_Q15, Winner_selector); + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + psSample = &psDelDec->Samples[*smpl_buf_idx]; + psDelDec->Seed = _mm_add_epi32(psDelDec->Seed, silk_mm_srai_round_epi32(_mm256_castsi256_si128(SS_Q_Q10), 10)); + psDelDec->LF_AR_Q14 = _mm256_castsi256_si128(SS_LF_AR_Q14); + psDelDec->Diff_Q14 = _mm256_castsi256_si128(SS_Diff_Q14); + psDelDec->sLPC_Q14[i + NSQ_LPC_BUF_LENGTH] = _mm256_castsi256_si128(SS_xq_Q14); + psDelDec->RD_Q10 = _mm256_castsi256_si128(SS_RD_Q10); + psSample->Xq_Q14 = _mm256_castsi256_si128(SS_xq_Q14); + psSample->Q_Q10 = _mm256_castsi256_si128(SS_Q_Q10); + psSample->Pred_Q15 = _mm256_castsi256_si128(SS_LPC_exc_Q14); + psSample->Shape_Q14 = _mm256_castsi256_si128(SS_sLTP_shp_Q14); + psSample->RandState = psDelDec->Seed; + delayedGain_Q10[*smpl_buf_idx] = Gain_Q10; + } + /* Update LPC states */ + for (i = 0; i < NSQ_LPC_BUF_LENGTH; i++) + { + psDelDec->sLPC_Q14[i] = (&psDelDec->sLPC_Q14[length])[i]; + } +} + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_avx2( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct *psDelDec, /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[MAX_SUB_FRAME_LENGTH], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[MAX_NB_SUBFR], /* I */ + const opus_int pitchL[MAX_NB_SUBFR], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +) +{ + int i; + opus_int lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + NSQ_del_dec_sample_struct *psSample; + + lag = pitchL[subfr]; + inv_gain_Q31 = silk_INVERSE32_varQ(silk_max(Gains_Q16[subfr], 1), 47); + silk_assert(inv_gain_Q31 != 0); + + /* Scale input */ + inv_gain_Q26 = silk_sar_round_32(inv_gain_Q31, 5); + for (i = 0; i < psEncC->subfr_length; i+=4) + { + __m256i x = _mm256_cvtepi16_epi64(_mm_loadu_si64(&x16[i])); + x = _mm256_slli_epi64(_mm256_mul_epi32(x, _mm256_set1_epi32(inv_gain_Q26)), 16); + _mm_storeu_si128((__m128i*)&x_sc_Q10[i], silk_cvtepi64_epi32_high(x)); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if (NSQ->rewhite_flag) + { + if (subfr == 0) + { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT(silk_SMULWB(inv_gain_Q31, LTP_scale_Q14), 2); + } + for (i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++) + { + silk_assert(i < MAX_FRAME_LENGTH); + sLTP_Q15[i] = silk_SMULWB(inv_gain_Q31, sLTP[i]); + } + } + + /* Adjust for changing gain */ + if (Gains_Q16[subfr] != NSQ->prev_gain_Q16) + { + gain_adj_Q16 = silk_DIV32_varQ(NSQ->prev_gain_Q16, Gains_Q16[subfr], 16); + + /* Scale long-term shaping state */ + for (i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i+=4) + { + opus_int32 *p = &NSQ->sLTP_shp_Q14[i]; + _mm_storeu_si128((__m128i*)p, silk_mm_smulww_epi32(_mm_loadu_si128((__m128i*)p), gain_adj_Q16)); + } + + /* Scale long-term prediction state */ + if (signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0) + { + for (i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay; i++) + { + sLTP_Q15[i] = ((opus_int64)sLTP_Q15[i]) * ((opus_int64)gain_adj_Q16) >> 16; + } + } + + /* Scale scalar states */ + psDelDec->LF_AR_Q14 = silk_mm_smulww_epi32(psDelDec->LF_AR_Q14, gain_adj_Q16); + psDelDec->Diff_Q14 = silk_mm_smulww_epi32(psDelDec->Diff_Q14, gain_adj_Q16); + + /* Scale short-term prediction and shaping states */ + for (i = 0; i < NSQ_LPC_BUF_LENGTH; i++) + { + psDelDec->sLPC_Q14[i] = silk_mm_smulww_epi32(psDelDec->sLPC_Q14[i], gain_adj_Q16); + } + for (i = 0; i < DECISION_DELAY; i++) + { + psSample = &psDelDec->Samples[i]; + psSample->Pred_Q15 = silk_mm_smulww_epi32(psSample->Pred_Q15, gain_adj_Q16); + psSample->Shape_Q14 = silk_mm_smulww_epi32(psSample->Shape_Q14, gain_adj_Q16); + } + for (i = 0; i < MAX_SHAPE_LPC_ORDER; i++) + { + psDelDec->sAR2_Q14[i] = silk_mm_smulww_epi32(psDelDec->sAR2_Q14[i], gain_adj_Q16); + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[subfr]; + } +} + +static OPUS_INLINE void silk_LPC_analysis_filter_avx2( + opus_int16 *out, /* O Output signal */ + const opus_int16 *in, /* I Input signal */ + const opus_int16 *B, /* I MA prediction coefficients, Q12 [order] */ + const opus_int32 len, /* I Signal length */ + const opus_int32 order /* I Filter order */ +) +{ + int i; + opus_int32 out32_Q12, out32; + silk_assert(order == 10 || order == 16); + + for(i = order; i < len; i++ ) + { + const opus_int16 *in_ptr = &in[ i ]; + /* Allowing wrap around so that two wraps can cancel each other. The rare + cases where the result wraps around can only be triggered by invalid streams*/ + + __m256i in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&in_ptr[-8])); + __m256i B_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)& B[0])); + __m256i sum = _mm256_mullo_epi32(in_v, silk_mm256_reverse_epi32(B_v)); + if (order > 10) + { + in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&in_ptr[-16])); + B_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&B [8])); + B_v = silk_mm256_reverse_epi32(B_v); + } + else + { + in_v = _mm256_cvtepi16_epi32(_mm_loadu_si32(&in_ptr[-10])); + B_v = _mm256_cvtepi16_epi32(_mm_loadu_si32(&B [8])); + B_v = _mm256_shuffle_epi32(B_v, 0x01); + } + sum = _mm256_add_epi32(sum, _mm256_mullo_epi32(in_v, B_v)); + + out32_Q12 = silk_mm256_hsum_epi32(sum); + + /* Subtract prediction */ + out32_Q12 = silk_SUB32_ovflw( silk_LSHIFT( (opus_int32)*in_ptr, 12 ), out32_Q12 ); + + /* Scale to Q0 */ + out32 = silk_sar_round_32(out32_Q12, 12); + + /* Saturate output */ + out[ i ] = silk_sat16(out32); + } + + /* Set first d output samples to zero */ + silk_memset( out, 0, order * sizeof( opus_int16 ) ); +} diff --git a/code/opus-1.2.1/silk/x86/NSQ_del_dec_sse.c b/code/opus-1.5.2/silk/x86/NSQ_del_dec_sse4_1.c similarity index 86% rename from code/opus-1.2.1/silk/x86/NSQ_del_dec_sse.c rename to code/opus-1.5.2/silk/x86/NSQ_del_dec_sse4_1.c index c5212bee..5937682d 100644 --- a/code/opus-1.2.1/silk/x86/NSQ_del_dec_sse.c +++ b/code/opus-1.5.2/silk/x86/NSQ_del_dec_sse4_1.c @@ -1,5 +1,5 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -46,6 +46,7 @@ typedef struct { opus_int32 Shape_Q14[ DECISION_DELAY ]; opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ]; opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; opus_int32 Seed; opus_int32 SeedInit; opus_int32 RD_Q10; @@ -56,6 +57,7 @@ typedef struct { opus_int32 RD_Q10; opus_int32 xq_Q14; opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; opus_int32 sLTP_shp_Q14; opus_int32 LPC_exc_Q14; } NSQ_sample_struct; @@ -66,7 +68,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ - const opus_int32 x_Q3[], /* I Input in Q3 */ + const opus_int16 x16[], /* I Input */ opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ @@ -112,21 +114,21 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( ); void silk_NSQ_del_dec_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) { opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; @@ -142,8 +144,39 @@ void silk_NSQ_del_dec_sse4_1( VARDECL( opus_int32, delayedGain_Q10 ); VARDECL( NSQ_del_dec_struct, psDelDec ); NSQ_del_dec_struct *psDD; +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[ MAX_FRAME_LENGTH ]; + const opus_int8 *const pulses_a = pulses; +#endif SAVE_STACK; +#ifdef OPUS_CHECK_ASM + ( void )pulses_a; + silk_memcpy( &NSQ_c, NSQ, sizeof( NSQ_c ) ); + silk_memcpy( &psIndices_c, psIndices, sizeof( psIndices_c ) ); + silk_assert( psEncC->nb_subfr * psEncC->subfr_length <= MAX_FRAME_LENGTH ); + silk_memcpy( pulses_c, pulses, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ); + silk_NSQ_del_dec_c( + psEncC, + &NSQ_c, + &psIndices_c, + x16, + pulses_c, + PredCoef_Q12, + LTPCoef_Q14, + AR_Q13, + HarmShapeGain_Q14, + Tilt_Q14, + LF_shp_Q14, + Gains_Q16, + pitchL, + Lambda_Q10, + LTP_scale_Q14 + ); +#endif + /* Set unvoiced lag to the previous one, overwrite later for voiced */ lag = NSQ->lagPrev; @@ -158,6 +191,7 @@ void silk_NSQ_del_dec_sse4_1( psDD->SeedInit = psDD->Seed; psDD->RD_Q10 = 0; psDD->LF_AR_Q14 = NSQ->sLF_AR_shp_Q14; + psDD->Diff_Q14 = NSQ->sDiff_shp_Q14; psDD->Shape_Q14[ 0 ] = NSQ->sLTP_shp_Q14[ psEncC->ltp_mem_length - 1 ]; silk_memcpy( psDD->sLPC_Q14, NSQ->sLPC_Q14, NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); silk_memcpy( psDD->sAR2_Q14, NSQ->sAR2_Q14, sizeof( NSQ->sAR2_Q14 ) ); @@ -185,8 +219,7 @@ void silk_NSQ_del_dec_sse4_1( LSF_interpolation_flag = 1; } - ALLOC( sLTP_Q15, - psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); @@ -198,7 +231,7 @@ void silk_NSQ_del_dec_sse4_1( for( k = 0; k < psEncC->nb_subfr; k++ ) { A_Q12 = &PredCoef_Q12[ ( ( k >> 1 ) | ( 1 - LSF_interpolation_flag ) ) * MAX_LPC_ORDER ]; B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; - AR_shp_Q13 = &AR2_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; /* Noise shape parameters */ silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); @@ -247,7 +280,7 @@ void silk_NSQ_del_dec_sse4_1( /* Rewhiten with new A coefs */ start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; - silk_assert( start_idx > 0 ); + celt_assert( start_idx > 0 ); silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); @@ -257,7 +290,7 @@ void silk_NSQ_del_dec_sse4_1( } } - silk_nsq_del_dec_scale_states_sse4_1( psEncC, NSQ, psDelDec, x_Q3, x_sc_Q10, sLTP, sLTP_Q15, k, + silk_nsq_del_dec_scale_states_sse4_1( psEncC, NSQ, psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, psEncC->nStatesDelayedDecision, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay ); silk_noise_shape_quantizer_del_dec_sse4_1( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, @@ -265,7 +298,7 @@ void silk_NSQ_del_dec_sse4_1( Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay ); - x_Q3 += psEncC->subfr_length; + x16 += psEncC->subfr_length; pulses += psEncC->subfr_length; pxq += psEncC->subfr_length; } @@ -288,6 +321,7 @@ void silk_NSQ_del_dec_sse4_1( for( i = 0; i < decisionDelay; i++ ) { last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gain_Q10 ), 8 ) ); @@ -298,11 +332,19 @@ void silk_NSQ_del_dec_sse4_1( /* Update states */ NSQ->sLF_AR_shp_Q14 = psDD->LF_AR_Q14; + NSQ->sDiff_shp_Q14 = psDD->Diff_Q14; NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; /* Save quantized speech signal */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( &NSQ_c, NSQ, sizeof( NSQ_c ) ) ); + silk_assert( !memcmp( &psIndices_c, psIndices, sizeof( psIndices_c ) ) ); + silk_assert( !memcmp( pulses_c, pulses_a, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ) ); +#endif + RESTORE_STACK; } @@ -345,6 +387,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; + int rdo_offset; + VARDECL( NSQ_sample_pair, psSampleState ); NSQ_del_dec_struct *psDD; NSQ_sample_struct *psSS; @@ -353,9 +397,11 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( __m128i b_Q12_0123, b_sr_Q12_0123; SAVE_STACK; - silk_assert( nStatesDelayedDecision > 0 ); + celt_assert( nStatesDelayedDecision > 0 ); ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); + rdo_offset = (Lambda_Q10 >> 1) - 512; + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); @@ -382,7 +428,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( LTP_pred_Q14 = 2; { __m128i tmpa, tmpb, pred_lag_ptr_tmp; - pred_lag_ptr_tmp = _mm_loadu_si128( (__m128i *)(&pred_lag_ptr[ -3 ] ) ); + pred_lag_ptr_tmp = _mm_loadu_si128( (__m128i *)(void*)(&pred_lag_ptr[ -3 ] ) ); pred_lag_ptr_tmp = _mm_shuffle_epi32( pred_lag_ptr_tmp, 0x1B ); tmpa = _mm_mul_epi32( pred_lag_ptr_tmp, b_Q12_0123 ); tmpa = _mm_srli_si128( tmpa, 2 ); @@ -407,8 +453,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( /* Long-term shaping */ if( lag > 0 ) { /* Symmetric, packed FIR coefficients */ - n_LTP_Q14 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); - n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ shp_lag_ptr++; } else { @@ -437,7 +483,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( tmpb = _mm_setzero_si128(); /* step 1 */ - psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -3 ] ) ); /* -3, -2 , -1, 0 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -3 ] ) ); /* -3, -2 , -1, 0 */ psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); /* 0, -1, -2, -3 */ tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_0123 ); /* 0, -1, -2, -3 * 0123 -> 0*0, 2*-2 */ @@ -451,7 +497,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); /* step 2 */ - psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -7 ] ) ); + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -7 ] ) ); psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_4567 ); tmpa = _mm_srli_epi64( tmpa, 16 ); @@ -466,7 +512,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( if ( opus_likely( predictLPCOrder == 16 ) ) { /* step 3 */ - psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -11 ] ) ); + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -11 ] ) ); psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_89AB ); tmpa = _mm_srli_epi64( tmpa, 16 ); @@ -478,8 +524,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( psLPC_Q14_tmp = _mm_srli_epi64( psLPC_Q14_tmp, 16 ); tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); - /* setp 4 */ - psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -15 ] ) ); + /* step 4 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -15 ] ) ); psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_CDEF ); tmpa = _mm_srli_epi64( tmpa, 16 ); @@ -511,22 +557,22 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ /* Noise shape feedback */ - silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ /* Output of lowpass section */ - tmp2 = silk_SMLAWB( psLPC_Q14[ 0 ], psDD->sAR2_Q14[ 0 ], warping_Q16 ); + tmp2 = silk_SMLAWB( psDD->Diff_Q14, psDD->sAR2_Q14[ 0 ], warping_Q16 ); /* Output of allpass section */ - tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], psDD->sAR2_Q14[ 1 ] - tmp2, warping_Q16 ); + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ 1 ], tmp2), warping_Q16 ); psDD->sAR2_Q14[ 0 ] = tmp2; n_AR_Q14 = silk_RSHIFT( shapingLPCOrder, 1 ); n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ 0 ] ); /* Loop over allpass sections */ for( j = 2; j < shapingLPCOrder; j += 2 ) { /* Output of allpass section */ - tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], psDD->sAR2_Q14[ j + 0 ] - tmp1, warping_Q16 ); + tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 0 ], tmp1), warping_Q16 ); psDD->sAR2_Q14[ j - 1 ] = tmp1; n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ j - 1 ] ); /* Output of allpass section */ - tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], psDD->sAR2_Q14[ j + 1 ] - tmp2, warping_Q16 ); + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 1 ], tmp2), warping_Q16 ); psDD->sAR2_Q14[ j + 0 ] = tmp2; n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ j ] ); } @@ -543,9 +589,9 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( /* Input minus prediction plus noise feedback */ /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ - tmp1 = silk_ADD32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ - tmp2 = silk_ADD32( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ - tmp1 = silk_SUB32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_ADD_SAT32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32_ovflw( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB_SAT32( tmp2, tmp1 ); /* Q13 */ tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ @@ -559,6 +605,18 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( /* Find two quantization level candidates and measure their rate-distortion */ q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } if( q1_Q0 > 0 ) { q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); @@ -609,11 +667,12 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( /* Add predictions */ LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); - xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); /* Update states */ - sLF_AR_shp_Q14 = silk_SUB32( xq_Q14, n_AR_Q14 ); - psSS[ 0 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; psSS[ 0 ].xq_Q14 = xq_Q14; @@ -626,14 +685,14 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( exc_Q14 = -exc_Q14; } - /* Add predictions */ LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); - xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); /* Update states */ - sLF_AR_shp_Q14 = silk_SUB32( xq_Q14, n_AR_Q14 ); - psSS[ 1 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; psSS[ 1 ].xq_Q14 = xq_Q14; @@ -705,6 +764,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( psDD = &psDelDec[ k ]; psSS = &psSampleState[ k ][ 0 ]; psDD->LF_AR_Q14 = psSS->LF_AR_Q14; + psDD->Diff_Q14 = psSS->Diff_Q14; psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ] = psSS->xq_Q14; psDD->Xq_Q14[ *smpl_buf_idx ] = psSS->xq_Q14; psDD->Q_Q10[ *smpl_buf_idx ] = psSS->Q_Q10; @@ -728,7 +788,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ - const opus_int32 x_Q3[], /* I Input in Q3 */ + const opus_int16 x16[], /* I Input */ opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ @@ -742,51 +802,41 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( ) { opus_int i, k, lag; - opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q23; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; NSQ_del_dec_struct *psDD; - __m128i xmm_inv_gain_Q23, xmm_x_Q3_x2x0, xmm_x_Q3_x3x1; + __m128i xmm_inv_gain_Q26, xmm_x16_x2x0, xmm_x16_x3x1; lag = pitchL[ subfr ]; inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); - silk_assert( inv_gain_Q31 != 0 ); - /* Calculate gain adjustment factor */ - if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { - gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); - } else { - gain_adj_Q16 = (opus_int32)1 << 16; - } - /* Scale input */ - inv_gain_Q23 = silk_RSHIFT_ROUND( inv_gain_Q31, 8 ); + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); - /* prepare inv_gain_Q23 in packed 4 32-bits */ - xmm_inv_gain_Q23 = _mm_set1_epi32(inv_gain_Q23); + /* prepare inv_gain_Q26 in packed 4 32-bits */ + xmm_inv_gain_Q26 = _mm_set1_epi32(inv_gain_Q26); for( i = 0; i < psEncC->subfr_length - 3; i += 4 ) { - xmm_x_Q3_x2x0 = _mm_loadu_si128( (__m128i *)(&(x_Q3[ i ] ) ) ); + xmm_x16_x2x0 = OP_CVTEPI16_EPI32_M64( &(x16[ i ] ) ); + /* equal shift right 4 bytes*/ - xmm_x_Q3_x3x1 = _mm_shuffle_epi32( xmm_x_Q3_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_x16_x3x1 = _mm_shuffle_epi32( xmm_x16_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); - xmm_x_Q3_x2x0 = _mm_mul_epi32( xmm_x_Q3_x2x0, xmm_inv_gain_Q23 ); - xmm_x_Q3_x3x1 = _mm_mul_epi32( xmm_x_Q3_x3x1, xmm_inv_gain_Q23 ); + xmm_x16_x2x0 = _mm_mul_epi32( xmm_x16_x2x0, xmm_inv_gain_Q26 ); + xmm_x16_x3x1 = _mm_mul_epi32( xmm_x16_x3x1, xmm_inv_gain_Q26 ); - xmm_x_Q3_x2x0 = _mm_srli_epi64( xmm_x_Q3_x2x0, 16 ); - xmm_x_Q3_x3x1 = _mm_slli_epi64( xmm_x_Q3_x3x1, 16 ); + xmm_x16_x2x0 = _mm_srli_epi64( xmm_x16_x2x0, 16 ); + xmm_x16_x3x1 = _mm_slli_epi64( xmm_x16_x3x1, 16 ); - xmm_x_Q3_x2x0 = _mm_blend_epi16( xmm_x_Q3_x2x0, xmm_x_Q3_x3x1, 0xCC ); + xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC ); - _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ])), xmm_x_Q3_x2x0 ); + _mm_storeu_si128( (__m128i *)(void*)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 ); } for( ; i < psEncC->subfr_length; i++ ) { - x_sc_Q10[ i ] = silk_SMULWW( x_Q3[ i ], inv_gain_Q23 ); + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); } - /* Save inverse gain */ - NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; - /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ if( NSQ->rewhite_flag ) { if( subfr == 0 ) { @@ -800,7 +850,9 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( } /* Adjust for changing gain */ - if( gain_adj_Q16 != (opus_int32)1 << 16 ) { + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + /* Scale long-term shaping state */ { __m128i xmm_gain_adj_Q16, xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1; @@ -810,7 +862,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 3; i += 4 ) { - xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); + xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(void*)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); /* equal shift right 4 bytes*/ xmm_sLTP_shp_Q14_x3x1 = _mm_shuffle_epi32( xmm_sLTP_shp_Q14_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); @@ -822,7 +874,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( xmm_sLTP_shp_Q14_x2x0 = _mm_blend_epi16( xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1, 0xCC ); - _mm_storeu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); + _mm_storeu_si128( (__m128i *)(void*)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); } for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { @@ -841,6 +893,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( /* Scale scalar states */ psDD->LF_AR_Q14 = silk_SMULWW( gain_adj_Q16, psDD->LF_AR_Q14 ); + psDD->Diff_Q14 = silk_SMULWW( gain_adj_Q16, psDD->Diff_Q14 ); /* Scale short-term prediction and shaping states */ for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { @@ -855,5 +908,8 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( } } } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; } } diff --git a/code/opus-1.2.1/silk/x86/NSQ_sse.c b/code/opus-1.5.2/silk/x86/NSQ_sse4_1.c similarity index 77% rename from code/opus-1.2.1/silk/x86/NSQ_sse.c rename to code/opus-1.5.2/silk/x86/NSQ_sse4_1.c index 6a9e6e96..3c9aca7b 100644 --- a/code/opus-1.2.1/silk/x86/NSQ_sse.c +++ b/code/opus-1.5.2/silk/x86/NSQ_sse4_1.c @@ -1,5 +1,5 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -37,17 +37,17 @@ #include "stack_alloc.h" static OPUS_INLINE void silk_nsq_scale_states_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - const opus_int32 x_Q3[], /* I input in Q3 */ - opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ - const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ - opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ - opus_int subfr, /* I subframe number */ - const opus_int LTP_scale_Q14, /* I */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ - const opus_int signal_type /* I Signal type */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ ); static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( @@ -65,27 +65,28 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( opus_int Tilt_Q14, /* I Spectral tilt */ opus_int32 LF_shp_Q14, /* I */ opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ opus_int offset_Q10, /* I */ opus_int length, /* I Input length */ opus_int32 table[][4] /* I */ ); void silk_NSQ_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) { opus_int k, lag, start_idx, LSF_interpolation_flag; @@ -101,8 +102,41 @@ void silk_NSQ_sse4_1( opus_int32 tmp1; opus_int32 q1_Q10, q2_Q10, rd1_Q20, rd2_Q20; +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[ MAX_FRAME_LENGTH ]; + const opus_int8 *const pulses_a = pulses; +#endif + SAVE_STACK; +#ifdef OPUS_CHECK_ASM + ( void )pulses_a; + silk_memcpy( &NSQ_c, NSQ, sizeof( NSQ_c ) ); + silk_memcpy( &psIndices_c, psIndices, sizeof( psIndices_c ) ); + silk_assert( psEncC->nb_subfr * psEncC->subfr_length <= MAX_FRAME_LENGTH ); + silk_memcpy( pulses_c, pulses, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ); + + silk_NSQ_c( + psEncC, + &NSQ_c, + &psIndices_c, + x16, + pulses_c, + PredCoef_Q12, + LTPCoef_Q14, + AR_Q13, + HarmShapeGain_Q14, + Tilt_Q14, + LF_shp_Q14, + Gains_Q16, + pitchL, + Lambda_Q10, + LTP_scale_Q14 + ); +#endif + NSQ->rand_seed = psIndices->Seed; /* Set unvoiced lag to the previous one, overwrite later for voiced */ @@ -172,8 +206,7 @@ void silk_NSQ_sse4_1( LSF_interpolation_flag = 1; } - ALLOC( sLTP_Q15, - psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); /* Set up pointers to start of sub frame */ @@ -183,7 +216,7 @@ void silk_NSQ_sse4_1( for( k = 0; k < psEncC->nb_subfr; k++ ) { A_Q12 = &PredCoef_Q12[ (( k >> 1 ) | ( 1 - LSF_interpolation_flag )) * MAX_LPC_ORDER ]; B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; - AR_shp_Q13 = &AR2_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; /* Noise shape parameters */ silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); @@ -199,7 +232,7 @@ void silk_NSQ_sse4_1( if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { /* Rewhiten with new A coefs */ start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; - silk_assert( start_idx > 0 ); + celt_assert( start_idx > 0 ); silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); @@ -209,12 +242,12 @@ void silk_NSQ_sse4_1( } } - silk_nsq_scale_states_sse4_1( psEncC, NSQ, x_Q3, x_sc_Q10, sLTP, sLTP_Q15, k, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType ); + silk_nsq_scale_states_sse4_1( psEncC, NSQ, x16, x_sc_Q10, sLTP, sLTP_Q15, k, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType ); if ( opus_likely( ( 10 == psEncC->shapingLPCOrder ) && ( 16 == psEncC->predictLPCOrder) ) ) { silk_noise_shape_quantizer_10_16_sse4_1( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, - AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], + AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, &(table[32]) ); } else @@ -224,7 +257,7 @@ void silk_NSQ_sse4_1( offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); } - x_Q3 += psEncC->subfr_length; + x16 += psEncC->subfr_length; pulses += psEncC->subfr_length; pxq += psEncC->subfr_length; } @@ -235,12 +268,19 @@ void silk_NSQ_sse4_1( /* Save quantized speech and noise shaping signals */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( &NSQ_c, NSQ, sizeof( NSQ_c ) ) ); + silk_assert( !memcmp( &psIndices_c, psIndices, sizeof( psIndices_c ) ) ); + silk_assert( !memcmp( pulses_c, pulses_a, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ) ); +#endif + RESTORE_STACK; } -/***********************************/ -/* silk_noise_shape_quantizer_10_16 */ -/***********************************/ +/************************************/ +/* silk_noise_shape_quantizer_10_16 */ +/************************************/ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( silk_nsq_state *NSQ, /* I/O NSQ state */ opus_int signalType, /* I Signal type */ @@ -256,6 +296,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( opus_int Tilt_Q14, /* I Spectral tilt */ opus_int32 LF_shp_Q14, /* I */ opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ opus_int offset_Q10, /* I */ opus_int length, /* I Input length */ opus_int32 table[][4] /* I */ @@ -264,7 +305,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( opus_int i; opus_int32 LTP_pred_Q13, LPC_pred_Q10, n_AR_Q12, n_LTP_Q13; opus_int32 n_LF_Q12, r_Q10, q1_Q0, q1_Q10, q2_Q10; - opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10, sDiff_shp_Q14; opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; opus_int32 *psLPC_Q14, *shp_lag_ptr, *pred_lag_ptr; @@ -279,6 +320,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( __m128i sAR2_Q14_hi_76543210, sAR2_Q14_lo_76543210; __m128i AR_shp_Q13_76543210; + int rdo_offset = (Lambda_Q10 >> 1) - 512; + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); @@ -288,27 +331,28 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( sLF_AR_shp_Q14 = NSQ->sLF_AR_shp_Q14; xq_Q14 = psLPC_Q14[ 0 ]; + sDiff_shp_Q14 = NSQ->sDiff_shp_Q14; LTP_pred_Q13 = 0; /* load a_Q12 */ xmm_one = _mm_set_epi8( 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 ); /* load a_Q12[0] - a_Q12[7] */ - a_Q12_01234567 = _mm_loadu_si128( (__m128i *)(&a_Q12[ 0 ] ) ); + a_Q12_01234567 = _mm_loadu_si128( (__m128i *)(void*)(&a_Q12[ 0 ] ) ); /* load a_Q12[ 8 ] - a_Q12[ 15 ] */ - a_Q12_89ABCDEF = _mm_loadu_si128( (__m128i *)(&a_Q12[ 8 ] ) ); + a_Q12_89ABCDEF = _mm_loadu_si128( (__m128i *)(void*)(&a_Q12[ 8 ] ) ); a_Q12_01234567 = _mm_shuffle_epi8( a_Q12_01234567, xmm_one ); a_Q12_89ABCDEF = _mm_shuffle_epi8( a_Q12_89ABCDEF, xmm_one ); /* load AR_shp_Q13 */ - AR_shp_Q13_76543210 = _mm_loadu_si128( (__m128i *)(&AR_shp_Q13[0] ) ); + AR_shp_Q13_76543210 = _mm_loadu_si128( (__m128i *)(void*)(&AR_shp_Q13[0] ) ); /* load psLPC_Q14 */ xmm_one = _mm_set_epi8(15, 14, 11, 10, 7, 6, 3, 2, 13, 12, 9, 8, 5, 4, 1, 0 ); - xmm_tempa = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[-16]) ); - xmm_tempb = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[-12]) ); + xmm_tempa = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[-16]) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[-12]) ); xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); @@ -316,8 +360,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( psLPC_Q14_hi_89ABCDEF = _mm_unpackhi_epi64( xmm_tempa, xmm_tempb ); psLPC_Q14_lo_89ABCDEF = _mm_unpacklo_epi64( xmm_tempa, xmm_tempb ); - xmm_tempa = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -8 ]) ); - xmm_tempb = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -4 ]) ); + xmm_tempa = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -8 ]) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(void*)(&psLPC_Q14[ -4 ]) ); xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); @@ -326,8 +370,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( psLPC_Q14_lo_01234567 = _mm_unpacklo_epi64( xmm_tempa, xmm_tempb ); /* load sAR2_Q14 */ - xmm_tempa = _mm_loadu_si128( (__m128i *)(&(NSQ->sAR2_Q14[ 0 ]) ) ); - xmm_tempb = _mm_loadu_si128( (__m128i *)(&(NSQ->sAR2_Q14[ 4 ]) ) ); + xmm_tempa = _mm_loadu_si128( (__m128i *)(void*)(&(NSQ->sAR2_Q14[ 0 ]) ) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(void*)(&(NSQ->sAR2_Q14[ 4 ]) ) ); xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); @@ -399,7 +443,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( b_Q14_0123 = _mm_shuffle_epi32( b_Q14_3210, 0x1B ); /* loaded: [0] [-1] [-2] [-3] */ - pred_lag_ptr_0123 = _mm_loadu_si128( (__m128i *)(&pred_lag_ptr[ -3 ] ) ); + pred_lag_ptr_0123 = _mm_loadu_si128( (__m128i *)(void*)(&pred_lag_ptr[ -3 ] ) ); /* shuffle to [-3] [-2] [-1] [0] and to new xmm */ xmm_tempa = _mm_shuffle_epi32( pred_lag_ptr_0123, 0x1B ); /*64-bit multiply, a[2] * b[-2], a[0] * b[0] */ @@ -430,8 +474,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( sAR2_Q14_hi_76543210 = _mm_slli_si128( sAR2_Q14_hi_76543210, 2 ); sAR2_Q14_lo_76543210 = _mm_slli_si128( sAR2_Q14_lo_76543210, 2 ); - sAR2_Q14_hi_76543210 = _mm_insert_epi16( sAR2_Q14_hi_76543210, (xq_Q14 >> 16), 0 ); - sAR2_Q14_lo_76543210 = _mm_insert_epi16( sAR2_Q14_lo_76543210, (xq_Q14), 0 ); + sAR2_Q14_hi_76543210 = _mm_insert_epi16( sAR2_Q14_hi_76543210, (sDiff_shp_Q14 >> 16), 0 ); + sAR2_Q14_lo_76543210 = _mm_insert_epi16( sAR2_Q14_lo_76543210, (sDiff_shp_Q14), 0 ); /* high part, use pmaddwd, results in 4 32-bit */ xmm_hi_07 = _mm_madd_epi16( sAR2_Q14_hi_76543210, AR_shp_Q13_76543210 ); @@ -462,14 +506,14 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); n_LF_Q12 = silk_SMLAWT( n_LF_Q12, sLF_AR_shp_Q14, LF_shp_Q14 ); - silk_assert( lag > 0 || signalType != TYPE_VOICED ); + celt_assert( lag > 0 || signalType != TYPE_VOICED ); /* Combine prediction and noise shaping signals */ tmp1 = silk_SUB32( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ tmp1 = silk_SUB32( tmp1, n_LF_Q12 ); /* Q12 */ if( lag > 0 ) { /* Symmetric, packed FIR coefficients */ - n_LTP_Q13 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); n_LTP_Q13 = silk_SMLAWT( n_LTP_Q13, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); n_LTP_Q13 = silk_LSHIFT( n_LTP_Q13, 1 ); shp_lag_ptr++; @@ -495,6 +539,18 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( /* Find two quantization level candidates and measure their rate-distortion */ q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } q1_Q10 = table[q1_Q0][0]; q2_Q10 = table[q1_Q0][1]; @@ -519,7 +575,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( /* Update states */ psLPC_Q14++; *psLPC_Q14 = xq_Q14; - sLF_AR_shp_Q14 = silk_SUB_LSHIFT32( xq_Q14, n_AR_Q12, 2 ); + NSQ->sDiff_shp_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_sc_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB_LSHIFT32( NSQ->sDiff_shp_Q14, n_AR_Q12, 2 ); NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB_LSHIFT32( sLF_AR_shp_Q14, n_LF_Q12, 2 ); sLTP_Q15[ NSQ->sLTP_buf_idx ] = silk_LSHIFT( LPC_exc_Q14, 1 ); @@ -538,8 +595,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( /* write back sAR2_Q14 */ xmm_tempa = _mm_unpackhi_epi16( sAR2_Q14_lo_76543210, sAR2_Q14_hi_76543210 ); xmm_tempb = _mm_unpacklo_epi16( sAR2_Q14_lo_76543210, sAR2_Q14_hi_76543210 ); - _mm_storeu_si128( (__m128i *)(&NSQ->sAR2_Q14[ 4 ]), xmm_tempa ); - _mm_storeu_si128( (__m128i *)(&NSQ->sAR2_Q14[ 0 ]), xmm_tempb ); + _mm_storeu_si128( (__m128i *)(void*)(&NSQ->sAR2_Q14[ 4 ]), xmm_tempa ); + _mm_storeu_si128( (__m128i *)(void*)(&NSQ->sAR2_Q14[ 0 ]), xmm_tempb ); /* xq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psLPC_Q14[ i ], Gain_Q10 ), 8 ) ); */ { @@ -555,8 +612,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( /* process xq */ for (i = 0; i < length - 7; i += 8) { - xmm_xq_Q14_3210 = _mm_loadu_si128( (__m128i *)(&(psLPC_Q14[ i + 0 ] ) ) ); - xmm_xq_Q14_7654 = _mm_loadu_si128( (__m128i *)(&(psLPC_Q14[ i + 4 ] ) ) ); + xmm_xq_Q14_3210 = _mm_loadu_si128( (__m128i *)(void*)(&(psLPC_Q14[ i + 0 ] ) ) ); + xmm_xq_Q14_7654 = _mm_loadu_si128( (__m128i *)(void*)(&(psLPC_Q14[ i + 4 ] ) ) ); /* equal shift right 4 bytes*/ xmm_xq_Q14_x3x1 = _mm_shuffle_epi32( xmm_xq_Q14_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); @@ -587,7 +644,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( xmm_xq_Q14_3210 = _mm_packs_epi32( xmm_xq_Q14_3210, xmm_xq_Q14_7654 ); /* save to xq */ - _mm_storeu_si128( (__m128i *)(&xq[ i ] ), xmm_xq_Q14_3210 ); + _mm_storeu_si128( (__m128i *)(void*)(&xq[ i ] ), xmm_xq_Q14_3210 ); } } for ( ; i < length; i++) @@ -600,64 +657,54 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( } static OPUS_INLINE void silk_nsq_scale_states_sse4_1( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - const opus_int32 x_Q3[], /* I input in Q3 */ - opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ - const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ - opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ - opus_int subfr, /* I subframe number */ - const opus_int LTP_scale_Q14, /* I */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ - const opus_int signal_type /* I Signal type */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ ) { opus_int i, lag; - opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q23; - __m128i xmm_inv_gain_Q23, xmm_x_Q3_x2x0, xmm_x_Q3_x3x1; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + __m128i xmm_inv_gain_Q26, xmm_x16_x2x0, xmm_x16_x3x1; lag = pitchL[ subfr ]; inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); silk_assert( inv_gain_Q31 != 0 ); - /* Calculate gain adjustment factor */ - if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { - gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); - } else { - gain_adj_Q16 = (opus_int32)1 << 16; - } - /* Scale input */ - inv_gain_Q23 = silk_RSHIFT_ROUND( inv_gain_Q31, 8 ); + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); - /* prepare inv_gain_Q23 in packed 4 32-bits */ - xmm_inv_gain_Q23 = _mm_set1_epi32(inv_gain_Q23); + /* prepare inv_gain_Q26 in packed 4 32-bits */ + xmm_inv_gain_Q26 = _mm_set1_epi32(inv_gain_Q26); for( i = 0; i < psEncC->subfr_length - 3; i += 4 ) { - xmm_x_Q3_x2x0 = _mm_loadu_si128( (__m128i *)(&(x_Q3[ i ] ) ) ); + xmm_x16_x2x0 = OP_CVTEPI16_EPI32_M64( &(x16[ i ] ) ); /* equal shift right 4 bytes*/ - xmm_x_Q3_x3x1 = _mm_shuffle_epi32( xmm_x_Q3_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_x16_x3x1 = _mm_shuffle_epi32( xmm_x16_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); - xmm_x_Q3_x2x0 = _mm_mul_epi32( xmm_x_Q3_x2x0, xmm_inv_gain_Q23 ); - xmm_x_Q3_x3x1 = _mm_mul_epi32( xmm_x_Q3_x3x1, xmm_inv_gain_Q23 ); + xmm_x16_x2x0 = _mm_mul_epi32( xmm_x16_x2x0, xmm_inv_gain_Q26 ); + xmm_x16_x3x1 = _mm_mul_epi32( xmm_x16_x3x1, xmm_inv_gain_Q26 ); - xmm_x_Q3_x2x0 = _mm_srli_epi64( xmm_x_Q3_x2x0, 16 ); - xmm_x_Q3_x3x1 = _mm_slli_epi64( xmm_x_Q3_x3x1, 16 ); + xmm_x16_x2x0 = _mm_srli_epi64( xmm_x16_x2x0, 16 ); + xmm_x16_x3x1 = _mm_slli_epi64( xmm_x16_x3x1, 16 ); - xmm_x_Q3_x2x0 = _mm_blend_epi16( xmm_x_Q3_x2x0, xmm_x_Q3_x3x1, 0xCC ); + xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC ); - _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ] ) ), xmm_x_Q3_x2x0 ); + _mm_storeu_si128( (__m128i *)(void*)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 ); } for( ; i < psEncC->subfr_length; i++ ) { - x_sc_Q10[ i ] = silk_SMULWW( x_Q3[ i ], inv_gain_Q23 ); + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); } - /* Save inverse gain */ - NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; - /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ if( NSQ->rewhite_flag ) { if( subfr == 0 ) { @@ -671,16 +718,18 @@ static OPUS_INLINE void silk_nsq_scale_states_sse4_1( } /* Adjust for changing gain */ - if( gain_adj_Q16 != (opus_int32)1 << 16 ) { - /* Scale long-term shaping state */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { __m128i xmm_gain_adj_Q16, xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1; + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ /* prepare gain_adj_Q16 in packed 4 32-bits */ xmm_gain_adj_Q16 = _mm_set1_epi32(gain_adj_Q16); for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 3; i += 4 ) { - xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); + xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(void*)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); /* equal shift right 4 bytes*/ xmm_sLTP_shp_Q14_x3x1 = _mm_shuffle_epi32( xmm_sLTP_shp_Q14_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); @@ -692,7 +741,7 @@ static OPUS_INLINE void silk_nsq_scale_states_sse4_1( xmm_sLTP_shp_Q14_x2x0 = _mm_blend_epi16( xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1, 0xCC ); - _mm_storeu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); + _mm_storeu_si128( (__m128i *)(void*)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); } for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { @@ -707,6 +756,7 @@ static OPUS_INLINE void silk_nsq_scale_states_sse4_1( } NSQ->sLF_AR_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sLF_AR_shp_Q14 ); + NSQ->sDiff_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sDiff_shp_Q14 ); /* Scale short-term prediction and shaping states */ for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { @@ -715,5 +765,8 @@ static OPUS_INLINE void silk_nsq_scale_states_sse4_1( for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { NSQ->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sAR2_Q14[ i ] ); } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; } } diff --git a/code/opus-1.2.1/silk/x86/SigProc_FIX_sse.h b/code/opus-1.5.2/silk/x86/SigProc_FIX_sse.h similarity index 75% rename from code/opus-1.2.1/silk/x86/SigProc_FIX_sse.h rename to code/opus-1.5.2/silk/x86/SigProc_FIX_sse.h index 61efa8da..89a5ec88 100644 --- a/code/opus-1.2.1/silk/x86/SigProc_FIX_sse.h +++ b/code/opus-1.5.2/silk/x86/SigProc_FIX_sse.h @@ -26,13 +26,13 @@ */ #ifndef SIGPROC_FIX_SSE_H -#define SIGPROC_FIX_SSE_H +# define SIGPROC_FIX_SSE_H -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif -#if defined(OPUS_X86_MAY_HAVE_SSE4_1) +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) void silk_burg_modified_sse4_1( opus_int32 *res_nrg, /* O Residual energy */ opus_int *res_nrg_Q, /* O Residual energy Q value */ @@ -45,11 +45,13 @@ void silk_burg_modified_sse4_1( int arch /* I Run-time architecture */ ); -#if defined(OPUS_X86_PRESUME_SSE4_1) -#define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ - ((void)(arch), silk_burg_modified_sse4_1(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) +# if defined(OPUS_X86_PRESUME_SSE4_1) -#else +# define OVERRIDE_silk_burg_modified +# define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ + ((void)(arch), silk_burg_modified_sse4_1(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) + +# elif defined(OPUS_HAVE_RTCD) extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])( opus_int32 *res_nrg, /* O Residual energy */ @@ -62,33 +64,36 @@ extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])( const opus_int D, /* I Order */ int arch /* I Run-time architecture */); -# define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ - ((*SILK_BURG_MODIFIED_IMPL[(arch) & OPUS_ARCHMASK])(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) +# define OVERRIDE_silk_burg_modified +# define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ + ((*SILK_BURG_MODIFIED_IMPL[(arch) & OPUS_ARCHMASK])(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) -#endif +# endif -opus_int64 silk_inner_prod16_aligned_64_sse4_1( +opus_int64 silk_inner_prod16_sse4_1( const opus_int16 *inVec1, const opus_int16 *inVec2, const opus_int len ); -#if defined(OPUS_X86_PRESUME_SSE4_1) +# if defined(OPUS_X86_PRESUME_SSE4_1) -#define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \ - ((void)(arch),silk_inner_prod16_aligned_64_sse4_1(inVec1, inVec2, len)) +# define OVERRIDE_silk_inner_prod16 +# define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((void)(arch),silk_inner_prod16_sse4_1(inVec1, inVec2, len)) -#else +# elif defined(OPUS_HAVE_RTCD) -extern opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[OPUS_ARCHMASK + 1])( +extern opus_int64 (*const SILK_INNER_PROD16_IMPL[OPUS_ARCHMASK + 1])( const opus_int16 *inVec1, const opus_int16 *inVec2, const opus_int len); -# define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \ - ((*SILK_INNER_PROD16_ALIGNED_64_IMPL[(arch) & OPUS_ARCHMASK])(inVec1, inVec2, len)) +# define OVERRIDE_silk_inner_prod16 +# define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((*SILK_INNER_PROD16_IMPL[(arch) & OPUS_ARCHMASK])(inVec1, inVec2, len)) -#endif -#endif +# endif +# endif #endif diff --git a/code/opus-1.2.1/silk/x86/VAD_sse.c b/code/opus-1.5.2/silk/x86/VAD_sse4_1.c similarity index 92% rename from code/opus-1.2.1/silk/x86/VAD_sse.c rename to code/opus-1.5.2/silk/x86/VAD_sse4_1.c index 4e90f441..9e06bc79 100644 --- a/code/opus-1.2.1/silk/x86/VAD_sse.c +++ b/code/opus-1.5.2/silk/x86/VAD_sse4_1.c @@ -1,5 +1,5 @@ -/* Copyright (c) 2014, Cisco Systems, INC - Written by XiangMingZhu WeiZhou MinPeng YanWang +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -63,11 +63,19 @@ opus_int silk_VAD_GetSA_Q8_sse4_1( /* O Return value, 0 if s SAVE_STACK; +#ifdef OPUS_CHECK_ASM + silk_encoder_state psEncC_c; + opus_int ret_c; + + silk_memcpy( &psEncC_c, psEncC, sizeof( psEncC_c ) ); + ret_c = silk_VAD_GetSA_Q8_c( &psEncC_c, pIn ); +#endif + /* Safety checks */ silk_assert( VAD_N_BANDS == 4 ); - silk_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); - silk_assert( psEncC->frame_length <= 512 ); - silk_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); + celt_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); + celt_assert( psEncC->frame_length <= 512 ); + celt_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); /***********************/ /* Filter and Decimate */ @@ -136,7 +144,7 @@ opus_int silk_VAD_GetSA_Q8_sse4_1( /* O Return value, 0 if s for( i = 0; i < dec_subframe_length - 7; i += 8 ) { - xmm_X = _mm_loadu_si128( (__m128i *)&(X[ X_offset[ b ] + i + dec_subframe_offset ] ) ); + xmm_X = _mm_loadu_si128( (__m128i *)(void*)&(X[ X_offset[ b ] + i + dec_subframe_offset ] ) ); xmm_X = _mm_srai_epi16( xmm_X, 3 ); xmm_X = _mm_madd_epi16( xmm_X, xmm_X ); xmm_acc = _mm_add_epi32( xmm_acc, xmm_X ); @@ -233,15 +241,14 @@ opus_int silk_VAD_GetSA_Q8_sse4_1( /* O Return value, 0 if s speech_nrg += ( b + 1 ) * silk_RSHIFT( Xnrg[ b ] - psSilk_VAD->NL[ b ], 4 ); } + if( psEncC->frame_length == 20 * psEncC->fs_kHz ) { + speech_nrg = silk_RSHIFT32( speech_nrg, 1 ); + } /* Power scaling */ if( speech_nrg <= 0 ) { SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); - } else if( speech_nrg < 32768 ) { - if( psEncC->frame_length == 10 * psEncC->fs_kHz ) { - speech_nrg = silk_LSHIFT_SAT32( speech_nrg, 16 ); - } else { - speech_nrg = silk_LSHIFT_SAT32( speech_nrg, 15 ); - } + } else if( speech_nrg < 16384 ) { + speech_nrg = silk_LSHIFT32( speech_nrg, 16 ); /* square-root */ speech_nrg = silk_SQRT_APPROX( speech_nrg ); @@ -272,6 +279,11 @@ opus_int silk_VAD_GetSA_Q8_sse4_1( /* O Return value, 0 if s psEncC->input_quality_bands_Q15[ b ] = silk_sigm_Q15( silk_RSHIFT( SNR_Q7 - 16 * 128, 4 ) ); } +#ifdef OPUS_CHECK_ASM + silk_assert( ret == ret_c ); + silk_assert( !memcmp( &psEncC_c, psEncC, sizeof( psEncC_c ) ) ); +#endif + RESTORE_STACK; return( ret ); } diff --git a/code/opus-1.5.2/silk/x86/VQ_WMat_EC_sse4_1.c b/code/opus-1.5.2/silk/x86/VQ_WMat_EC_sse4_1.c new file mode 100644 index 00000000..df4626b6 --- /dev/null +++ b/code/opus-1.5.2/silk/x86/VQ_WMat_EC_sse4_1.c @@ -0,0 +1,173 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include "main.h" +#include "celt/x86/x86cpu.h" + +/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */ +void silk_VQ_WMat_EC_sse4_1( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +) +{ + opus_int k, gain_tmp_Q7; + const opus_int8 *cb_row_Q7; + opus_int32 neg_xX_Q24[ 5 ]; + opus_int32 sum1_Q15, sum2_Q24; + opus_int32 bits_res_Q8, bits_tot_Q8; + __m128i v_XX_31_Q17, v_XX_42_Q17, v_cb_row_31_Q7, v_cb_row_42_Q7, v_acc1_Q24, v_acc2_Q24; + + /* Negate and convert to new Q domain */ + neg_xX_Q24[ 0 ] = -silk_LSHIFT32( xX_Q17[ 0 ], 7 ); + neg_xX_Q24[ 1 ] = -silk_LSHIFT32( xX_Q17[ 1 ], 7 ); + neg_xX_Q24[ 2 ] = -silk_LSHIFT32( xX_Q17[ 2 ], 7 ); + neg_xX_Q24[ 3 ] = -silk_LSHIFT32( xX_Q17[ 3 ], 7 ); + neg_xX_Q24[ 4 ] = -silk_LSHIFT32( xX_Q17[ 4 ], 7 ); + + v_XX_31_Q17 = _mm_loadu_si128( (__m128i *)(void*)(&XX_Q17[ 1 ] ) ); + v_XX_42_Q17 = _mm_shuffle_epi32( v_XX_31_Q17, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + /* Loop over codebook */ + *rate_dist_Q8 = silk_int32_MAX; + *res_nrg_Q15 = silk_int32_MAX; + cb_row_Q7 = cb_Q7; + /* If things go really bad, at least *ind is set to something safe. */ + *ind = 0; + for( k = 0; k < L; k++ ) { + opus_int32 penalty; + gain_tmp_Q7 = cb_gain_Q7[k]; + /* Weighted rate */ + /* Quantization error: 1 - 2 * xX * cb + cb' * XX * cb */ + sum1_Q15 = SILK_FIX_CONST( 1.001, 15 ); + + /* Penalty for too large gain */ + penalty = silk_LSHIFT32( silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 11 ); + + /* first row of XX_Q17 */ + v_cb_row_31_Q7 = OP_CVTEPI8_EPI32_M32( &cb_row_Q7[ 1 ] ); + v_cb_row_42_Q7 = _mm_shuffle_epi32( v_cb_row_31_Q7, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + v_cb_row_31_Q7 = _mm_mul_epi32( v_XX_31_Q17, v_cb_row_31_Q7 ); + v_cb_row_42_Q7 = _mm_mul_epi32( v_XX_42_Q17, v_cb_row_42_Q7 ); + v_acc1_Q24 = _mm_add_epi64( v_cb_row_31_Q7, v_cb_row_42_Q7); + v_acc2_Q24 = _mm_shuffle_epi32( v_acc1_Q24, _MM_SHUFFLE( 1, 0, 3, 2 ) ); + v_acc1_Q24 = _mm_add_epi64( v_acc1_Q24, v_acc2_Q24); + sum2_Q24 = _mm_cvtsi128_si32( v_acc1_Q24 ); + sum2_Q24 = silk_ADD32( neg_xX_Q24[ 0 ], sum2_Q24 ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 0 ], cb_row_Q7[ 0 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 0 ] ); + + /* second row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 1 ], XX_Q17[ 7 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 8 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 9 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 6 ], cb_row_Q7[ 1 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 1 ] ); + + /* third row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 2 ], XX_Q17[ 13 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 14 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 12 ], cb_row_Q7[ 2 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 2 ] ); + + /* fourth row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 3 ], XX_Q17[ 19 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 18 ], cb_row_Q7[ 3 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 3 ] ); + + /* last row of XX_Q17 */ + sum2_Q24 = silk_LSHIFT32( neg_xX_Q24[ 4 ], 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 24 ], cb_row_Q7[ 4 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 4 ] ); + + /* find best */ + if( sum1_Q15 >= 0 ) { + /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */ + bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) ); + /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */ + bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 ); + if( bits_tot_Q8 <= *rate_dist_Q8 ) { + *rate_dist_Q8 = bits_tot_Q8; + *res_nrg_Q15 = sum1_Q15 + penalty; + *ind = (opus_int8)k; + *gain_Q7 = gain_tmp_Q7; + } + } + + /* Go to next cbk vector */ + cb_row_Q7 += LTP_ORDER; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int8 ind_c = 0; + opus_int32 res_nrg_Q15_c = 0; + opus_int32 rate_dist_Q8_c = 0; + opus_int gain_Q7_c = 0; + + silk_VQ_WMat_EC_c( + &ind_c, + &res_nrg_Q15_c, + &rate_dist_Q8_c, + &gain_Q7_c, + XX_Q17, + xX_Q17, + cb_Q7, + cb_gain_Q7, + cl_Q5, + subfr_len, + max_gain_Q7, + L + ); + + silk_assert( *ind == ind_c ); + silk_assert( *res_nrg_Q15 == res_nrg_Q15_c ); + silk_assert( *rate_dist_Q8 == rate_dist_Q8_c ); + silk_assert( *gain_Q7 == gain_Q7_c ); + } +#endif +} diff --git a/code/opus-1.5.2/silk/x86/main_sse.h b/code/opus-1.5.2/silk/x86/main_sse.h new file mode 100644 index 00000000..b254d53e --- /dev/null +++ b/code/opus-1.5.2/silk/x86/main_sse.h @@ -0,0 +1,299 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MAIN_SSE_H +# define MAIN_SSE_H + +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) + +void silk_VQ_WMat_EC_sse4_1( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +); + +# if defined OPUS_X86_PRESUME_SSE4_1 + +# define OVERRIDE_silk_VQ_WMat_EC +# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L, arch) \ + ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +); + +# define OVERRIDE_silk_VQ_WMat_EC +# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L, arch) \ + ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L)) + +# endif + +void silk_NSQ_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# if defined OPUS_X86_PRESUME_SSE4_1 + +# define OVERRIDE_silk_NSQ +# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# define OVERRIDE_silk_NSQ +# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# endif + +void silk_NSQ_del_dec_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +void silk_NSQ_del_dec_avx2( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[MAX_NB_SUBFR], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[MAX_NB_SUBFR], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[MAX_NB_SUBFR], /* I Quantization step sizes */ + const opus_int32 pitchL[MAX_NB_SUBFR], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# if defined (OPUS_X86_PRESUME_AVX2) + +# define OVERRIDE_silk_NSQ_del_dec +# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_del_dec_avx2(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# elif defined (OPUS_X86_PRESUME_SSE4_1) && !defined(OPUS_X86_MAY_HAVE_AVX2) + +# define OVERRIDE_silk_NSQ_del_dec +# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# define OVERRIDE_silk_NSQ_del_dec +# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# endif + +void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +); + +/**************************/ +/* Noise level estimation */ +/**************************/ +void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +); + +opus_int silk_VAD_GetSA_Q8_sse4_1( + silk_encoder_state *psEnC, + const opus_int16 pIn[] +); + +# if defined(OPUS_X86_PRESUME_SSE4_1) + +# define OVERRIDE_silk_VAD_GetSA_Q8 +# define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) + +# elif defined(OPUS_HAVE_RTCD) + +extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( + silk_encoder_state *psEnC, + const opus_int16 pIn[]); + +# define OVERRIDE_silk_VAD_GetSA_Q8 +# define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ + ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) + +# endif + +#ifndef FIXED_POINT +double silk_inner_product_FLP_avx2( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +); + +#if defined (OPUS_X86_PRESUME_AVX2) + +#define OVERRIDE_inner_product_FLP +#define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,silk_inner_product_FLP_avx2(data1, data2, dataSize)) + +#elif defined(OPUS_HAVE_RTCD) && defined(OPUS_X86_MAY_HAVE_AVX2) + +#define OVERRIDE_inner_product_FLP +extern double (*const SILK_INNER_PRODUCT_FLP_IMPL[OPUS_ARCHMASK + 1])( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +); + +#define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,(*SILK_INNER_PRODUCT_FLP_IMPL[(arch) & OPUS_ARCHMASK])(data1, data2, dataSize)) + +#endif +#endif + +# endif +#endif diff --git a/code/opus-1.2.1/silk/x86/x86_silk_map.c b/code/opus-1.5.2/silk/x86/x86_silk_map.c similarity index 64% rename from code/opus-1.2.1/silk/x86/x86_silk_map.c rename to code/opus-1.5.2/silk/x86/x86_silk_map.c index 32dcc3ca..39ad7527 100644 --- a/code/opus-1.2.1/silk/x86/x86_silk_map.c +++ b/code/opus-1.5.2/silk/x86/x86_silk_map.c @@ -32,25 +32,28 @@ #include "celt/x86/x86cpu.h" #include "structs.h" #include "SigProc_FIX.h" +#ifndef FIXED_POINT +#include "SigProc_FLP.h" +#endif #include "pitch.h" #include "main.h" -#if !defined(OPUS_X86_PRESUME_SSE4_1) +#if defined(OPUS_HAVE_RTCD) && !defined(OPUS_X86_PRESUME_AVX2) #if defined(FIXED_POINT) #include "fixed/main_FIX.h" -opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[ OPUS_ARCHMASK + 1 ] )( +opus_int64 (*const SILK_INNER_PROD16_IMPL[ OPUS_ARCHMASK + 1 ] )( const opus_int16 *inVec1, const opus_int16 *inVec2, const opus_int len ) = { - silk_inner_prod16_aligned_64_c, /* non-sse */ - silk_inner_prod16_aligned_64_c, - silk_inner_prod16_aligned_64_c, - MAY_HAVE_SSE4_1( silk_inner_prod16_aligned_64 ), /* sse4.1 */ - MAY_HAVE_SSE4_1( silk_inner_prod16_aligned_64 ) /* avx */ + silk_inner_prod16_c, /* non-sse */ + silk_inner_prod16_c, + silk_inner_prod16_c, + MAY_HAVE_SSE4_1( silk_inner_prod16 ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_inner_prod16 ) /* avx */ }; #endif @@ -66,23 +69,22 @@ opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ) /* avx */ }; -#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) = { silk_NSQ_c, /* non-sse */ silk_NSQ_c, @@ -90,21 +92,20 @@ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_NSQ ), /* sse4.1 */ MAY_HAVE_SSE4_1( silk_NSQ ) /* avx */ }; -#endif -#if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( opus_int8 *ind, /* O index of best codebook vector */ - opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ - const opus_int16 *in_Q14, /* I input vector to be quantized */ - const opus_int32 *W_Q18, /* I weighting matrix */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ const opus_int8 *cb_Q7, /* I codebook */ const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ - const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ + const opus_int subfr_len, /* I number of samples per subframe */ const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ - opus_int L /* I number of vectors in codebook */ + const opus_int L /* I number of vectors in codebook */ ) = { silk_VQ_WMat_EC_c, /* non-sse */ silk_VQ_WMat_EC_c, @@ -112,33 +113,30 @@ void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ), /* sse4.1 */ MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ) /* avx */ }; -#endif -#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )( - const silk_encoder_state *psEncC, /* I Encoder State */ - silk_nsq_state *NSQ, /* I/O NSQ state */ - SideInfoIndices *psIndices, /* I/O Quantization Indices */ - const opus_int32 x_Q3[], /* I Prefiltered input signal */ - opus_int8 pulses[], /* O Quantized pulse signal */ - const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ - const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ - const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ - const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ - const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ - const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ - const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ - const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ - const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ - const opus_int LTP_scale_Q14 /* I LTP state scaling */ + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ ) = { silk_NSQ_del_dec_c, /* non-sse */ silk_NSQ_del_dec_c, silk_NSQ_del_dec_c, MAY_HAVE_SSE4_1( silk_NSQ_del_dec ), /* sse4.1 */ - MAY_HAVE_SSE4_1( silk_NSQ_del_dec ) /* avx */ + MAY_HAVE_AVX2( silk_NSQ_del_dec ) /* avx */ }; -#endif #if defined(FIXED_POINT) @@ -161,4 +159,21 @@ void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )( }; #endif + +#ifndef FIXED_POINT + +double (*const SILK_INNER_PRODUCT_FLP_IMPL[ OPUS_ARCHMASK + 1 ] )( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +) = { + silk_inner_product_FLP_c, /* non-sse */ + silk_inner_product_FLP_c, + silk_inner_product_FLP_c, + silk_inner_product_FLP_c, /* sse4.1 */ + MAY_HAVE_AVX2( silk_inner_product_FLP ) /* avx */ +}; + +#endif + #endif diff --git a/code/opus-1.2.1/src/analysis.c b/code/opus-1.5.2/src/analysis.c similarity index 77% rename from code/opus-1.2.1/src/analysis.c rename to code/opus-1.5.2/src/analysis.c index f4160e4b..1f580138 100644 --- a/code/opus-1.2.1/src/analysis.c +++ b/code/opus-1.5.2/src/analysis.c @@ -31,7 +31,9 @@ #define ANALYSIS_C +#ifdef MLP_TRAINING #include +#endif #include "mathops.h" #include "kiss_fft.h" @@ -50,6 +52,8 @@ #ifndef DISABLE_FLOAT_API +#define TRANSITION_PENALTY 10 + static const float dct_table[128] = { 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, @@ -224,25 +228,38 @@ void tonality_analysis_reset(TonalityAnalysisState *tonal) /* Clear non-reusable fields. */ char *start = (char*)&tonal->TONALITY_ANALYSIS_RESET_START; OPUS_CLEAR(start, sizeof(TonalityAnalysisState) - (start - (char*)tonal)); - tonal->music_confidence = .9f; - tonal->speech_confidence = .1f; } void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len) { int pos; int curr_lookahead; - float psum; float tonality_max; float tonality_avg; int tonality_count; int i; + int pos0; + float prob_avg; + float prob_count; + float prob_min, prob_max; + float vad_prob; + int mpos, vpos; + int bandwidth_span; pos = tonal->read_pos; curr_lookahead = tonal->write_pos-tonal->read_pos; if (curr_lookahead<0) curr_lookahead += DETECT_SIZE; + tonal->read_subframe += len/(tonal->Fs/400); + while (tonal->read_subframe>=8) + { + tonal->read_subframe -= 8; + tonal->read_pos++; + } + if (tonal->read_pos>=DETECT_SIZE) + tonal->read_pos-=DETECT_SIZE; + /* On long frames, look at the second analysis window rather than the first. */ if (len > tonal->Fs/50 && pos != tonal->write_pos) { @@ -254,9 +271,14 @@ void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int pos--; if (pos<0) pos = DETECT_SIZE-1; + pos0 = pos; OPUS_COPY(info_out, &tonal->info[pos], 1); + if (!info_out->valid) + return; tonality_max = tonality_avg = info_out->tonality; tonality_count = 1; + /* Look at the neighbouring frames and pick largest bandwidth found (to be safe). */ + bandwidth_span = 6; /* If possible, look ahead for a tone to compensate for the delay in the tone detector. */ for (i=0;i<3;i++) { @@ -268,31 +290,122 @@ void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int tonality_max = MAX32(tonality_max, tonal->info[pos].tonality); tonality_avg += tonal->info[pos].tonality; tonality_count++; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); + bandwidth_span--; + } + pos = pos0; + /* Look back in time to see if any has a wider bandwidth than the current frame. */ + for (i=0;iwrite_pos) + break; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); } info_out->tonality = MAX32(tonality_avg/tonality_count, tonality_max-.2f); - tonal->read_subframe += len/(tonal->Fs/400); - while (tonal->read_subframe>=8) + + mpos = vpos = pos0; + /* If we have enough look-ahead, compensate for the ~5-frame delay in the music prob and + ~1 frame delay in the VAD prob. */ + if (curr_lookahead > 15) { - tonal->read_subframe -= 8; - tonal->read_pos++; + mpos += 5; + if (mpos>=DETECT_SIZE) + mpos -= DETECT_SIZE; + vpos += 1; + if (vpos>=DETECT_SIZE) + vpos -= DETECT_SIZE; } - if (tonal->read_pos>=DETECT_SIZE) - tonal->read_pos-=DETECT_SIZE; - /* The -1 is to compensate for the delay in the features themselves. */ - curr_lookahead = IMAX(curr_lookahead-1, 0); + /* The following calculations attempt to minimize a "badness function" + for the transition. When switching from speech to music, the badness + of switching at frame k is + b_k = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + where + v_i is the activity probability (VAD) at frame i, + p_i is the music probability at frame i + T is the probability threshold for switching + S is the penalty for switching during active audio rather than silence + the current frame has index i=0 - psum=0; - /* Summing the probability of transition patterns that involve music at - time (DETECT_SIZE-curr_lookahead-1) */ - for (i=0;ipmusic[i]; - for (;ipspeech[i]; - psum = psum*tonal->music_confidence + (1-psum)*tonal->speech_confidence; - /*printf("%f %f %f %f %f\n", psum, info_out->music_prob, info_out->vad_prob, info_out->activity_probability, info_out->tonality);*/ + Rather than apply badness to directly decide when to switch, what we compute + instead is the threshold for which the optimal switching point is now. When + considering whether to switch now (frame 0) or at frame k, we have: + S*v_0 = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + which gives us: + T = ( \sum_{i=0}^{k-1} v_i*p_i + S*(v_k-v_0) ) / ( \sum_{i=0}^{k-1} v_i ) + We take the min threshold across all positive values of k (up to the maximum + amount of lookahead we have) to give us the threshold for which the current + frame is the optimal switch point. - info_out->music_prob = psum; + The last step is that we need to consider whether we want to switch at all. + For that we use the average of the music probability over the entire window. + If the threshold is higher than that average we're not going to + switch, so we compute a min with the average as well. The result of all these + min operations is music_prob_min, which gives the threshold for switching to music + if we're currently encoding for speech. + + We do the exact opposite to compute music_prob_max which is used for switching + from music to speech. + */ + prob_min = 1.f; + prob_max = 0.f; + vad_prob = tonal->info[vpos].activity_probability; + prob_count = MAX16(.1f, vad_prob); + prob_avg = MAX16(.1f, vad_prob)*tonal->info[mpos].music_prob; + while (1) + { + float pos_vad; + mpos++; + if (mpos==DETECT_SIZE) + mpos = 0; + if (mpos == tonal->write_pos) + break; + vpos++; + if (vpos==DETECT_SIZE) + vpos = 0; + if (vpos == tonal->write_pos) + break; + pos_vad = tonal->info[vpos].activity_probability; + prob_min = MIN16((prob_avg - TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_min); + prob_max = MAX16((prob_avg + TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_max); + prob_count += MAX16(.1f, pos_vad); + prob_avg += MAX16(.1f, pos_vad)*tonal->info[mpos].music_prob; + } + info_out->music_prob = prob_avg/prob_count; + prob_min = MIN16(prob_avg/prob_count, prob_min); + prob_max = MAX16(prob_avg/prob_count, prob_max); + prob_min = MAX16(prob_min, 0.f); + prob_max = MIN16(prob_max, 1.f); + + /* If we don't have enough look-ahead, do our best to make a decent decision. */ + if (curr_lookahead < 10) + { + float pmin, pmax; + pmin = prob_min; + pmax = prob_max; + pos = pos0; + /* Look for min/max in the past. */ + for (i=0;icount-1, 15);i++) + { + pos--; + if (pos < 0) + pos = DETECT_SIZE-1; + pmin = MIN16(pmin, tonal->info[pos].music_prob); + pmax = MAX16(pmax, tonal->info[pos].music_prob); + } + /* Bias against switching on active audio. */ + pmin = MAX16(0.f, pmin - .1f*vad_prob); + pmax = MIN16(1.f, pmax + .1f*vad_prob); + prob_min += (1.f-.1f*curr_lookahead)*(pmin - prob_min); + prob_max += (1.f-.1f*curr_lookahead)*(pmax - prob_max); + } + info_out->music_prob_min = prob_min; + info_out->music_prob_max = prob_max; + + /* printf("%f %f %f %f %f\n", prob_min, prob_max, prob_avg/prob_count, vad_prob, info_out->music_prob); */ } static const float std_feature_bias[9] = { @@ -312,6 +425,24 @@ static const float std_feature_bias[9] = { #define SCALE_ENER(e) (e) #endif +#ifdef FIXED_POINT +static int is_digital_silence32(const opus_val32* pcm, int frame_size, int channels, int lsb_depth) +{ + int silence = 0; + opus_val32 sample_max = 0; +#ifdef MLP_TRAINING + return 0; +#endif + sample_max = celt_maxabs32(pcm, frame_size*channels); + + silence = (sample_max == 0); + (void)lsb_depth; + return silence; +} +#else +#define is_digital_silence32(pcm, frame_size, channels, lsb_depth) is_digital_silence(pcm, frame_size, channels, lsb_depth) +#endif + static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix) { int i, b; @@ -340,6 +471,7 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt float alpha, alphaE, alphaE2; float frame_loudness; float bandwidth_mask; + int is_masked[NB_TBANDS+1]; int bandwidth=0; float maxE = 0; float noise_floor; @@ -352,11 +484,22 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt float band_log2[NB_TBANDS+1]; float leakage_from[NB_TBANDS+1]; float leakage_to[NB_TBANDS+1]; + float layer_out[MAX_NEURONS]; + float below_max_pitch; + float above_max_pitch; + int is_silence; SAVE_STACK; + if (!tonal->initialized) + { + tonal->mem_fill = 240; + tonal->initialized = 1; + } alpha = 1.f/IMIN(10, 1+tonal->count); alphaE = 1.f/IMIN(25, 1+tonal->count); - alphaE2 = 1.f/IMIN(500, 1+tonal->count); + /* Noise floor related decay for bandwidth detection: -2.2 dB/second */ + alphaE2 = 1.f/IMIN(100, 1+tonal->count); + if (tonal->count <= 1) alphaE2 = 1; if (tonal->Fs == 48000) { @@ -368,15 +511,7 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt offset = 3*offset/2; } - if (tonal->count<4) { - if (tonal->application == OPUS_APPLICATION_VOIP) - tonal->music_prob = .1f; - else - tonal->music_prob = .625f; - } kfft = celt_mode->mdct.kfft[0]; - if (tonal->count==0) - tonal->mem_fill = 240; tonal->hp_ener_accum += (float)downmix_and_resample(downmix, x, &tonal->inmem[tonal->mem_fill], tonal->downmix_state, IMIN(len, ANALYSIS_BUF_SIZE-tonal->mem_fill), offset, c1, c2, C, tonal->Fs); @@ -392,6 +527,8 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt if (tonal->write_pos>=DETECT_SIZE) tonal->write_pos-=DETECT_SIZE; + is_silence = is_digital_silence32(tonal->inmem, ANALYSIS_BUF_SIZE, 1, lsb_depth); + ALLOC(in, 480, kiss_fft_cpx); ALLOC(out, 480, kiss_fft_cpx); ALLOC(tonality, 240, float); @@ -410,6 +547,16 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt &tonal->inmem[240], tonal->downmix_state, remaining, offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C, tonal->Fs); tonal->mem_fill = 240 + remaining; + if (is_silence) + { + /* On silence, copy the previous analysis. */ + int prev_pos = tonal->write_pos-2; + if (prev_pos < 0) + prev_pos += DETECT_SIZE; + OPUS_COPY(info, &tonal->info[prev_pos], 1); + RESTORE_STACK; + return; + } opus_fft(kfft, in, out, tonal->arch); #ifndef FIXED_POINT /* If there's any NaN on the input, the entire output will be NaN, so we only need to check one value. */ @@ -546,7 +693,7 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt tonal->lowE[b] = logE[b]; tonal->highE[b] = MIN32(tonal->lowE[b]+15, tonal->highE[b]); } - relativeE += (logE[b]-tonal->lowE[b])/(1e-15f + (tonal->highE[b]-tonal->lowE[b])); + relativeE += (logE[b]-tonal->lowE[b])/(1e-5f + (tonal->highE[b]-tonal->lowE[b])); L1=L2=0; for (i=0;imeanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); - E = MAX32(E, tonal->meanE[b]); - /* Use a simple follower with 13 dB/Bark slope for spreading function */ - bandwidth_mask = MAX32(.05f*bandwidth_mask, E); + Em = MAX32(E, tonal->meanE[b]); /* Consider the band "active" only if all these conditions are met: - 1) less than 10 dB below the simple follower - 2) less than 90 dB below the peak band (maximal masking possible considering + 1) less than 90 dB below the peak band (maximal masking possible considering both the ATH and the loudness-dependent slope of the spreading function) - 3) above the PCM quantization noise floor + 2) above the PCM quantization noise floor We use b+1 because the first CELT band isn't included in tbands[] */ - if (E>.1*bandwidth_mask && E*1e9f > maxE && E > noise_floor*(band_end-band_start)) + if (E*1e9f > maxE && (Em > 3*noise_floor*(band_end-band_start) || E > noise_floor*(band_end-band_start))) bandwidth = b+1; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth >= b+1 ? .01f : .05f)*bandwidth_mask; + /* Use a simple follower with 13 dB/Bark slope for spreading function. */ + bandwidth_mask = MAX32(.05f*bandwidth_mask, E); } /* Special case for the last two bands, for which we don't have spectrum but only - the energy above 12 kHz. */ + the energy above 12 kHz. The difficulty here is that the high-pass we use + leaks some LF energy, so we need to increase the threshold without accidentally cutting + off the band. */ if (tonal->Fs == 48000) { - float ratio; - float E = hp_ener*(1.f/(240*240)); - ratio = tonal->prev_bandwidth==20 ? 0.03f : 0.07f; + float noise_ratio; + float Em; + float E = hp_ener*(1.f/(60*60)); + noise_ratio = tonal->prev_bandwidth==20 ? 10.f : 30.f; + #ifdef FIXED_POINT /* silk_resampler_down2_hp() shifted right by an extra 8 bits. */ E *= 256.f*(1.f/Q15ONE)*(1.f/Q15ONE); #endif - maxE = MAX32(maxE, E); + above_max_pitch += E; tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); - E = MAX32(E, tonal->meanE[b]); - /* Use a simple follower with 13 dB/Bark slope for spreading function */ - bandwidth_mask = MAX32(.05f*bandwidth_mask, E); - if (E>ratio*bandwidth_mask && E*1e9f > maxE && E > noise_floor*160) - bandwidth = 20; - /* This detector is unreliable, so if the bandwidth is close to SWB, assume it's FB. */ - if (bandwidth >= 17) + Em = MAX32(E, tonal->meanE[b]); + if (Em > 3*noise_ratio*noise_floor*160 || E > noise_ratio*noise_floor*160) bandwidth = 20; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth == 20 ? .01f : .05f)*bandwidth_mask; } + if (above_max_pitch > below_max_pitch) + info->max_pitch_ratio = below_max_pitch/above_max_pitch; + else + info->max_pitch_ratio = 1; + /* In some cases, resampling aliasing can create a small amount of energy in the first band + being cut. So if the last band is masked, we don't include it. */ + if (bandwidth == 20 && is_masked[NB_TBANDS]) + bandwidth-=2; + else if (bandwidth > 0 && bandwidth <= NB_TBANDS && is_masked[bandwidth-1]) + bandwidth--; if (tonal->count<=2) bandwidth = 20; frame_loudness = 20*(float)log10(frame_loudness); @@ -761,139 +929,15 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt features[23] = info->tonality_slope + 0.069216f; features[24] = tonal->lowECount - 0.067930f; - mlp_process(&net, features, frame_probs); - frame_probs[0] = .5f*(frame_probs[0]+1); - /* Curve fitting between the MLP probability and the actual probability */ - /*frame_probs[0] = .01f + 1.21f*frame_probs[0]*frame_probs[0] - .23f*(float)pow(frame_probs[0], 10);*/ - /* Probability of active audio (as opposed to silence) */ - frame_probs[1] = .5f*frame_probs[1]+.5f; - frame_probs[1] *= frame_probs[1]; + analysis_compute_dense(&layer0, layer_out, features); + analysis_compute_gru(&layer1, tonal->rnn_state, layer_out); + analysis_compute_dense(&layer2, frame_probs, tonal->rnn_state); /* Probability of speech or music vs noise */ info->activity_probability = frame_probs[1]; + info->music_prob = frame_probs[0]; - /*printf("%f %f\n", frame_probs[0], frame_probs[1]);*/ - { - /* Probability of state transition */ - float tau; - /* Represents independence of the MLP probabilities, where - beta=1 means fully independent. */ - float beta; - /* Denormalized probability of speech (p0) and music (p1) after update */ - float p0, p1; - /* Probabilities for "all speech" and "all music" */ - float s0, m0; - /* Probability sum for renormalisation */ - float psum; - /* Instantaneous probability of speech and music, with beta pre-applied. */ - float speech0; - float music0; - float p, q; - - /* More silence transitions for speech than for music. */ - tau = .001f*tonal->music_prob + .01f*(1-tonal->music_prob); - p = MAX16(.05f,MIN16(.95f,frame_probs[1])); - q = MAX16(.05f,MIN16(.95f,tonal->vad_prob)); - beta = .02f+.05f*ABS16(p-q)/(p*(1-q)+q*(1-p)); - /* p0 and p1 are the probabilities of speech and music at this frame - using only information from previous frame and applying the - state transition model */ - p0 = (1-tonal->vad_prob)*(1-tau) + tonal->vad_prob *tau; - p1 = tonal->vad_prob *(1-tau) + (1-tonal->vad_prob)*tau; - /* We apply the current probability with exponent beta to work around - the fact that the probability estimates aren't independent. */ - p0 *= (float)pow(1-frame_probs[1], beta); - p1 *= (float)pow(frame_probs[1], beta); - /* Normalise the probabilities to get the Marokv probability of music. */ - tonal->vad_prob = p1/(p0+p1); - info->vad_prob = tonal->vad_prob; - /* Consider that silence has a 50-50 probability of being speech or music. */ - frame_probs[0] = tonal->vad_prob*frame_probs[0] + (1-tonal->vad_prob)*.5f; - - /* One transition every 3 minutes of active audio */ - tau = .0001f; - /* Adapt beta based on how "unexpected" the new prob is */ - p = MAX16(.05f,MIN16(.95f,frame_probs[0])); - q = MAX16(.05f,MIN16(.95f,tonal->music_prob)); - beta = .02f+.05f*ABS16(p-q)/(p*(1-q)+q*(1-p)); - /* p0 and p1 are the probabilities of speech and music at this frame - using only information from previous frame and applying the - state transition model */ - p0 = (1-tonal->music_prob)*(1-tau) + tonal->music_prob *tau; - p1 = tonal->music_prob *(1-tau) + (1-tonal->music_prob)*tau; - /* We apply the current probability with exponent beta to work around - the fact that the probability estimates aren't independent. */ - p0 *= (float)pow(1-frame_probs[0], beta); - p1 *= (float)pow(frame_probs[0], beta); - /* Normalise the probabilities to get the Marokv probability of music. */ - tonal->music_prob = p1/(p0+p1); - info->music_prob = tonal->music_prob; - - /*printf("%f %f %f %f\n", frame_probs[0], frame_probs[1], tonal->music_prob, tonal->vad_prob);*/ - /* This chunk of code deals with delayed decision. */ - psum=1e-20f; - /* Instantaneous probability of speech and music, with beta pre-applied. */ - speech0 = (float)pow(1-frame_probs[0], beta); - music0 = (float)pow(frame_probs[0], beta); - if (tonal->count==1) - { - if (tonal->application == OPUS_APPLICATION_VOIP) - tonal->pmusic[0] = .1f; - else - tonal->pmusic[0] = .625f; - tonal->pspeech[0] = 1-tonal->pmusic[0]; - } - /* Updated probability of having only speech (s0) or only music (m0), - before considering the new observation. */ - s0 = tonal->pspeech[0] + tonal->pspeech[1]; - m0 = tonal->pmusic [0] + tonal->pmusic [1]; - /* Updates s0 and m0 with instantaneous probability. */ - tonal->pspeech[0] = s0*(1-tau)*speech0; - tonal->pmusic [0] = m0*(1-tau)*music0; - /* Propagate the transition probabilities */ - for (i=1;ipspeech[i] = tonal->pspeech[i+1]*speech0; - tonal->pmusic [i] = tonal->pmusic [i+1]*music0; - } - /* Probability that the latest frame is speech, when all the previous ones were music. */ - tonal->pspeech[DETECT_SIZE-1] = m0*tau*speech0; - /* Probability that the latest frame is music, when all the previous ones were speech. */ - tonal->pmusic [DETECT_SIZE-1] = s0*tau*music0; - - /* Renormalise probabilities to 1 */ - for (i=0;ipspeech[i] + tonal->pmusic[i]; - psum = 1.f/psum; - for (i=0;ipspeech[i] *= psum; - tonal->pmusic [i] *= psum; - } - psum = tonal->pmusic[0]; - for (i=1;ipspeech[i]; - - /* Estimate our confidence in the speech/music decisions */ - if (frame_probs[1]>.75) - { - if (tonal->music_prob>.9) - { - float adapt; - adapt = 1.f/(++tonal->music_confidence_count); - tonal->music_confidence_count = IMIN(tonal->music_confidence_count, 500); - tonal->music_confidence += adapt*MAX16(-.2f,frame_probs[0]-tonal->music_confidence); - } - if (tonal->music_prob<.1) - { - float adapt; - adapt = 1.f/(++tonal->speech_confidence_count); - tonal->speech_confidence_count = IMIN(tonal->speech_confidence_count, 500); - tonal->speech_confidence += adapt*MIN16(.2f,frame_probs[0]-tonal->speech_confidence); - } - } - } - tonal->last_music = tonal->music_prob>.5f; + /*printf("%f %f %f\n", frame_probs[0], frame_probs[1], info->music_prob);*/ #ifdef MLP_TRAINING for (i=0;i<25;i++) printf("%f ", features[i]); @@ -933,7 +977,6 @@ void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, co analysis->analysis_offset -= frame_size; } - analysis_info->valid = 0; tonality_get_info(analysis, analysis_info, frame_size); } diff --git a/code/opus-1.2.1/src/analysis.h b/code/opus-1.5.2/src/analysis.h similarity index 86% rename from code/opus-1.2.1/src/analysis.h rename to code/opus-1.5.2/src/analysis.h index cac51dfa..0b66555f 100644 --- a/code/opus-1.2.1/src/analysis.h +++ b/code/opus-1.5.2/src/analysis.h @@ -30,6 +30,7 @@ #include "celt.h" #include "opus_private.h" +#include "mlp.h" #define NB_FRAMES 8 #define NB_TBANDS 18 @@ -64,28 +65,17 @@ typedef struct { float mem[32]; float cmean[8]; float std[9]; - float music_prob; - float vad_prob; float Etracker; float lowECount; int E_count; - int last_music; int count; int analysis_offset; - /** Probability of having speech for time i to DETECT_SIZE-1 (and music before). - pspeech[0] is the probability that all frames in the window are speech. */ - float pspeech[DETECT_SIZE]; - /** Probability of having music for time i to DETECT_SIZE-1 (and speech before). - pmusic[0] is the probability that all frames in the window are music. */ - float pmusic[DETECT_SIZE]; - float speech_confidence; - float music_confidence; - int speech_confidence_count; - int music_confidence_count; int write_pos; int read_pos; int read_subframe; float hp_ener_accum; + int initialized; + float rnn_state[MAX_NEURONS]; opus_val32 downmix_state[3]; AnalysisInfo info[DETECT_SIZE]; } TonalityAnalysisState; diff --git a/code/opus-1.5.2/src/extensions.c b/code/opus-1.5.2/src/extensions.c new file mode 100644 index 00000000..bb6c0b02 --- /dev/null +++ b/code/opus-1.5.2/src/extensions.c @@ -0,0 +1,315 @@ +/* Copyright (c) 2022 Amazon */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + + +#include "opus_types.h" +#include "opus_defines.h" +#include "arch.h" +#include "os_support.h" +#include "opus_private.h" + + +/* Given an extension payload, advance data to the next extension and return the + length of the remaining extensions. */ +opus_int32 skip_extension(const unsigned char **data, opus_int32 len, opus_int32 *header_size) +{ + int id, L; + if (len==0) + return 0; + id = **data>>1; + L = **data&1; + if (id == 0 && L == 1) + { + *header_size = 1; + if (len < 1) + return -1; + (*data)++; + len--; + return len; + } else if (id > 0 && id < 32) + { + if (len < 1+L) + return -1; + *data += 1+L; + len -= 1+L; + *header_size = 1; + return len; + } else { + if (L==0) + { + *data += len; + *header_size = 1; + return 0; + } else { + opus_int32 bytes=0; + *header_size = 1; + do { + (*data)++; + len--; + if (len == 0) + return -1; + bytes += **data; + (*header_size)++; + } while (**data == 255); + (*data)++; + len--; + if (bytes <= len) + { + len -= bytes; + *data += bytes; + } else { + return -1; + } + return len; + } + } +} + +/* Count the number of extensions, excluding real padding and separators. */ +opus_int32 opus_packet_extensions_count(const unsigned char *data, opus_int32 len) +{ + opus_int32 curr_len; + opus_int32 count=0; + const unsigned char *curr_data = data; + + celt_assert(len >= 0); + celt_assert(data != NULL || len == 0); + + curr_len = len; + while (curr_len > 0) + { + int id; + opus_int32 header_size; + id = *curr_data>>1; + curr_len = skip_extension(&curr_data, curr_len, &header_size); + if (curr_len < 0) + return OPUS_INVALID_PACKET; + if (id > 1) + count++; + } + return count; +} + +/* Extract extensions from Opus padding (excluding real padding and separators) */ +opus_int32 opus_packet_extensions_parse(const unsigned char *data, opus_int32 len, opus_extension_data *extensions, opus_int32 *nb_extensions) +{ + const unsigned char *curr_data; + opus_int32 curr_len; + int curr_frame=0; + opus_int32 count=0; + + celt_assert(len >= 0); + celt_assert(data != NULL || len == 0); + celt_assert(nb_extensions != NULL); + celt_assert(extensions != NULL || *nb_extensions == 0); + + curr_data = data; + curr_len = len; + while (curr_len > 0) + { + int id; + opus_int32 header_size; + opus_extension_data curr_ext; + id = *curr_data>>1; + if (id > 1) + { + curr_ext.id = id; + curr_ext.frame = curr_frame; + curr_ext.data = curr_data; + } else if (id == 1) + { + int L = *curr_data&1; + if (L==0) + curr_frame++; + else { + if (curr_len >= 2) + curr_frame += curr_data[1]; + /* Else we're at the end and it doesn't matter. */ + } + if (curr_frame >= 48) + { + *nb_extensions = count; + return OPUS_INVALID_PACKET; + } + } + curr_len = skip_extension(&curr_data, curr_len, &header_size); + /* printf("curr_len = %d, header_size = %d\n", curr_len, header_size); */ + if (curr_len < 0) + { + *nb_extensions = count; + return OPUS_INVALID_PACKET; + } + celt_assert(curr_data - data == len - curr_len); + if (id > 1) + { + if (count == *nb_extensions) + { + return OPUS_BUFFER_TOO_SMALL; + } + curr_ext.len = curr_data - curr_ext.data - header_size; + curr_ext.data += header_size; + extensions[count++] = curr_ext; + } + } + celt_assert(curr_len == 0); + *nb_extensions = count; + return OPUS_OK; +} + +opus_int32 opus_packet_extensions_generate(unsigned char *data, opus_int32 len, const opus_extension_data *extensions, opus_int32 nb_extensions, int pad) +{ + int max_frame=0; + opus_int32 i; + int frame; + int curr_frame = 0; + opus_int32 pos = 0; + opus_int32 written = 0; + + celt_assert(len >= 0); + + for (i=0;i 127) + return OPUS_BAD_ARG; + } + if (max_frame >= 48) return OPUS_BAD_ARG; + for (frame=0;frame<=max_frame;frame++) + { + for (i=0;i 1) + return OPUS_BAD_ARG; + if (len-pos < extensions[i].len+1) + return OPUS_BUFFER_TOO_SMALL; + if (data) data[pos] = (extensions[i].id<<1) + extensions[i].len; + pos++; + if (extensions[i].len > 0) { + if (data) data[pos] = extensions[i].data[0]; + pos++; + } + } else { + int last; + opus_int32 length_bytes; + if (extensions[i].len < 0) + return OPUS_BAD_ARG; + last = (written == nb_extensions - 1); + length_bytes = 1 + extensions[i].len/255; + if (last) + length_bytes = 0; + if (len-pos < 1 + length_bytes + extensions[i].len) + return OPUS_BUFFER_TOO_SMALL; + if (data) data[pos] = (extensions[i].id<<1) + !last; + pos++; + if (!last) + { + opus_int32 j; + for (j=0;j +int main() +{ + opus_extension_data ext[] = {{2, 0, (const unsigned char *)"a", 1}, + {32, 10, (const unsigned char *)"DRED", 4}, + {33, 1, (const unsigned char *)"NOT DRED", 8}, + {3, 4, (const unsigned char *)NULL, 0} + }; + opus_extension_data ext2[10]; + int i, len; + int nb_ext = 10; + unsigned char packet[10000]; + len = opus_packet_extensions_generate(packet, 32, ext, 4, 1); + for (i=0;i 255 || cols > 255) + return 0; + size = rows * (opus_int32)cols * sizeof(opus_int16); + if (size > 65004) + return 0; + + return align(sizeof(MappingMatrix)) + align(size); +} + +opus_int16 *mapping_matrix_get_data(const MappingMatrix *matrix) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (opus_int16*)(void*)((char*)matrix + align(sizeof(MappingMatrix))); +} + +void mapping_matrix_init(MappingMatrix * const matrix, + int rows, int cols, int gain, const opus_int16 *data, opus_int32 data_size) +{ + int i; + opus_int16 *ptr; + +#if !defined(ENABLE_ASSERTIONS) + (void)data_size; +#endif + celt_assert(align(data_size) == align(rows * cols * sizeof(opus_int16))); + + matrix->rows = rows; + matrix->cols = cols; + matrix->gain = gain; + ptr = mapping_matrix_get_data(matrix); + for (i = 0; i < rows * cols; i++) + { + ptr[i] = data[i]; + } +} + +#ifndef DISABLE_FLOAT_API +void mapping_matrix_multiply_channel_in_float( + const MappingMatrix *matrix, + const float *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, col; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { + float tmp = 0; + for (col = 0; col < input_rows; col++) + { + tmp += + matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + input[MATRIX_INDEX(input_rows, col, i)]; + } +#if defined(FIXED_POINT) + output[output_rows * i] = FLOAT2INT16((1/32768.f)*tmp); +#else + output[output_rows * i] = (1/32768.f)*tmp; +#endif + } +} + +void mapping_matrix_multiply_channel_out_float( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + float *output, + int output_rows, + int frame_size +) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, row; + float input_sample; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { +#if defined(FIXED_POINT) + input_sample = (1/32768.f)*input[input_rows * i]; +#else + input_sample = input[input_rows * i]; +#endif + for (row = 0; row < output_rows; row++) + { + float tmp = + (1/32768.f)*matrix_data[MATRIX_INDEX(matrix->rows, row, input_row)] * + input_sample; + output[MATRIX_INDEX(output_rows, row, i)] += tmp; + } + } +} +#endif /* DISABLE_FLOAT_API */ + +void mapping_matrix_multiply_channel_in_short( + const MappingMatrix *matrix, + const opus_int16 *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, col; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { + opus_val32 tmp = 0; + for (col = 0; col < input_rows; col++) + { +#if defined(FIXED_POINT) + tmp += + ((opus_int32)matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + (opus_int32)input[MATRIX_INDEX(input_rows, col, i)]) >> 8; +#else + tmp += + matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + input[MATRIX_INDEX(input_rows, col, i)]; +#endif + } +#if defined(FIXED_POINT) + output[output_rows * i] = (opus_int16)((tmp + 64) >> 7); +#else + output[output_rows * i] = (1/(32768.f*32768.f))*tmp; +#endif + } +} + +void mapping_matrix_multiply_channel_out_short( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + opus_int16 *output, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, row; + opus_int32 input_sample; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { +#if defined(FIXED_POINT) + input_sample = (opus_int32)input[input_rows * i]; +#else + input_sample = (opus_int32)FLOAT2INT16(input[input_rows * i]); +#endif + for (row = 0; row < output_rows; row++) + { + opus_int32 tmp = + (opus_int32)matrix_data[MATRIX_INDEX(matrix->rows, row, input_row)] * + input_sample; + output[MATRIX_INDEX(output_rows, row, i)] += (tmp + 16384) >> 15; + } + } +} + +const MappingMatrix mapping_matrix_foa_mixing = { 6, 6, 0 }; +const opus_int16 mapping_matrix_foa_mixing_data[36] = { + 16384, 0, -16384, 23170, 0, 0, 16384, 23170, + 16384, 0, 0, 0, 16384, 0, -16384, -23170, + 0, 0, 16384, -23170, 16384, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_soa_mixing = { 11, 11, 0 }; +const opus_int16 mapping_matrix_soa_mixing_data[121] = { + 10923, 7723, 13377, -13377, 11585, 9459, 7723, -16384, + -6689, 0, 0, 10923, 7723, 13377, 13377, -11585, + 9459, 7723, 16384, -6689, 0, 0, 10923, -15447, + 13377, 0, 0, -18919, 7723, 0, 13377, 0, + 0, 10923, 7723, -13377, -13377, 11585, -9459, 7723, + 16384, -6689, 0, 0, 10923, -7723, 0, 13377, + -16384, 0, -15447, 0, 9459, 0, 0, 10923, + -7723, 0, -13377, 16384, 0, -15447, 0, 9459, + 0, 0, 10923, 15447, 0, 0, 0, 0, + -15447, 0, -18919, 0, 0, 10923, 7723, -13377, + 13377, -11585, -9459, 7723, -16384, -6689, 0, 0, + 10923, -15447, -13377, 0, 0, 18919, 7723, 0, + 13377, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767 +}; + +const MappingMatrix mapping_matrix_toa_mixing = { 18, 18, 0 }; +const opus_int16 mapping_matrix_toa_mixing_data[324] = { + 8208, 0, -881, 14369, 0, 0, -8192, -4163, + 13218, 0, 0, 0, 11095, -8836, -6218, 14833, + 0, 0, 8208, -10161, 881, 10161, -13218, -2944, + -8192, 2944, 0, -10488, -6218, 6248, -11095, -6248, + 0, -10488, 0, 0, 8208, 10161, 881, -10161, + -13218, 2944, -8192, -2944, 0, 10488, -6218, -6248, + -11095, 6248, 0, 10488, 0, 0, 8176, 5566, + -11552, 5566, 9681, -11205, 8192, -11205, 0, 4920, + -15158, 9756, -3334, 9756, 0, -4920, 0, 0, + 8176, 7871, 11552, 0, 0, 15846, 8192, 0, + -9681, -6958, 0, 13797, 3334, 0, -15158, 0, + 0, 0, 8176, 0, 11552, 7871, 0, 0, + 8192, 15846, 9681, 0, 0, 0, 3334, 13797, + 15158, 6958, 0, 0, 8176, 5566, -11552, -5566, + -9681, -11205, 8192, 11205, 0, 4920, 15158, 9756, + -3334, -9756, 0, 4920, 0, 0, 8208, 14369, + -881, 0, 0, -4163, -8192, 0, -13218, -14833, + 0, -8836, 11095, 0, 6218, 0, 0, 0, + 8208, 10161, 881, 10161, 13218, 2944, -8192, 2944, + 0, 10488, 6218, -6248, -11095, -6248, 0, -10488, + 0, 0, 8208, -14369, -881, 0, 0, 4163, + -8192, 0, -13218, 14833, 0, 8836, 11095, 0, + 6218, 0, 0, 0, 8208, 0, -881, -14369, + 0, 0, -8192, 4163, 13218, 0, 0, 0, + 11095, 8836, -6218, -14833, 0, 0, 8176, -5566, + -11552, 5566, -9681, 11205, 8192, -11205, 0, -4920, + 15158, -9756, -3334, 9756, 0, -4920, 0, 0, + 8176, 0, 11552, -7871, 0, 0, 8192, -15846, + 9681, 0, 0, 0, 3334, -13797, 15158, -6958, + 0, 0, 8176, -7871, 11552, 0, 0, -15846, + 8192, 0, -9681, 6958, 0, -13797, 3334, 0, + -15158, 0, 0, 0, 8176, -5566, -11552, -5566, + 9681, 11205, 8192, 11205, 0, -4920, -15158, -9756, + -3334, -9756, 0, 4920, 0, 0, 8208, -10161, + 881, -10161, 13218, -2944, -8192, -2944, 0, -10488, + 6218, 6248, -11095, 6248, 0, 10488, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_fourthoa_mixing = { 27, 27, 0 }; +const opus_int16 mapping_matrix_fourthoa_mixing_data[729] = { + 9243, 0, 16010, 0, 0, 0, 20669, 0, + 0, 0, 0, 0, 24456, 0, 0, 0, + 0, 0, 0, 0, 27731, 0, 0, 0, + 0, 0, 0, 9243, 0, 10884, 11741, 0, + 0, 3995, 17849, 9626, 0, 0, 0, -5727, + 14399, 17315, 7625, 0, 0, 0, 0, -11747, + 2574, 18637, 15552, 5930, 0, 0, 9243, -14302, + -2682, -6677, 13337, 5357, -9464, 2501, -11170, 4770, + -5911, 11501, 5858, 5369, 4951, 17901, -19071, -2397, + -9281, -9198, 7576, -4294, 7773, -8997, -3399, 0, + 0, 9243, 9940, 11991, -3705, -5144, 16647, 7057, + -6206, -5941, -2698, -10194, 16781, -1788, -6256, -11772, + 4935, 3912, -6062, -13039, 9446, -9758, -3521, -15058, + 11089, 565, 0, 0, 9243, -15376, 3720, 2461, + -5285, -7989, -8660, 1278, -16087, 15811, -3249, 10500, + -7757, -1680, -9890, -8153, 10884, 11022, 2847, 12828, + 5137, -2053, 8666, -5684, 14776, 0, 0, 9243, + -10577, 10304, -6186, 9139, -15222, 2507, -8902, -5140, + -145, 15562, -10596, -7311, -6197, -8753, 8667, -6014, + -281, 15033, 938, -11859, 548, -8456, 16735, -3654, + 0, 0, 9243, 8974, 4839, -12343, -15472, 6066, + -7501, -8343, 5015, 15920, -12374, -4559, -9400, 6271, + 4011, 5191, -9932, 14438, 4828, -8768, 1909, 12059, + -1565, 4707, -13711, 0, 0, 9243, 15799, 2085, + -1534, -3386, 4602, -9808, -447, -17267, -18054, -1167, + -13525, -4644, 1313, -5951, 5397, 7485, -7056, 2584, + -8120, 8669, 788, 13177, 2109, 18349, 0, 0, + 9243, 12371, -10036, 1597, 2760, -17341, 1848, -2239, + -10509, -8474, -4577, 11164, 7935, 1441, 17430, -3436, + -3713, 15936, 4184, 2647, -11730, 341, -15934, 6462, + 6581, 0, 0, 9243, -8963, 2184, 13084, -16381, + -2734, -9757, 3991, 6345, -18297, -5912, 7604, -4849, + -11100, 2290, -4304, -13305, -7488, 12338, 4805, 8505, + -7014, -4779, -1761, -14597, 0, 0, 9243, 1301, + -15498, 3799, 690, -2816, 18718, -8223, 889, 255, + -1768, 4485, -19951, 13097, -2278, 167, 78, -740, + 3324, -6139, 19488, -17925, 4283, -486, 20, 0, + 0, 9243, -13470, -6719, 5452, -10257, 12641, -4873, + -5116, -10595, 5856, 11389, 1502, 10876, -608, 11765, + -13218, 13911, -7373, -2070, -13679, -4154, 5536, -2138, + 16643, 451, 0, 0, 9243, 2455, -3679, -15387, + -5277, -1261, -8697, 7906, 16112, 8147, 3208, -1690, + 7687, 10593, -9796, -15852, -10884, -5616, 2881, 2032, + 5246, -12735, -8796, 10928, 14833, 0, 0, 9243, + -6849, 2775, -14202, 13586, -2655, -9402, -5505, 10809, + -18013, 6231, 5444, -6041, 11288, 4958, -4078, 18799, + -9368, -9291, 4535, 7383, 9405, -7391, -2121, -4336, + 0, 0, 9243, 6423, -9040, 11548, 10359, -8109, + -450, -14580, 6431, 10857, -15475, 3569, 9707, 6416, + -9607, 521, 8528, -18391, 11049, 3815, -10423, 6860, + 6860, -883, -4221, 0, 0, 9243, 11932, -5968, + -8850, -14749, -9946, -6026, 7377, -4472, 5206, 14547, + -3406, 10508, 2526, 4411, 14543, 8444, -5822, 347, + 12347, -1709, -9158, 105, -16265, -12642, 0, 0, + 9243, 13044, -150, 9282, 16910, -274, -10332, -194, + -5864, 5428, -420, -12196, 344, -8679, 145, -18554, + -12695, -152, -14635, 503, 10389, 358, 5076, 522, + -16100, 0, 0, 9243, -8374, -13590, -1221, 1428, + 15896, 12005, 2318, -4793, 2590, -3209, -20390, -6256, + -2974, 10766, 1202, -876, -6597, 5004, 19896, -1541, + 2902, -16788, -3062, 1340, 0, 0, 9243, 9879, + 10267, 7300, 10073, 14167, 2416, 10469, -3094, 2899, + 17092, 9762, -7400, 7214, -5250, -8238, -3989, 5578, + 16392, -1050, -11848, -776, -5034, -15850, -5882, 0, + 0, 9243, -4974, -9068, 12221, -8490, 6299, -388, + -15478, 8702, -9920, 12723, -2810, 9668, 6905, -13040, + 4325, -9456, 16856, -9159, -2909, -10476, 7149, 9387, + -7350, 233, 0, 0, 9243, 3627, -13823, -7218, + -3656, -7002, 12776, 13935, 2719, 2446, 8352, 9252, + -7676, -18413, -6212, -429, -1272, -6335, -13356, -9510, + 295, 18926, 9934, 1112, -382, 0, 0, 9243, + -6383, -9343, -11326, 10097, 8329, 223, 14780, 6114, + -10348, -15590, -4195, 9257, -7445, -9439, -323, 7902, + 18117, 12101, -3142, -10944, -5577, 7327, 566, -4133, + 0, 0, 9243, 2626, 865, 15769, 5783, 317, + -10244, 1905, 16884, 9144, 826, -2420, -1972, -14536, + 2413, 16939, 12500, 1482, -4906, -578, 10096, -3476, + -14323, 2745, 16105, 0, 0, 9243, -8975, 12086, + 5450, -6832, -15149, 7333, 9200, -3550, -362, -13645, + -15525, -1391, 9428, -7091, -5442, 3105, -820, -17685, + -9175, -9462, 5572, -9191, -12325, -2180, 0, 0, + 9243, -114, 11576, -11058, 177, -185, 5875, -17880, + 8539, -198, 339, -173, -3411, -16698, 16336, -6369, + 193, -430, 408, -75, -10806, -7225, 19670, -13817, + 4665, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767 +}; + +const MappingMatrix mapping_matrix_fifthoa_mixing = { 38, 38, 0 }; +const opus_int16 mapping_matrix_fifthoa_mixing_data[1444] = { + 9243, 0, 16010, 0, 0, 0, 20669, 0, + 0, 0, 0, 0, 24456, 0, 0, 0, + 0, 0, 0, 0, 27731, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 30657, 0, + 0, 0, 0, 0, 0, 0, 9243, 0, + -7023, 14387, 0, 0, -4369, -14112, 14455, 0, + 0, 0, 10931, -510, -16777, 14031, 0, 0, + 0, 0, -5118, 14286, 4343, -18465, 13374, 0, + 0, 0, 0, 0, -6494, -12221, 11761, 8513, + -19458, 12605, 0, 0, 9243, -14128, 5093, 5547, + -10946, -10050, -7197, 3945, -11790, 7142, -9213, 6529, + -9701, -2563, -9923, -14846, 16521, 6816, 2764, 14103, + 1118, -5537, 2977, -14168, 1228, 4866, 17430, -528, + 10639, 2641, 10437, -1037, 11460, 1098, 1296, 15737, + 0, 0, 9243, 1128, -14775, 6062, 955, -2329, + 16069, -12511, 2477, 579, -2333, 3440, -14197, 18478, + -6050, 940, 303, -1604, 4106, -4223, 9829, -22688, + 10647, -2604, 334, 145, -927, 3203, -6017, 4507, + -3812, 24212, -15600, 5198, -1023, 110, 0, 0, + 9243, 1158, 12997, 9277, 1501, 2103, 10097, 16840, + 5916, 1402, 3225, 2488, 2929, 19916, 12706, 3585, + 1137, 3415, 4698, 2078, -5442, 16634, 18511, 8731, + 2095, 850, 3061, 5733, 5225, 960, -11728, 7689, + 20588, 14659, 5642, 1187, 0, 0, 9243, -4663, + -3081, -15003, 9771, 2007, -9185, 6457, 14199, -14357, + -4976, 3554, 6625, 11434, -7231, -11297, 17760, 8291, + -6267, -3368, 6712, -10837, -9107, 6524, 6793, -19531, + -11338, 7934, 7335, -2205, -9215, -7094, 10659, 6243, + -4337, -1250, 0, 0, 9243, -13515, 7679, -3831, + 7232, -14496, -3201, -4109, -11731, 8828, 9178, -1901, + -10848, -539, -14888, 9626, -10860, 12703, 3824, 12334, + -7104, 3496, -6203, 13852, 5461, -2109, -17277, 7837, + -4714, 13901, 4097, 3940, 7647, 8546, 8688, -10986, + 0, 0, 9243, 8113, -9860, 9657, 10943, -11174, + 1426, -13300, 1915, 8178, -17833, 6805, 8309, 8100, + -3121, -4742, 2683, -15111, 15688, 2358, -11590, 2807, + 2746, 8762, -7430, -2251, -5481, 16370, -4081, -9694, + 5872, -11539, -714, -9492, 15177, -6126, 0, 0, + 9243, 9933, -9215, -8528, -11831, -12785, -62, 10976, + -1811, 5593, 18018, 6100, 9455, -5237, 2758, 8971, + 2743, -9659, -13517, 5330, -10737, -4576, -2069, -15491, + -8749, -7226, -5237, 9191, -181, -12277, 2815, 10540, + -27, 14741, 16703, 3103, 0, 0, 9243, -10067, + -8881, -8723, 12265, 12487, -793, 10821, -1762, -6021, + -18002, -5072, 9912, -4395, 2587, 9368, -2767, 10021, + 12259, -6468, -10113, -5605, -1761, -15590, -9430, 7800, + 5092, -8835, 2293, 12314, 1222, 10671, -329, 13745, + 17349, 3563, 0, 0, 9243, -6485, 12991, -6743, + 6108, -11768, 10080, -12236, 238, -2883, 13115, -13907, + 2900, -14460, 511, 2564, 186, -7019, 19094, -11597, + -5472, -12058, 744, 6243, -2384, 930, 501, -11778, + 21214, -5330, -11746, -5542, 827, 10475, -6418, 1132, + 0, 0, 9243, 3862, 5238, -14627, -7891, 2826, + -7015, -10701, 13900, 11410, -6831, -1679, -9861, 6359, + 12032, -11660, -14041, 11199, 1713, -3895, 657, 14749, + -3017, -11445, 8380, 15575, -15236, -346, 7690, -923, + 10317, 3498, -13545, 354, 9093, -4476, 0, 0, + 9243, -8417, 13183, 3418, -4018, -15498, 10685, 6294, + -4132, 1419, -8755, -18818, 3926, 7642, -9001, -3235, + 2125, 3506, -13037, -16570, -4337, 6729, -13404, -7991, + 59, 443, 5804, 6005, -15011, -9060, -11044, 3679, + -15434, -13685, 161, 1185, 0, 0, 9243, -5288, + 6773, -13508, 9977, -5002, -4784, -12780, 10790, -12942, + 11168, 519, -10890, 1326, 12078, -6274, 13780, -16427, + 2186, 5352, -4328, 13671, 2364, -7963, 1080, -12568, + 19336, -6557, -8574, 4084, 7277, 10433, -9273, -3178, + 1516, 3817, 0, 0, 9243, 9660, 7817, 10093, + 13619, 10548, -2942, 11021, 597, 9663, 17594, 1736, + -10794, 1814, 771, -8469, 1041, 14155, 7891, -8597, + -7498, -8982, 346, -12407, -11848, -6809, 1686, 9181, + -8306, -10247, 3538, -10706, -364, -8047, -19188, -8493, + 0, 0, 9243, -7163, -1020, 14282, -14289, 1021, + -10208, -2036, 10660, -18919, 2410, 6564, 2323, -13088, + -1798, 3365, -19498, 3619, 12022, -1858, 9978, 3705, + -8969, -643, -5794, -15523, 4123, 15113, -3949, -6265, + -3596, 12490, 2946, -2688, 1225, -14570, 0, 0, + 9243, -12187, 772, -10354, 17623, -1314, -10262, -1117, + -2885, -9937, 2249, 11267, -1763, 9572, -368, 16506, + -6510, -1438, -15014, 2402, 10157, 2041, 2458, 2389, + -19346, 19860, -1041, 8067, -3704, -10931, 2743, -9286, + 606, -13399, -3095, 7924, 0, 0, 9243, 15545, + -2367, -3011, -6538, -5139, -9657, 995, -16242, -15706, + 2557, -12952, 5226, 2508, 6353, 10156, 13593, 6966, + 4795, 8960, 8183, -1735, 11914, -4504, 14149, 11727, + -6665, 10460, -3962, 10145, -7648, -1965, -9845, -6764, + -6938, -16633, 0, 0, 9243, 3098, 12983, -8841, + -3826, 5618, 10053, -16031, 4787, 3283, -8209, 6632, + 2856, -18922, 10272, -2055, -2344, 7987, -11939, 5516, + -5520, -15739, 14940, -5001, 530, 1465, -6306, 13388, + -13243, 2513, -11772, -7170, 16572, -8384, 1426, 168, + 0, 0, 9243, -15767, -2008, -1916, 4220, 4422, + -9846, 537, -17105, 17650, -1400, 13589, 4481, 1651, + 5677, 6701, -9241, -6642, -3252, -7827, 8792, -951, + 13182, -2522, 17586, -17005, 3845, -12562, 2213, -11472, + -6688, -1394, -8970, -4769, -7316, -11753, 0, 0, + 9243, -13344, -3829, 7975, -14863, 7136, -8561, -4265, + -7992, -801, 9405, 8912, 7937, -5326, 5057, -17681, + 15207, 575, 7717, -11360, 4847, 6789, 4150, 12686, + -10050, 16730, -12063, 322, -12920, -3313, -10267, 1980, + -6948, 7112, 7972, 8042, 0, 0, 9243, 7791, + -1021, 13949, 15180, -1111, -10208, -1989, 9348, 19199, + -2561, -7140, 2323, -12782, -1577, 817, 18164, -3673, + -12771, 2022, 9978, 3620, -7865, -156, -9155, 11924, + -3842, -15336, 4196, 6814, -3596, 12199, 2583, -652, + 1936, -17637, 0, 0, 9243, -4810, -15144, -1958, + 1315, 10175, 17406, 4142, -1348, 263, -3292, -15632, + -17046, -6363, 3374, 605, -227, -748, 5997, 20334, + 14481, 8277, -6146, -1717, 5, 27, 712, 1542, + -9197, -23572, -10163, -9595, 9425, 3539, -17, -72, + 0, 0, 9243, -7366, 8261, 11568, -11901, -8499, + -2079, 13347, 5556, -12049, -16247, -2282, -10529, 3584, + 7585, -1577, -8464, -18652, -8902, 5913, -8688, -9287, + 4156, -2442, -7089, -2993, -14485, -13949, 5422, 8459, + 1638, -13285, -2531, -1826, -12132, -9456, 0, 0, + 9243, 11716, 698, -10889, -17818, 1143, -10275, -1062, + -1305, 12057, -2057, -10855, -1595, 10088, -150, 15043, + 2978, 1578, 15225, -2090, 10201, 1943, 1115, 1969, + -20211, -17636, 430, -9826, 3391, 10572, 2485, -9826, + 248, -12259, -2924, 12131, 0, 0, 9243, 4361, + -4594, -14703, -8956, -2798, -7781, 9434, 13769, 12936, + 6800, -2400, 9082, 8091, -10453, -11023, -15786, -11136, + 3285, 4153, 2658, -14002, -5051, 9489, 7000, 17206, + 15024, -2777, -8491, -42, -10626, 141, 13053, 2366, + -6662, -2231, 0, 0, 9243, -752, -11933, -10646, + 1119, 1254, 6890, 17745, 7875, -1203, -2207, -1251, + 2024, -17706, -15532, -5600, 1128, 2691, 2800, 683, + -9927, 9661, 19706, 12522, 3889, -978, -2789, -3992, + -2440, 206, 12695, 2921, -17173, -18575, -9616, -2657, + 0, 0, 9243, 4791, -15001, -2887, -1931, -10037, + 16885, 6048, -1020, 46, 4789, 15191, -15922, -9154, + 2530, 823, 252, -130, -8608, -19335, 12613, 11651, + -4549, -2314, -172, -101, -784, 265, 12975, 21741, + -7551, -13101, 6856, 4710, 535, -46, 0, 0, + 9243, -12153, -10395, 754, -1281, 17644, 2735, -1095, + -10274, 8359, 2200, -12593, 7083, 782, 17650, -1573, + 1685, -16282, -2164, -530, -11878, 32, -17359, 3065, + 6651, -5212, -3628, 19365, 965, 13180, 8243, -818, + 7746, -3645, -14323, 1670, 0, 0, 9243, -6961, + -11198, 9081, -8829, 10887, 4833, -14202, 2374, -6524, + 16339, -9417, 4737, 12284, -4394, -2691, -2683, 13690, + -18539, 2830, -11438, -3692, 4985, 5648, -4628, 514, + 6225, -18409, 12672, 5311, 11170, -6928, -3407, -7595, + 10737, -3977, 0, 0, 9243, 12099, -10405, 1294, + 2187, -17582, 2760, -1880, -10105, -8058, -3760, 12583, + 7058, 1346, 17376, -2667, -2829, 15710, 3705, 468, + -11880, 50, -17123, 5201, 6230, 4698, 6098, -18716, + -1665, -13088, 8285, -1400, 7696, -6196, -13429, 2770, + 0, 0, 9243, 8602, 13392, 1722, 2070, 16090, + 11359, 3222, -4960, -2638, 4581, 20106, 5099, 4026, + -10978, -1778, -1314, -6620, 6988, 18701, -2965, 3745, + -16745, -4461, 1300, 584, -3646, -11588, 8350, 11847, + -10050, 2372, -20010, -7809, 3608, 887, 0, 0, + 9243, 14252, -1958, 7026, 13986, -3899, -9870, -1922, + -10736, -3693, -4527, -12333, 4376, -6080, 3475, -18537, + -19222, 1355, -10843, 6913, 8869, 3408, 8323, 6804, + -5141, -13648, 7800, 2649, 7171, 10505, -6548, 5179, + -5505, 13299, 2086, 15579, 0, 0, 9243, 11323, + 9021, -6835, -10810, 14267, -489, -8613, -5689, 639, + -16117, 6224, -9731, -3757, -8482, 10882, 7873, 1080, + -11447, -6791, -10388, 4099, -6025, 18396, -5407, -7536, + 14714, 984, 1267, -13940, -1889, 8416, 666, 16762, + -10106, -3418, 0, 0, 9243, 871, 4833, 15238, + 1855, 588, -7508, 10287, 16162, 2857, 1481, -443, + -9392, -7758, 12910, 16506, 3837, 2588, -581, -851, + 1928, -14879, -5066, 14950, 16498, 4773, 3842, -425, + -1785, -82, 10578, -1435, -15554, -2459, 16520, 16250, + 0, 0, 9243, 14762, 5967, 1673, 3450, 12303, + -6027, 1394, -15022, -14571, 3402, -4217, -10507, -478, + -14813, -5131, -6634, -16293, -82, -15276, -1705, -1731, + 358, -5738, 13681, 12503, -8200, -3023, -3290, -7384, + 9272, -837, 14328, -1064, 16913, 7915, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_foa_demixing = { 6, 6, 0 }; +const opus_int16 mapping_matrix_foa_demixing_data[36] = { + 16384, 16384, 16384, 16384, 0, 0, 0, 23170, + 0, -23170, 0, 0, -16384, 16384, -16384, 16384, + 0, 0, 23170, 0, -23170, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_soa_demixing = { 11, 11, 3050 }; +const opus_int16 mapping_matrix_soa_demixing_data[121] = { + 2771, 2771, 2771, 2771, 2771, 2771, 2771, 2771, + 2771, 0, 0, 10033, 10033, -20066, 10033, 14189, + 14189, -28378, 10033, -20066, 0, 0, 3393, 3393, + 3393, -3393, 0, 0, 0, -3393, -3393, 0, + 0, -17378, 17378, 0, -17378, -24576, 24576, 0, + 17378, 0, 0, 0, -14189, 14189, 0, -14189, + -28378, 28378, 0, 14189, 0, 0, 0, 2399, + 2399, -4799, -2399, 0, 0, 0, -2399, 4799, + 0, 0, 1959, 1959, 1959, 1959, -3918, -3918, + -3918, 1959, 1959, 0, 0, -4156, 4156, 0, + 4156, 0, 0, 0, -4156, 0, 0, 0, + 8192, 8192, -16384, 8192, 16384, 16384, -32768, 8192, + -16384, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 8312, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 8312 +}; + +const MappingMatrix mapping_matrix_toa_demixing = { 18, 18, 0 }; +const opus_int16 mapping_matrix_toa_demixing_data[324] = { + 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, + 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, + 0, 0, 0, -9779, 9779, 6263, 8857, 0, + 6263, 13829, 9779, -13829, 0, -6263, 0, -8857, + -6263, -9779, 0, 0, -3413, 3413, 3413, -11359, + 11359, 11359, -11359, -3413, 3413, -3413, -3413, -11359, + 11359, 11359, -11359, 3413, 0, 0, 13829, 9779, + -9779, 6263, 0, 8857, -6263, 0, 9779, 0, + -13829, 6263, -8857, 0, -6263, -9779, 0, 0, + 0, -15617, -15617, 6406, 0, 0, -6406, 0, + 15617, 0, 0, -6406, 0, 0, 6406, 15617, + 0, 0, 0, -5003, 5003, -10664, 15081, 0, + -10664, -7075, 5003, 7075, 0, 10664, 0, -15081, + 10664, -5003, 0, 0, -8176, -8176, -8176, 8208, + 8208, 8208, 8208, -8176, -8176, -8176, -8176, 8208, + 8208, 8208, 8208, -8176, 0, 0, -7075, 5003, + -5003, -10664, 0, 15081, 10664, 0, 5003, 0, + 7075, -10664, -15081, 0, 10664, -5003, 0, 0, + 15617, 0, 0, 0, -6406, 6406, 0, -15617, + 0, -15617, 15617, 0, 6406, -6406, 0, 0, + 0, 0, 0, -11393, 11393, 2993, -4233, 0, + 2993, -16112, 11393, 16112, 0, -2993, 0, 4233, + -2993, -11393, 0, 0, 0, -9974, -9974, -13617, + 0, 0, 13617, 0, 9974, 0, 0, 13617, + 0, 0, -13617, 9974, 0, 0, 0, 5579, + -5579, 10185, 14403, 0, 10185, -7890, -5579, 7890, + 0, -10185, 0, -14403, -10185, 5579, 0, 0, + 11826, -11826, -11826, -901, 901, 901, -901, 11826, + -11826, 11826, 11826, -901, 901, 901, -901, -11826, + 0, 0, -7890, -5579, 5579, 10185, 0, 14403, + -10185, 0, -5579, 0, 7890, 10185, -14403, 0, + -10185, 5579, 0, 0, -9974, 0, 0, 0, + -13617, 13617, 0, 9974, 0, 9974, -9974, 0, + 13617, -13617, 0, 0, 0, 0, 16112, -11393, + 11393, -2993, 0, 4233, 2993, 0, -11393, 0, + -16112, -2993, -4233, 0, 2993, 11393, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_fourthoa_demixing = { 27, 27, 0 }; +const opus_int16 mapping_matrix_fourthoa_demixing_data[729] = { + 4870, 4484, 4870, 4347, 4440, 4726, 4683, 4821, + 4883, 4842, 4603, 4484, 4683, 4698, 4234, 4368, + 4603, 4783, 4783, 4820, 4821, 4347, 4820, 4440, + 4698, 0, 0, 101, 84, -7818, 4640, -7178, + -5492, 4629, 8384, 6547, -4966, 617, -6345, 1061, + -3241, 2939, 5549, 6390, -4434, 4994, -2610, 1993, + -2873, 1547, -4356, -164, 0, 0, 8797, 5074, + -1553, 5383, 1906, 5297, 2722, 1158, -5226, 1311, + -7760, -3327, -1940, 1586, -4093, -2951, -214, -6873, + 5450, -4875, -7193, -4438, 558, 5593, 5607, 0, + 0, -26, 5761, -3723, -1460, 1195, -3065, -6357, + -1175, 608, 6965, 2310, 2759, -8023, -7138, 5162, + -3624, 5006, -809, 3592, 6209, -4159, -4968, 8150, + 2513, -5702, 0, 0, 301, 109, 7161, -2462, + -2443, 5044, -7125, -2256, 1967, -9107, 259, -4928, + -2592, 6514, 4111, -7236, 8695, 635, 5009, -4025, + -1937, 4794, 3420, -3507, -400, 0, 0, -134, + 85, 2771, 7842, -3649, -8225, 2866, 2586, -9200, + -1945, -1563, 6155, -720, -1061, -3494, -4513, -487, + 8389, 7317, 3348, -3721, 3806, 371, -6896, 70, + 0, 0, 10919, 2072, -4867, 3472, -4429, 1721, + -4066, -5193, 1032, -5253, 9501, -2017, -3971, -5261, + -306, -2737, -5137, 5713, 1237, -8, 6387, 364, + -5423, 3364, 2888, 0, 0, -48, 8946, 1048, + -2691, 602, -4332, -4302, -514, -1730, 2459, -4328, + -2156, 3335, -2748, -6029, 4023, 155, 897, 5268, + -8380, 7625, 7395, 508, 3945, -8951, 0, 0, + 39, 4151, -5965, -3398, -7006, -3534, 2697, -8989, + -5237, 2913, 46, -5540, 8196, 5766, 2711, -2520, + -3043, -2146, -948, 4965, 1806, 2472, 8988, -1266, + 4840, 0, 0, -407, -189, 2179, -1627, 6516, + 259, 7196, -9449, -4905, -9766, 561, 4021, 3371, + -8650, 5032, 3329, 2534, 641, 2224, -5747, 1047, + -4074, 5252, -24, 674, 0, 0, 664, 237, + -2837, -4072, -1205, 8252, -5875, -1670, -2743, -3984, + 381, 5059, 1765, 2666, -8295, 7403, 1154, -2086, + 7622, 7105, 3677, -6943, 1050, -6632, -694, 0, + 0, 382, -133, 5699, 7650, 5154, -5713, -1645, + -6902, 6181, 4450, 1151, 410, -993, 3829, 2444, + -2405, -6618, -9514, 5366, -1896, 5844, -2886, -1524, + -7321, -1007, 0, 0, 12767, -2530, 3183, -1409, + -4015, -2894, -5155, -1710, 3841, -2107, -10274, 5119, + 3979, -4010, 5550, 4822, -746, -2507, -3080, 4289, + -3675, 4333, -1416, -1230, -1122, 0, 0, 17, + 8048, 2398, -2167, -73, -3606, 3125, 398, 731, + -5973, 5705, -1032, 4679, 7305, 3134, 1301, -3858, + -89, 2938, 4359, -9155, -4805, -8407, 3673, -8645, + 0, 0, 187, 7355, 3145, -6719, -4432, -5939, + 2541, -2810, 9723, 778, -1105, 5687, -4174, 2534, + -4461, 1017, -244, 5481, -1655, -6765, -3350, -4894, + 1592, -2318, 8827, 0, 0, 196, 3588, 9631, + 3063, -4564, 6043, 2683, 2595, -2488, -2186, 173, + -6059, -8270, -2386, 409, 7441, -8608, 376, -4364, + 2321, -280, 97, 8331, -3022, -4721, 0, 0, + 117, -748, -10833, 1533, 4200, -2875, -997, -109, + -3661, -6119, 4454, 8808, -9189, 8294, 1521, 7265, + -2348, -5094, -948, -5400, -3193, 8914, 5763, 1716, + -1070, 0, 0, 2497, 399, -5201, -2038, 7843, + -376, 7567, -5073, 7616, -5537, 2086, -3453, -5544, + -56, -11648, -1314, 3546, -3432, -117, 8694, -4245, + 9621, 3098, -2582, -2351, 0, 0, 4386, -3104, + -3132, -10512, 566, 5217, 5128, 4967, 1348, 7035, + -1470, 91, -125, -3548, 8244, -3029, -10033, 2186, + 9745, -6440, -2074, 3638, -1477, -7045, -562, 0, + 0, 2154, 8116, -6102, 6570, 12998, -712, -4126, + -4996, 30, 1571, -6393, -12794, 425, 5036, 1190, + 5763, 5653, 12933, -6671, 5197, -2964, -3316, -6354, + -10554, -2652, 0, 0, 12618, -3737, 93, -5901, + 4262, -3364, 4444, 3103, -2767, 3403, 4925, -2584, + -989, 4977, -3714, -1965, 3076, 326, -2946, -2568, + 1026, -2980, 3362, -6132, -5966, 0, 0, 6001, + 48, -1979, -7275, 3476, -2096, 10591, 3793, 2629, + -447, -14747, -3689, -5525, 8358, 6883, -9703, -4556, + 7471, 2965, 4056, 13221, -7327, -3073, -2353, -6720, + 0, 0, 621, 11034, -44, -2828, 5978, -1850, + -1772, 3894, -7471, -1397, 945, -2028, -2928, -2240, + 3172, 2222, 4544, -4243, -5645, 3745, 2573, 3511, + -8206, -7286, 5700, 0, 0, 321, 10818, -4982, + 7813, -749, 9907, 1360, -1443, 568, -1103, 2305, + 6045, 2270, -1063, -1920, -3073, 5893, -3476, -11346, + -1657, -588, 2957, -2287, -8527, -8041, 0, 0, + 119, -268, 2372, -3040, 4979, -3789, -5630, 10619, + 5900, -5109, -4585, -3862, 10467, -3527, -385, -10034, + -9991, 4860, 984, 2362, 2311, -6804, 6324, 433, + 5291, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767 +}; + +const MappingMatrix mapping_matrix_fifthoa_demixing = { 38, 38, 0 }; +const opus_int16 mapping_matrix_fifthoa_demixing_data[1444] = { + 3188, 3247, 3268, 3368, 3368, 3138, 3268, 3099, + 3211, 3368, 3099, 3247, 3211, 3368, 3368, 3368, + 3149, 3268, 3247, 3211, 3099, 3188, 3138, 3149, + 3099, 3188, 3368, 3149, 3188, 3247, 3268, 3138, + 3211, 3368, 3138, 3149, 0, 0, 118, -47, + -5011, 282, 333, -1497, -4584, 2908, 3388, -3647, + -2493, 1139, -2882, -1719, 3604, -2543, -4328, 5443, + 1286, -5498, -4583, 2510, -1743, -2556, 4168, 1446, + -290, 1812, -4074, -2377, 4152, 2847, 4991, 3980, + 393, 5072, 0, 0, 5489, -2235, 1507, -5326, + 4609, -1096, 2926, -3427, -3301, -3078, 4226, 1730, + 4627, 2561, 2966, -592, 143, -677, 4617, -755, + -956, -433, -5138, 3037, 157, -1394, -4498, -4984, + -3661, -4112, -3756, 4628, -570, 3356, 1605, 1803, + 0, 0, -162, 5162, 2132, 2392, 3556, -5141, + -1536, 2975, -3001, -3350, -2231, -5230, 1294, -4965, + 3494, 5230, -3292, -1359, -2945, -773, 2670, 4867, + -660, 3720, -3415, -5112, -3700, -1211, 407, 3013, + 763, 591, 2481, -2657, 5210, 784, 0, 0, + -156, 338, -4246, 510, 462, 3296, 2846, 3333, + -4292, 4574, 1940, -2986, -1275, 3701, 5022, -5250, + 5780, -2676, -1180, 1516, -4852, 4877, 342, -3923, + -5703, -2920, 379, -657, -361, -3346, 1044, 795, + 5257, -4004, 698, 1115, 0, 0, 47, -140, + -3292, -1097, 652, 855, -5260, -3691, -4470, 4521, + -3863, 1093, -5552, -2016, 3831, 334, -456, -1532, + 2068, 1788, 2054, -295, 3668, -2820, 328, -994, + 295, -3301, 5770, 4282, -6353, 5632, -1371, 5005, + 238, 4041, 0, 0, 6764, -1659, -2730, 5726, + 3715, -3216, -933, 531, -52, -345, 3022, -2818, + 4005, -1617, -1189, -3748, -3403, -3592, 4040, -3553, + -2806, -3444, 6023, -711, -3298, -2503, 2548, 5564, + 940, 1848, 1207, 4010, -3488, -358, -2511, -1966, + 0, 0, -64, -5039, 1403, -4455, 6240, 2189, + -1716, -4348, 4183, 3951, -4042, -3606, 2399, -4563, + 4050, -612, -395, 348, -5791, 391, -1440, -735, + 1398, 4359, -518, 2969, 6556, 1951, -518, -4993, + -925, 998, -569, -2934, 3460, 420, 0, 0, + 16, 5482, -4122, 770, 2082, 5020, -3961, 485, + -584, -793, 3, 5222, -1416, 3673, 78, 3549, + -937, -5723, 1673, -6162, -2540, 3082, -355, 1838, + -615, 4601, 2832, -359, -3346, 668, -3393, -1583, + -3774, -2206, 5754, -4961, 0, 0, -328, 299, + 2470, 317, 525, -4494, 2805, 2617, 2383, -2363, + -1037, 4085, 895, -4622, 3218, -6607, -3381, -5933, + 1397, 6394, -446, 5694, 14, -4510, 4329, 3690, + -334, 0, 2932, -2478, -2944, -577, -599, -230, + 1553, -4736, 0, 0, -324, 142, -3252, -867, + 1111, -1882, 3378, -6055, 6502, -6840, 4280, -2694, + -2876, 4190, 6454, 655, 1061, 626, -2669, -798, + 3192, -985, -898, -5482, -548, 2315, -558, 1302, + 900, 5747, -1325, 1599, -1384, -5749, 624, 1110, + 0, 0, 321, 312, 2188, 1322, 237, 708, + -304, 2463, 1500, -1094, -5112, -1010, -6799, 646, + 992, 1969, 3423, -3996, 2628, 4451, 3432, -2833, + -6101, -330, -3768, -3, -707, 5961, -4037, -3736, + 4080, 7254, -4113, 2151, 54, -2150, 0, 0, + 7735, 4064, -3884, -5240, 577, 2229, -3947, 2914, + 3555, 4011, 774, -3519, 1985, -3701, -3824, 330, + -905, 2085, 1155, 2176, 3006, 340, -5533, -3264, + -902, 3114, 344, -5060, 1524, 1805, 1926, 2350, + 1905, -3203, -2762, -4162, 0, 0, 193, -151, + -1434, 6289, 7354, 4234, 169, 2868, -1977, -1375, + -4987, 2345, 2742, 599, 939, -4837, 2688, 991, + -6907, 716, -1542, -4346, -1833, 1493, 3134, 2903, + -7019, -2835, 93, 4395, 621, 870, -2357, -975, + -2933, -127, 0, 0, -616, -5968, -3479, -1651, + 4932, -2445, -5512, -1451, 691, 739, 479, 4227, + -2886, 3853, 8, -501, 188, 1990, 3842, 2270, + 1662, -174, 1290, 2456, 67, -3267, -5535, 483, + 5721, -1642, 6501, -3432, 1184, -3246, 4101, -4880, + 0, 0, -465, 5264, -4812, 682, 1683, -4539, + 2916, -1985, 2899, 3324, 1060, -4398, -745, -2137, + -3827, 1044, 6225, 3609, -532, 1980, -6001, 564, + -209, -1299, 5336, -3605, -1484, 37, 19, -1295, + -665, -385, -6773, 3651, 6153, -1291, 0, 0, + 193, -415, 5166, -110, 626, 6743, -2860, 1425, + 1101, -1341, 80, -4533, 249, 4231, -119, -6009, + -2970, 5170, -822, -2610, 4527, 5948, 182, -2589, + 837, -5471, 371, -43, 373, -665, -1233, -626, + -7353, 2606, 1339, -1398, 0, 0, -533, 147, + 2075, -672, 1043, 3503, 4402, -4971, -3287, 3731, + -2606, 3817, 1972, -5603, 5114, 1185, -1318, 1906, + 3018, -1999, 343, -1943, 207, -6744, 913, -4060, + 645, -349, -5667, 4766, 5575, -1733, 1116, 160, + 1534, -5690, 0, 0, -137, -36, 1556, 1325, + 1553, -2230, 1188, 5296, -5104, 4673, 6295, 498, + -4723, 933, 2994, 4067, -4700, 1758, -4116, -1252, + 2444, -4092, 1653, -2802, 5069, 1133, 790, -2355, + -934, -6304, 1642, 2045, -4259, -3873, -213, 215, + 0, 0, -364, 423, 4888, -1316, 118, -950, + 4027, 114, 2961, -3136, -3012, -883, -6192, 1340, + -3210, -1193, 1376, 3128, 1596, -2994, -3194, 533, + 8502, 2487, -1485, 1032, 301, -8007, -577, 887, + 297, 7778, 3121, -1901, -94, -6401, 0, 0, + 9260, -1845, 668, 2787, -2255, 2699, -2512, -3737, + -3675, -3601, -1803, 210, -1701, -1442, -2700, 3457, + 2868, 2079, -2113, 3178, 1277, 3578, 5240, -2482, + 3324, 1020, -4027, 3835, -3758, -3633, -3170, -1310, + 2509, -3110, 713, 174, 0, 0, -399, 4969, + -2321, -7744, 6494, -3776, 1478, 758, -1794, -2233, + -4059, 4932, 2770, 4761, -3475, 1243, 829, -651, + -5358, -436, 2381, 1360, 2561, -3118, 858, -4366, + 3933, 3646, -43, -1310, -16, 924, 1197, 1415, + -5036, -376, 0, 0, 100, 1410, 1290, 3199, + 7091, -3638, -2641, 1118, 45, -441, 794, -974, + -5033, 889, 438, -3102, 895, 3555, 4672, 4795, + 1129, -2408, -2153, 1742, 159, -2040, 7578, -2006, + -5737, 1986, -5568, -6413, 2428, -1387, -2441, 667, + 0, 0, -37, -6031, -4434, -904, 3290, 1806, + 4736, 2516, -5905, -5927, 1754, -4300, -2468, -2203, + -4836, -672, 1444, -1591, -1631, -1789, 4311, -153, + -688, -1222, 1058, 3139, 4659, -353, 1543, 1838, + 2180, -1448, 2432, 6277, 5304, -1692, 0, 0, + -280, 4506, 807, -477, 823, 3550, 1427, -1856, + -3003, -3501, -1203, 2679, 933, 778, -4954, -1977, + -7458, 4687, 435, 7045, -4053, -3130, 257, -3917, + -6165, 1889, 927, 235, 1889, -1097, 1985, 630, + -2172, -2130, 7080, 4810, 0, 0, -300, 496, + 2808, 279, 667, -7179, -2661, -526, -2832, 1751, + 2849, 4829, -906, -4151, -1124, -3062, 8166, 5361, + -1656, -6017, 3265, 2551, -864, -432, -6966, 6295, + -168, 901, 442, -582, 269, 236, -3574, 799, + 472, 565, 0, 0, 805, -2466, 6208, -4592, + -170, -6701, -5610, 3678, -4242, 4561, -724, -5534, + 2415, 7354, 2761, 2699, -349, 3822, -2372, 1756, + -5523, -3445, -588, -5749, -3986, 9804, -3871, 5375, + -2308, 5504, -2766, -1651, 1472, 6832, 2705, -5104, + 0, 0, -700, -1179, 4402, 400, 1383, 939, + -1342, 6013, 2577, -3472, 472, 2883, 1450, -3917, + 2849, 5084, 4990, 5392, 342, -4925, -3329, -5372, + -2674, -6035, -5072, -836, 179, 2506, 7987, -3647, + -8202, -1437, 1891, 2400, 1607, -3611, 0, 0, + -4706, -4003, 9928, -379, 5557, 3738, -8789, 685, + 1937, -5157, 13388, 7995, -4119, -9909, -5079, 4804, + 5586, 774, -5430, 299, -9943, 3264, -3690, -3901, + -1133, -6199, 3182, 1544, 5467, 3686, -2639, 4068, + 1163, -185, -1299, -506, 0, 0, 843, 1005, + -1059, 467, -1279, -2259, 6057, -1694, -5885, 5342, + -5160, -3748, -1382, 4420, -697, -2000, -3808, 3100, + 2685, -4073, 531, 318, -7822, 2414, 2901, 3399, + -1340, 8449, 3685, 463, -3341, 2423, 2304, -2723, + 84, -2622, 0, 0, 12088, -265, 2562, -435, + -4348, -2426, 3538, 1552, 1279, 883, -4166, 2634, + -6130, 2994, 3729, -1570, -601, -1753, -5124, -2788, + -2096, -1920, -2649, 2793, -1079, -1952, 2983, -1530, + 2499, 1769, 1492, -6757, -2108, 2841, 1466, 2597, + 0, 0, -3830, -4093, 2448, 12720, 7737, -665, + -832, -9257, 2971, -2400, 791, 1873, 1072, -587, + -7440, 8055, 1531, -4736, 616, -1782, -2982, 9663, + -5057, -5926, 1610, -4489, 7033, -8658, 6010, -5673, + 5648, 812, -271, -1802, -4500, 4392, 0, 0, + -888, -327, 3373, -1084, 7959, 2430, 1898, -2360, + -1820, -1377, -1090, -4436, -3422, -1106, -3230, 3876, + -41, -5128, 6375, -1848, -3824, 5844, 617, -1957, + 4232, 1345, -1439, -83, 3046, -214, 5458, -5566, + -4387, -3738, -5740, 8657, 0, 0, 6978, 6239, + -3686, -981, -2854, 78, 5859, -357, 4618, 7391, + -138, 971, -5799, 2135, 4478, -7004, -5949, 1668, + -6933, -1163, 7010, -5624, 2990, 6192, -8075, 3567, + -8308, 2236, -5098, -2120, -4355, -4238, 4955, 10230, + 692, -5606, 0, 0, -1348, -7069, -12, -4927, + 1211, 651, 1360, 7744, 3404, 5069, -2438, -105, + 2332, 1494, -4686, 1336, -3628, -881, 2474, 1736, + -26, -257, 2135, -4452, 446, -641, -4704, 2605, + -6436, 6662, -4939, 990, -1100, -3782, 5028, 4753, + 0, 0, -2875, 6410, 3518, 3950, 1271, 869, + -2842, -5837, 1532, -2899, 1140, -597, 1712, -1988, + -4819, -4783, 4773, -8796, 2240, -4596, 3565, -4853, + -556, -3974, 7366, -4370, 3113, -3548, 3552, -5450, + 3869, 2514, 6736, -4570, 6074, 3151, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; diff --git a/code/opus-1.5.2/src/mapping_matrix.h b/code/opus-1.5.2/src/mapping_matrix.h new file mode 100644 index 00000000..53646cb1 --- /dev/null +++ b/code/opus-1.5.2/src/mapping_matrix.h @@ -0,0 +1,145 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file mapping_matrix.h + * @brief Opus reference implementation mapping matrix API + */ + +#ifndef MAPPING_MATRIX_H +#define MAPPING_MATRIX_H + +#include "opus_types.h" +#include "opus_projection.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct MappingMatrix +{ + int rows; /* number of channels outputted from matrix. */ + int cols; /* number of channels inputted to matrix. */ + int gain; /* in dB. S7.8-format. */ + /* Matrix cell data goes here using col-wise ordering. */ +} MappingMatrix; + +opus_int32 mapping_matrix_get_size(int rows, int cols); + +opus_int16 *mapping_matrix_get_data(const MappingMatrix *matrix); + +void mapping_matrix_init( + MappingMatrix * const matrix, + int rows, + int cols, + int gain, + const opus_int16 *data, + opus_int32 data_size +); + +#ifndef DISABLE_FLOAT_API +void mapping_matrix_multiply_channel_in_float( + const MappingMatrix *matrix, + const float *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size +); + +void mapping_matrix_multiply_channel_out_float( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + float *output, + int output_rows, + int frame_size +); +#endif /* DISABLE_FLOAT_API */ + +void mapping_matrix_multiply_channel_in_short( + const MappingMatrix *matrix, + const opus_int16 *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size +); + +void mapping_matrix_multiply_channel_out_short( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + opus_int16 *output, + int output_rows, + int frame_size +); + +/* Pre-computed mixing and demixing matrices for 1st to 3rd-order ambisonics. + * foa: first-order ambisonics + * soa: second-order ambisonics + * toa: third-order ambisonics + */ +extern const MappingMatrix mapping_matrix_foa_mixing; +extern const opus_int16 mapping_matrix_foa_mixing_data[36]; + +extern const MappingMatrix mapping_matrix_soa_mixing; +extern const opus_int16 mapping_matrix_soa_mixing_data[121]; + +extern const MappingMatrix mapping_matrix_toa_mixing; +extern const opus_int16 mapping_matrix_toa_mixing_data[324]; + +extern const MappingMatrix mapping_matrix_fourthoa_mixing; +extern const opus_int16 mapping_matrix_fourthoa_mixing_data[729]; + +extern const MappingMatrix mapping_matrix_fifthoa_mixing; +extern const opus_int16 mapping_matrix_fifthoa_mixing_data[1444]; + +extern const MappingMatrix mapping_matrix_foa_demixing; +extern const opus_int16 mapping_matrix_foa_demixing_data[36]; + +extern const MappingMatrix mapping_matrix_soa_demixing; +extern const opus_int16 mapping_matrix_soa_demixing_data[121]; + +extern const MappingMatrix mapping_matrix_toa_demixing; +extern const opus_int16 mapping_matrix_toa_demixing_data[324]; + +extern const MappingMatrix mapping_matrix_fourthoa_demixing; +extern const opus_int16 mapping_matrix_fourthoa_demixing_data[729]; + +extern const MappingMatrix mapping_matrix_fifthoa_demixing; +extern const opus_int16 mapping_matrix_fifthoa_demixing_data[1444]; + +#ifdef __cplusplus +} +#endif + +#endif /* MAPPING_MATRIX_H */ diff --git a/code/opus-1.5.2/src/mlp.c b/code/opus-1.5.2/src/mlp.c new file mode 100644 index 00000000..e658ccde --- /dev/null +++ b/code/opus-1.5.2/src/mlp.c @@ -0,0 +1,132 @@ +/* Copyright (c) 2008-2011 Octasic Inc. + 2012-2017 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include "opus_types.h" +#include "opus_defines.h" +#include "arch.h" +#include "mlp.h" + +#define fmadd(a, b, c) ((a)*(b)+(c)) +static OPUS_INLINE float tansig_approx(float x) +{ + const float N0 = 952.52801514f; + const float N1 = 96.39235687f; + const float N2 = 0.60863042f; + const float D0 = 952.72399902f; + const float D1 = 413.36801147f; + const float D2 = 11.88600922f; + float X2, num, den; + X2 = x*x; + num = fmadd(fmadd(N2, X2, N1), X2, N0); + den = fmadd(fmadd(D2, X2, D1), X2, D0); + num = num*x/den; + return MAX32(-1.f, MIN32(1.f, num)); +} + +static OPUS_INLINE float sigmoid_approx(float x) +{ + return .5f + .5f*tansig_approx(.5f*x); +} + +static void gemm_accum(float *out, const opus_int8 *weights, int rows, int cols, int col_stride, const float *x) +{ + int i, j; + for (i=0;inb_inputs; + N = layer->nb_neurons; + stride = N; + for (i=0;ibias[i]; + gemm_accum(output, layer->input_weights, N, M, stride, input); + for (i=0;isigmoid) { + for (i=0;inb_inputs; + N = gru->nb_neurons; + stride = 3*N; + /* Compute update gate. */ + for (i=0;ibias[i]; + gemm_accum(z, gru->input_weights, N, M, stride, input); + gemm_accum(z, gru->recurrent_weights, N, N, stride, state); + for (i=0;ibias[N + i]; + gemm_accum(r, &gru->input_weights[N], N, M, stride, input); + gemm_accum(r, &gru->recurrent_weights[N], N, N, stride, state); + for (i=0;ibias[2*N + i]; + for (i=0;iinput_weights[2*N], N, M, stride, input); + gemm_accum(h, &gru->recurrent_weights[2*N], N, N, stride, tmp); + for (i=0;i 5760) + if (count <= 0 || framesize*(opus_int32)count > 5760) return OPUS_INVALID_PACKET; len--; /* Padding flag is bit 6 */ @@ -337,6 +338,11 @@ int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, data += size[i]; } + if (padding != NULL) + { + *padding = data; + *padding_len = pad; + } if (packet_offset) *packet_offset = pad+(opus_int32)(data-data0); @@ -351,6 +357,6 @@ int opus_packet_parse(const unsigned char *data, opus_int32 len, opus_int16 size[48], int *payload_offset) { return opus_packet_parse_impl(data, len, 0, out_toc, - frames, size, payload_offset, NULL); + frames, size, payload_offset, NULL, NULL, NULL); } diff --git a/code/opus-1.5.2/src/opus_compare.c b/code/opus-1.5.2/src/opus_compare.c new file mode 100644 index 00000000..1956e08f --- /dev/null +++ b/code/opus-1.5.2/src/opus_compare.c @@ -0,0 +1,382 @@ +/* Copyright (c) 2011-2012 Xiph.Org Foundation, Mozilla Corporation + Written by Jean-Marc Valin and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include +#include +#include +#include + +#define OPUS_PI (3.14159265F) + +#define OPUS_COSF(_x) ((float)cos(_x)) +#define OPUS_SINF(_x) ((float)sin(_x)) + +static void *check_alloc(void *_ptr){ + if(_ptr==NULL){ + fprintf(stderr,"Out of memory.\n"); + exit(EXIT_FAILURE); + } + return _ptr; +} + +static void *opus_malloc(size_t _size){ + return check_alloc(malloc(_size)); +} + +static void *opus_realloc(void *_ptr,size_t _size){ + return check_alloc(realloc(_ptr,_size)); +} + +static size_t read_pcm16(float **_samples,FILE *_fin,int _nchannels){ + unsigned char buf[1024]; + float *samples; + size_t nsamples; + size_t csamples; + size_t xi; + size_t nread; + samples=NULL; + nsamples=csamples=0; + for(;;){ + nread=fread(buf,2*_nchannels,1024/(2*_nchannels),_fin); + if(nread<=0)break; + if(nsamples+nread>csamples){ + do csamples=csamples<<1|1; + while(nsamples+nread>csamples); + samples=(float *)opus_realloc(samples, + _nchannels*csamples*sizeof(*samples)); + } + for(xi=0;xi=_window_sz)ti-=_window_sz; + } + re*=_downsample; + im*=_downsample; + _ps[(xi*ps_sz+xj)*_nchannels+ci]=re*re+im*im+100000; + p[ci]+=_ps[(xi*ps_sz+xj)*_nchannels+ci]; + } + } + if(_out){ + _out[(xi*_nbands+bi)*_nchannels]=p[0]/(_bands[bi+1]-_bands[bi]); + if(_nchannels==2){ + _out[(xi*_nbands+bi)*_nchannels+1]=p[1]/(_bands[bi+1]-_bands[bi]); + } + } + } + } + free(window); +} + +#define NBANDS (21) +#define NFREQS (240) + +/*Bands on which we compute the pseudo-NMR (Bark-derived + CELT bands).*/ +static const int BANDS[NBANDS+1]={ + 0,2,4,6,8,10,12,14,16,20,24,28,32,40,48,56,68,80,96,120,156,200 +}; + +#define TEST_WIN_SIZE (480) +#define TEST_WIN_STEP (120) + +int main(int _argc,const char **_argv){ + FILE *fin1; + FILE *fin2; + float *x; + float *y; + float *xb; + float *X; + float *Y; + double err; + float Q; + size_t xlength; + size_t ylength; + size_t nframes; + size_t xi; + int ci; + int xj; + int bi; + int nchannels; + unsigned rate; + int downsample; + int ybands; + int yfreqs; + int max_compare; + if(_argc<3||_argc>6){ + fprintf(stderr,"Usage: %s [-s] [-r rate2] \n", + _argv[0]); + return EXIT_FAILURE; + } + nchannels=1; + if(strcmp(_argv[1],"-s")==0){ + nchannels=2; + _argv++; + } + rate=48000; + ybands=NBANDS; + yfreqs=NFREQS; + downsample=1; + if(strcmp(_argv[1],"-r")==0){ + rate=atoi(_argv[2]); + if(rate!=8000&&rate!=12000&&rate!=16000&&rate!=24000&&rate!=48000){ + fprintf(stderr, + "Sampling rate must be 8000, 12000, 16000, 24000, or 48000\n"); + return EXIT_FAILURE; + } + downsample=48000/rate; + switch(rate){ + case 8000:ybands=13;break; + case 12000:ybands=15;break; + case 16000:ybands=17;break; + case 24000:ybands=19;break; + } + yfreqs=NFREQS/downsample; + _argv+=2; + } + fin1=fopen(_argv[1],"rb"); + if(fin1==NULL){ + fprintf(stderr,"Error opening '%s'.\n",_argv[1]); + return EXIT_FAILURE; + } + fin2=fopen(_argv[2],"rb"); + if(fin2==NULL){ + fprintf(stderr,"Error opening '%s'.\n",_argv[2]); + fclose(fin1); + return EXIT_FAILURE; + } + /*Read in the data and allocate scratch space.*/ + xlength=read_pcm16(&x,fin1,2); + if(nchannels==1){ + for(xi=0;xi0;){ + for(ci=0;ci0){ + /*Temporal masking: -3 dB/2.5ms slope.*/ + for(bi=0;bi=79&&xj<=81)im*=0.1F; + if(xj==80)im*=0.1F; + Eb+=im; + } + } + Eb /= (BANDS[bi+1]-BANDS[bi])*nchannels; + Ef += Eb*Eb; + } + /*Using a fixed normalization value means we're willing to accept slightly + lower quality for lower sampling rates.*/ + Ef/=NBANDS; + Ef*=Ef; + err+=Ef*Ef; + } + free(xb); + free(X); + free(Y); + err=pow(err/nframes,1.0/16); + Q=100*(1-0.5*log(1+err)/log(1.13)); + if(Q<0){ + fprintf(stderr,"Test vector FAILS\n"); + fprintf(stderr,"Internal weighted error is %f\n",err); + return EXIT_FAILURE; + } + else{ + fprintf(stderr,"Test vector PASSES\n"); + fprintf(stderr, + "Opus quality metric: %.1f %% (internal weighted error is %f)\n",Q,err); + return EXIT_SUCCESS; + } +} diff --git a/code/opus-1.2.1/src/opus_decoder.c b/code/opus-1.5.2/src/opus_decoder.c similarity index 62% rename from code/opus-1.2.1/src/opus_decoder.c rename to code/opus-1.5.2/src/opus_decoder.c index 149ae7f2..b57c8094 100644 --- a/code/opus-1.2.1/src/opus_decoder.c +++ b/code/opus-1.5.2/src/opus_decoder.c @@ -52,6 +52,15 @@ #include "mathops.h" #include "cpu_support.h" +#ifdef ENABLE_DEEP_PLC +#include "dred_rdovae_dec_data.h" +#include "dred_rdovae_dec.h" +#endif + +#ifdef ENABLE_OSCE +#include "osce.h" +#endif + struct OpusDecoder { int celt_dec_offset; int silk_dec_offset; @@ -59,7 +68,11 @@ struct OpusDecoder { opus_int32 Fs; /** Sampling rate (at the API level) */ silk_DecControlStruct DecControl; int decode_gain; + int complexity; int arch; +#ifdef ENABLE_DEEP_PLC + LPCNetPLCState lpcnet; +#endif /* Everything beyond this point gets cleared on a reset */ #define OPUS_DECODER_RESET_START stream_channels @@ -78,6 +91,26 @@ struct OpusDecoder { opus_uint32 rangeFinal; }; +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +static void validate_opus_decoder(OpusDecoder *st) +{ + celt_assert(st->channels == 1 || st->channels == 2); + celt_assert(st->Fs == 48000 || st->Fs == 24000 || st->Fs == 16000 || st->Fs == 12000 || st->Fs == 8000); + celt_assert(st->DecControl.API_sampleRate == st->Fs); + celt_assert(st->DecControl.internalSampleRate == 0 || st->DecControl.internalSampleRate == 16000 || st->DecControl.internalSampleRate == 12000 || st->DecControl.internalSampleRate == 8000); + celt_assert(st->DecControl.nChannelsAPI == st->channels); + celt_assert(st->DecControl.nChannelsInternal == 0 || st->DecControl.nChannelsInternal == 1 || st->DecControl.nChannelsInternal == 2); + celt_assert(st->DecControl.payloadSize_ms == 0 || st->DecControl.payloadSize_ms == 10 || st->DecControl.payloadSize_ms == 20 || st->DecControl.payloadSize_ms == 40 || st->DecControl.payloadSize_ms == 60); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif + celt_assert(st->stream_channels == 1 || st->stream_channels == 2); +} +#define VALIDATE_OPUS_DECODER(st) validate_opus_decoder(st) +#else +#define VALIDATE_OPUS_DECODER(st) +#endif int opus_decoder_get_size(int channels) { @@ -104,7 +137,7 @@ int opus_decoder_init(OpusDecoder *st, opus_int32 Fs, int channels) return OPUS_BAD_ARG; OPUS_CLEAR((char*)st, opus_decoder_get_size(channels)); - /* Initialize SILK encoder */ + /* Initialize SILK decoder */ ret = silk_Get_Decoder_Size(&silkDecSizeBytes); if (ret) return OPUS_INTERNAL_ERROR; @@ -115,6 +148,7 @@ int opus_decoder_init(OpusDecoder *st, opus_int32 Fs, int channels) silk_dec = (char*)st+st->silk_dec_offset; celt_dec = (CELTDecoder*)((char*)st+st->celt_dec_offset); st->stream_channels = st->channels = channels; + st->complexity = 0; st->Fs = Fs; st->DecControl.API_sampleRate = st->Fs; @@ -132,6 +166,9 @@ int opus_decoder_init(OpusDecoder *st, opus_int32 Fs, int channels) st->prev_mode = 0; st->frame_size = Fs/400; +#ifdef ENABLE_DEEP_PLC + lpcnet_plc_init( &st->lpcnet); +#endif st->arch = opus_select_arch(); return OPUS_OK; } @@ -217,6 +254,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, int audiosize; int mode; + int bandwidth; int transition=0; int start_band; int redundancy=0; @@ -253,10 +291,13 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, { audiosize = st->frame_size; mode = st->mode; + bandwidth = st->bandwidth; ec_dec_init(&dec,(unsigned char*)data,len); } else { audiosize = frame_size; - mode = st->prev_mode; + /* Run PLC using last used mode (CELT if we ended with CELT redundancy) */ + mode = st->prev_redundancy ? MODE_CELT_ONLY : st->prev_mode; + bandwidth = 0; if (mode == 0) { @@ -346,7 +387,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, pcm_ptr = pcm_silk; if (st->prev_mode==MODE_CELT_ONLY) - silk_InitDecoder( silk_dec ); + silk_ResetDecoder( silk_dec ); /* The SILK PLC cannot produce frames of less than 10 ms */ st->DecControl.payloadSize_ms = IMAX(10, 1000 * audiosize / st->Fs); @@ -355,29 +396,43 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, { st->DecControl.nChannelsInternal = st->stream_channels; if( mode == MODE_SILK_ONLY ) { - if( st->bandwidth == OPUS_BANDWIDTH_NARROWBAND ) { + if( bandwidth == OPUS_BANDWIDTH_NARROWBAND ) { st->DecControl.internalSampleRate = 8000; - } else if( st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND ) { + } else if( bandwidth == OPUS_BANDWIDTH_MEDIUMBAND ) { st->DecControl.internalSampleRate = 12000; - } else if( st->bandwidth == OPUS_BANDWIDTH_WIDEBAND ) { + } else if( bandwidth == OPUS_BANDWIDTH_WIDEBAND ) { st->DecControl.internalSampleRate = 16000; } else { st->DecControl.internalSampleRate = 16000; - silk_assert( 0 ); + celt_assert( 0 ); } } else { /* Hybrid mode */ st->DecControl.internalSampleRate = 16000; } } + st->DecControl.enable_deep_plc = st->complexity >= 5; +#ifdef ENABLE_OSCE + st->DecControl.osce_method = OSCE_METHOD_NONE; +#ifndef DISABLE_LACE + if (st->complexity >= 6) {st->DecControl.osce_method = OSCE_METHOD_LACE;} +#endif +#ifndef DISABLE_NOLACE + if (st->complexity >= 7) {st->DecControl.osce_method = OSCE_METHOD_NOLACE;} +#endif +#endif - lost_flag = data == NULL ? 1 : 2 * decode_fec; + lost_flag = data == NULL ? 1 : 2 * !!decode_fec; decoded_samples = 0; do { /* Call SILK decoder */ int first_frame = decoded_samples == 0; silk_ret = silk_Decode( silk_dec, &st->DecControl, - lost_flag, first_frame, &dec, pcm_ptr, &silk_frame_size, st->arch ); + lost_flag, first_frame, &dec, pcm_ptr, &silk_frame_size, +#ifdef ENABLE_DEEP_PLC + &st->lpcnet, +#endif + st->arch ); if( silk_ret ) { if (lost_flag) { /* PLC failure should not be fatal */ @@ -396,7 +451,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, start_band = 0; if (!decode_fec && mode != MODE_CELT_ONLY && data != NULL - && ec_tell(&dec)+17+20*(st->mode == MODE_HYBRID) <= 8*len) + && ec_tell(&dec)+17+20*(mode == MODE_HYBRID) <= 8*len) { /* Check if we have a redundant 0-8 kHz band */ if (mode == MODE_HYBRID) @@ -427,29 +482,6 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, if (mode != MODE_CELT_ONLY) start_band = 17; - { - int endband=21; - - switch(st->bandwidth) - { - case OPUS_BANDWIDTH_NARROWBAND: - endband = 13; - break; - case OPUS_BANDWIDTH_MEDIUMBAND: - case OPUS_BANDWIDTH_WIDEBAND: - endband = 17; - break; - case OPUS_BANDWIDTH_SUPERWIDEBAND: - endband = 19; - break; - case OPUS_BANDWIDTH_FULLBAND: - endband = 21; - break; - } - celt_decoder_ctl(celt_dec, CELT_SET_END_BAND(endband)); - celt_decoder_ctl(celt_dec, CELT_SET_CHANNELS(st->stream_channels)); - } - if (redundancy) { transition = 0; @@ -464,6 +496,34 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); } + + if (bandwidth) + { + int endband=21; + + switch(bandwidth) + { + case OPUS_BANDWIDTH_NARROWBAND: + endband = 13; + break; + case OPUS_BANDWIDTH_MEDIUMBAND: + case OPUS_BANDWIDTH_WIDEBAND: + endband = 17; + break; + case OPUS_BANDWIDTH_SUPERWIDEBAND: + endband = 19; + break; + case OPUS_BANDWIDTH_FULLBAND: + endband = 21; + break; + default: + celt_assert(0); + break; + } + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_END_BAND(endband))); + } + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_CHANNELS(st->stream_channels))); + /* Only allocation memory for redundancy if/when needed */ redundant_audio_size = redundancy ? F5*st->channels : ALLOC_NONE; ALLOC(redundant_audio, redundant_audio_size, opus_val16); @@ -471,24 +531,33 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, /* 5 ms redundant frame for CELT->SILK*/ if (redundancy && celt_to_silk) { - celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0)); + /* If the previous frame did not use CELT (the first redundancy frame in + a transition from SILK may have been lost) then the CELT decoder is + stale at this point and the redundancy audio is not useful, however + the final range is still needed (for testing), so the redundancy is + always decoded but the decoded audio may not be used */ + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); celt_decode_with_ec(celt_dec, data+len, redundancy_bytes, redundant_audio, F5, NULL, 0); - celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng))); } /* MUST be after PLC */ - celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(start_band)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(start_band))); if (mode != MODE_SILK_ONLY) { int celt_frame_size = IMIN(F20, frame_size); /* Make sure to discard any previous CELT state */ if (mode != st->prev_mode && st->prev_mode > 0 && !st->prev_redundancy) - celt_decoder_ctl(celt_dec, OPUS_RESET_STATE); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_RESET_STATE)); /* Decode CELT */ - celt_ret = celt_decode_with_ec(celt_dec, decode_fec ? NULL : data, - len, pcm, celt_frame_size, &dec, celt_accum); + celt_ret = celt_decode_with_ec_dred(celt_dec, decode_fec ? NULL : data, + len, pcm, celt_frame_size, &dec, celt_accum +#ifdef ENABLE_DEEP_PLC + , &st->lpcnet +#endif + ); } else { unsigned char silence[2] = {0xFF, 0xFF}; if (!celt_accum) @@ -500,7 +569,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, do a fade-out by decoding a silence frame */ if (st->prev_mode == MODE_HYBRID && !(redundancy && celt_to_silk && st->prev_redundancy) ) { - celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); celt_decode_with_ec(celt_dec, silence, 2, pcm, F2_5, NULL, celt_accum); } } @@ -518,22 +587,25 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, { const CELTMode *celt_mode; - celt_decoder_ctl(celt_dec, CELT_GET_MODE(&celt_mode)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_GET_MODE(&celt_mode))); window = celt_mode->window; } /* 5 ms redundant frame for SILK->CELT */ if (redundancy && !celt_to_silk) { - celt_decoder_ctl(celt_dec, OPUS_RESET_STATE); - celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_RESET_STATE)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); celt_decode_with_ec(celt_dec, data+len, redundancy_bytes, redundant_audio, F5, NULL, 0); - celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng))); smooth_fade(pcm+st->channels*(frame_size-F2_5), redundant_audio+st->channels*F2_5, pcm+st->channels*(frame_size-F2_5), F2_5, st->channels, window, st->Fs); } - if (redundancy && celt_to_silk) + /* 5ms redundant frame for CELT->SILK; ignore if the previous frame did not + use CELT (the first redundancy frame in a transition from SILK may have + been lost) */ + if (redundancy && celt_to_silk && (st->prev_mode != MODE_SILK_ONLY || st->prev_redundancy)) { for (c=0;cchannels;c++) { @@ -597,7 +669,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, int opus_decode_native(OpusDecoder *st, const unsigned char *data, opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec, - int self_delimited, opus_int32 *packet_offset, int soft_clip) + int self_delimited, opus_int32 *packet_offset, int soft_clip, const OpusDRED *dred, opus_int32 dred_offset) { int i, nb_samples; int count, offset; @@ -605,11 +677,41 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, int packet_frame_size, packet_bandwidth, packet_mode, packet_stream_channels; /* 48 x 2.5 ms = 120 ms */ opus_int16 size[48]; + VALIDATE_OPUS_DECODER(st); if (decode_fec<0 || decode_fec>1) return OPUS_BAD_ARG; /* For FEC/PLC, frame_size has to be to have a multiple of 2.5 ms */ if ((decode_fec || len==0 || data==NULL) && frame_size%(st->Fs/400)!=0) return OPUS_BAD_ARG; +#ifdef ENABLE_DRED + if (dred != NULL && dred->process_stage == 2) { + int F10; + int features_per_frame; + int needed_feature_frames; + int init_frames; + lpcnet_plc_fec_clear(&st->lpcnet); + F10 = st->Fs/100; + /* if blend==0, the last PLC call was "update" and we need to feed two extra 10-ms frames. */ + init_frames = (st->lpcnet.blend == 0) ? 2 : 0; + features_per_frame = IMAX(1, frame_size/F10); + needed_feature_frames = init_frames + features_per_frame; + lpcnet_plc_fec_clear(&st->lpcnet); + for (i=0;idred_offset*F10/4)/F10); + if (feature_offset <= 4*dred->nb_latents-1 && feature_offset >= 0) { + lpcnet_plc_fec_add(&st->lpcnet, dred->fec_features+feature_offset*DRED_NUM_FEATURES); + } else { + if (feature_offset >= 0) lpcnet_plc_fec_add(&st->lpcnet, NULL); + } + + } + } +#else + (void)dred; + (void)dred_offset; +#endif if (len==0 || data==NULL) { int pcm_count=0; @@ -634,7 +736,7 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, packet_stream_channels = opus_packet_get_nb_channels(data); count = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, - size, &offset, packet_offset); + size, &offset, packet_offset, NULL, NULL); if (count<0) return count; @@ -646,12 +748,12 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, int ret; /* If no FEC can be present, run the PLC (recursive call) */ if (frame_size < packet_frame_size || packet_mode == MODE_CELT_ONLY || st->mode == MODE_CELT_ONLY) - return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL, soft_clip); + return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL, soft_clip, NULL, 0); /* Otherwise, run the PLC on everything except the size for which we might have FEC */ duration_copy = st->last_packet_duration; if (frame_size-packet_frame_size!=0) { - ret = opus_decode_native(st, NULL, 0, pcm, frame_size-packet_frame_size, 0, 0, NULL, soft_clip); + ret = opus_decode_native(st, NULL, 0, pcm, frame_size-packet_frame_size, 0, 0, NULL, soft_clip, NULL, 0); if (ret<0) { st->last_packet_duration = duration_copy; @@ -715,7 +817,7 @@ int opus_decode(OpusDecoder *st, const unsigned char *data, { if(frame_size<=0) return OPUS_BAD_ARG; - return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0, NULL, 0); } #ifndef DISABLE_FLOAT_API @@ -740,9 +842,10 @@ int opus_decode_float(OpusDecoder *st, const unsigned char *data, else return OPUS_INVALID_PACKET; } + celt_assert(st->channels == 1 || st->channels == 2); ALLOC(out, frame_size*st->channels, opus_int16); - ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 0); + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 0, NULL, 0); if (ret > 0) { for (i=0;ichannels;i++) @@ -777,9 +880,10 @@ int opus_decode(OpusDecoder *st, const unsigned char *data, else return OPUS_INVALID_PACKET; } + celt_assert(st->channels == 1 || st->channels == 2); ALLOC(out, frame_size*st->channels, float); - ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 1); + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 1, NULL, 0); if (ret > 0) { for (i=0;ichannels;i++) @@ -794,7 +898,7 @@ int opus_decode_float(OpusDecoder *st, const unsigned char *data, { if(frame_size<=0) return OPUS_BAD_ARG; - return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0, NULL, 0); } #endif @@ -824,6 +928,27 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) *value = st->bandwidth; } break; + case OPUS_SET_COMPLEXITY_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>10) + { + goto bad_arg; + } + st->complexity = value; + celt_decoder_ctl(celt_dec, OPUS_SET_COMPLEXITY(value)); + } + break; + case OPUS_GET_COMPLEXITY_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->complexity; + } + break; case OPUS_GET_FINAL_RANGE_REQUEST: { opus_uint32 *value = va_arg(ap, opus_uint32*); @@ -841,9 +966,12 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) ((char*)&st->OPUS_DECODER_RESET_START - (char*)st)); celt_decoder_ctl(celt_dec, OPUS_RESET_STATE); - silk_InitDecoder( silk_dec ); + silk_ResetDecoder( silk_dec ); st->stream_channels = st->channels; st->frame_size = st->Fs/400; +#ifdef ENABLE_DEEP_PLC + lpcnet_plc_reset( &st->lpcnet ); +#endif } break; case OPUS_GET_SAMPLE_RATE_REQUEST: @@ -864,7 +992,7 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) goto bad_arg; } if (st->prev_mode == MODE_CELT_ONLY) - celt_decoder_ctl(celt_dec, OPUS_GET_PITCH(value)); + ret = celt_decoder_ctl(celt_dec, OPUS_GET_PITCH(value)); else *value = st->DecControl.prevPitchLag; } @@ -906,7 +1034,7 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) { goto bad_arg; } - celt_decoder_ctl(celt_dec, OPUS_SET_PHASE_INVERSION_DISABLED(value)); + ret = celt_decoder_ctl(celt_dec, OPUS_SET_PHASE_INVERSION_DISABLED(value)); } break; case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: @@ -916,9 +1044,23 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) { goto bad_arg; } - celt_decoder_ctl(celt_dec, OPUS_GET_PHASE_INVERSION_DISABLED(value)); + ret = celt_decoder_ctl(celt_dec, OPUS_GET_PHASE_INVERSION_DISABLED(value)); } break; +#ifdef USE_WEIGHTS_FILE + case OPUS_SET_DNN_BLOB_REQUEST: + { + const unsigned char *data = va_arg(ap, const unsigned char *); + opus_int32 len = va_arg(ap, opus_int32); + if(len<0 || data == NULL) + { + goto bad_arg; + } + ret = lpcnet_plc_load_model(&st->lpcnet, data, len); + ret = silk_LoadOSCEModels(silk_dec, data, len) || ret; + } + break; +#endif default: /*fprintf(stderr, "unknown opus_decoder_ctl() request: %d", request);*/ ret = OPUS_UNIMPLEMENTED; @@ -994,8 +1136,373 @@ int opus_packet_get_nb_samples(const unsigned char packet[], opus_int32 len, return samples; } +int opus_packet_has_lbrr(const unsigned char packet[], opus_int32 len) +{ + int ret; + const unsigned char *frames[48]; + opus_int16 size[48]; + int packet_mode, packet_frame_size, packet_stream_channels; + int nb_frames=1; + int lbrr; + + packet_mode = opus_packet_get_mode(packet); + if (packet_mode == MODE_CELT_ONLY) + return 0; + packet_frame_size = opus_packet_get_samples_per_frame(packet, 48000); + if (packet_frame_size > 960) + nb_frames = packet_frame_size/960; + packet_stream_channels = opus_packet_get_nb_channels(packet); + ret = opus_packet_parse(packet, len, NULL, frames, size, NULL); + if (ret <= 0) + return ret; + lbrr = (frames[0][0] >> (7-nb_frames)) & 0x1; + if (packet_stream_channels == 2) + lbrr = lbrr || ((frames[0][0] >> (6-2*nb_frames)) & 0x1); + return lbrr; +} + int opus_decoder_get_nb_samples(const OpusDecoder *dec, const unsigned char packet[], opus_int32 len) { return opus_packet_get_nb_samples(packet, len, dec->Fs); } + +struct OpusDREDDecoder { +#ifdef ENABLE_DRED + RDOVAEDec model; +#endif + int loaded; + int arch; + opus_uint32 magic; +}; + +#if defined(ENABLE_DRED) && (defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS)) +static void validate_dred_decoder(OpusDREDDecoder *st) +{ + celt_assert(st->magic == 0xD8EDDEC0); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif +} +#define VALIDATE_DRED_DECODER(st) validate_dred_decoder(st) +#else +#define VALIDATE_DRED_DECODER(st) +#endif + + +int opus_dred_decoder_get_size(void) +{ + return sizeof(OpusDREDDecoder); +} + +#ifdef ENABLE_DRED +int dred_decoder_load_model(OpusDREDDecoder *dec, const unsigned char *data, int len) +{ + WeightArray *list; + int ret; + parse_weights(&list, data, len); + ret = init_rdovaedec(&dec->model, list); + opus_free(list); + if (ret == 0) dec->loaded = 1; + return (ret == 0) ? OPUS_OK : OPUS_BAD_ARG; +} +#endif + +int opus_dred_decoder_init(OpusDREDDecoder *dec) +{ + int ret = 0; + dec->loaded = 0; +#if defined(ENABLE_DRED) && !defined(USE_WEIGHTS_FILE) + ret = init_rdovaedec(&dec->model, rdovaedec_arrays); + if (ret == 0) dec->loaded = 1; +#endif + dec->arch = opus_select_arch(); + /* To make sure nobody forgets to init, use a magic number. */ + dec->magic = 0xD8EDDEC0; + return (ret == 0) ? OPUS_OK : OPUS_UNIMPLEMENTED; +} + +OpusDREDDecoder *opus_dred_decoder_create(int *error) +{ + int ret; + OpusDREDDecoder *dec; + dec = (OpusDREDDecoder *)opus_alloc(opus_dred_decoder_get_size()); + if (dec == NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_dred_decoder_init(dec); + if (error) + *error = ret; + if (ret != OPUS_OK) + { + opus_free(dec); + dec = NULL; + } + return dec; +} + +void opus_dred_decoder_destroy(OpusDREDDecoder *dec) +{ + if (dec) dec->magic = 0xDE57801D; + opus_free(dec); +} + +int opus_dred_decoder_ctl(OpusDREDDecoder *dred_dec, int request, ...) +{ +#ifdef ENABLE_DRED + int ret = OPUS_OK; + va_list ap; + + va_start(ap, request); + (void)dred_dec; + switch (request) + { +# ifdef USE_WEIGHTS_FILE + case OPUS_SET_DNN_BLOB_REQUEST: + { + const unsigned char *data = va_arg(ap, const unsigned char *); + opus_int32 len = va_arg(ap, opus_int32); + if(len<0 || data == NULL) + { + goto bad_arg; + } + return dred_decoder_load_model(dred_dec, data, len); + } + break; +# endif + default: + /*fprintf(stderr, "unknown opus_decoder_ctl() request: %d", request);*/ + ret = OPUS_UNIMPLEMENTED; + break; + } + va_end(ap); + return ret; +# ifdef USE_WEIGHTS_FILE +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +# endif +#else + (void)dred_dec; + (void)request; + return OPUS_UNIMPLEMENTED; +#endif +} + +#ifdef ENABLE_DRED +static int dred_find_payload(const unsigned char *data, opus_int32 len, const unsigned char **payload, int *dred_frame_offset) +{ + const unsigned char *data0; + int len0; + int frame = 0; + int ret; + const unsigned char *frames[48]; + opus_int16 size[48]; + int frame_size; + + *payload = NULL; + /* Get the padding section of the packet. */ + ret = opus_packet_parse_impl(data, len, 0, NULL, frames, size, NULL, NULL, &data0, &len0); + if (ret < 0) + return ret; + frame_size = opus_packet_get_samples_per_frame(data, 48000); + data = data0; + len = len0; + /* Scan extensions in order until we find the earliest frame with DRED data. */ + while (len > 0) + { + opus_int32 header_size; + int id, L; + len0 = len; + data0 = data; + id = *data0 >> 1; + L = *data0 & 0x1; + len = skip_extension(&data, len, &header_size); + if (len < 0) + break; + if (id == 1) + { + if (L==0) + { + frame++; + } else { + frame += data0[1]; + } + } else if (id == DRED_EXTENSION_ID) + { + const unsigned char *curr_payload; + opus_int32 curr_payload_len; + curr_payload = data0+header_size; + curr_payload_len = (data-data0)-header_size; + /* DRED position in the packet, in units of 2.5 ms like for the signaled DRED offset. */ + *dred_frame_offset = frame*frame_size/120; +#ifdef DRED_EXPERIMENTAL_VERSION + /* Check that temporary extension type and version match. + This check will be removed once extension is finalized. */ + if (curr_payload_len > DRED_EXPERIMENTAL_BYTES && curr_payload[0] == 'D' && curr_payload[1] == DRED_EXPERIMENTAL_VERSION) { + *payload = curr_payload+2; + return curr_payload_len-2; + } +#else + if (curr_payload_len > 0) { + *payload = curr_payload; + return curr_payload_len; + } +#endif + } + } + return 0; +} +#endif + +int opus_dred_get_size(void) +{ +#ifdef ENABLE_DRED + return sizeof(OpusDRED); +#else + return 0; +#endif +} + +OpusDRED *opus_dred_alloc(int *error) +{ +#ifdef ENABLE_DRED + OpusDRED *dec; + dec = (OpusDRED *)opus_alloc(opus_dred_get_size()); + if (dec == NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + return dec; +#else + if (error) + *error = OPUS_UNIMPLEMENTED; + return NULL; +#endif +} + +void opus_dred_free(OpusDRED *dec) +{ +#ifdef ENABLE_DRED + opus_free(dec); +#else + (void)dec; +#endif +} + +int opus_dred_parse(OpusDREDDecoder *dred_dec, OpusDRED *dred, const unsigned char *data, opus_int32 len, opus_int32 max_dred_samples, opus_int32 sampling_rate, int *dred_end, int defer_processing) +{ +#ifdef ENABLE_DRED + const unsigned char *payload; + opus_int32 payload_len; + int dred_frame_offset=0; + VALIDATE_DRED_DECODER(dred_dec); + if (!dred_dec->loaded) return OPUS_UNIMPLEMENTED; + dred->process_stage = -1; + payload_len = dred_find_payload(data, len, &payload, &dred_frame_offset); + if (payload_len < 0) + return payload_len; + if (payload != NULL) + { + int offset; + int min_feature_frames; + offset = 100*max_dred_samples/sampling_rate; + min_feature_frames = IMIN(2 + offset, 2*DRED_NUM_REDUNDANCY_FRAMES); + dred_ec_decode(dred, payload, payload_len, min_feature_frames, dred_frame_offset); + if (!defer_processing) + opus_dred_process(dred_dec, dred, dred); + if (dred_end) *dred_end = IMAX(0, -dred->dred_offset*sampling_rate/400); + return IMAX(0, dred->nb_latents*sampling_rate/25 - dred->dred_offset* sampling_rate/400); + } + if (dred_end) *dred_end = 0; + return 0; +#else + (void)dred_dec; + (void)dred; + (void)data; + (void)len; + (void)max_dred_samples; + (void)sampling_rate; + (void)defer_processing; + (void)dred_end; + return OPUS_UNIMPLEMENTED; +#endif +} + +int opus_dred_process(OpusDREDDecoder *dred_dec, const OpusDRED *src, OpusDRED *dst) +{ +#ifdef ENABLE_DRED + if (dred_dec == NULL || src == NULL || dst == NULL || (src->process_stage != 1 && src->process_stage != 2)) + return OPUS_BAD_ARG; + VALIDATE_DRED_DECODER(dred_dec); + if (!dred_dec->loaded) return OPUS_UNIMPLEMENTED; + if (src != dst) + OPUS_COPY(dst, src, 1); + if (dst->process_stage == 2) + return OPUS_OK; + DRED_rdovae_decode_all(&dred_dec->model, dst->fec_features, dst->state, dst->latents, dst->nb_latents, dred_dec->arch); + dst->process_stage = 2; + return OPUS_OK; +#else + (void)dred_dec; + (void)src; + (void)dst; + return OPUS_UNIMPLEMENTED; +#endif +} + +int opus_decoder_dred_decode(OpusDecoder *st, const OpusDRED *dred, opus_int32 dred_offset, opus_int16 *pcm, opus_int32 frame_size) +{ +#ifdef ENABLE_DRED + VARDECL(float, out); + int ret, i; + ALLOC_STACK; + + if(frame_size<=0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + celt_assert(st->channels == 1 || st->channels == 2); + ALLOC(out, frame_size*st->channels, float); + + ret = opus_decode_native(st, NULL, 0, out, frame_size, 0, 0, NULL, 1, dred, dred_offset); + if (ret > 0) + { + for (i=0;ichannels;i++) + pcm[i] = FLOAT2INT16(out[i]); + } + RESTORE_STACK; + return ret; +#else + (void)st; + (void)dred; + (void)dred_offset; + (void)pcm; + (void)frame_size; + return OPUS_UNIMPLEMENTED; +#endif +} + +int opus_decoder_dred_decode_float(OpusDecoder *st, const OpusDRED *dred, opus_int32 dred_offset, float *pcm, opus_int32 frame_size) +{ +#ifdef ENABLE_DRED + if(frame_size<=0) + return OPUS_BAD_ARG; + return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL, 0, dred, dred_offset); +#else + (void)st; + (void)dred; + (void)dred_offset; + (void)pcm; + (void)frame_size; + return OPUS_UNIMPLEMENTED; +#endif +} diff --git a/code/opus-1.5.2/src/opus_demo.c b/code/opus-1.5.2/src/opus_demo.c new file mode 100644 index 00000000..2876fff8 --- /dev/null +++ b/code/opus-1.5.2/src/opus_demo.c @@ -0,0 +1,1127 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include +#include "opus.h" +#include "debug.h" +#include "opus_types.h" +#include "opus_private.h" +#include "opus_multistream.h" +#ifdef ENABLE_LOSSGEN +#include "lossgen.h" +#endif + +#define MAX_PACKET 1500 + +#ifdef USE_WEIGHTS_FILE +# if __unix__ +# include +# include +# include +# include +/* When available, mmap() is preferable to reading the file, as it leads to + better resource utilization, especially if multiple processes are using the same + file (mapping will be shared in cache). */ +void *load_blob(const char *filename, int *len) { + int fd; + void *data; + struct stat st; + if (stat(filename, &st)) { + *len = 0; + return NULL; + } + *len = st.st_size; + fd = open(filename, O_RDONLY); + if (fd<0) { + *len = 0; + return NULL; + } + data = mmap(NULL, *len, PROT_READ, MAP_SHARED, fd, 0); + if (data == MAP_FAILED) { + *len = 0; + data = NULL; + } + close(fd); + return data; +} +void free_blob(void *blob, int len) { + if (blob) munmap(blob, len); +} +# else +void *load_blob(const char *filename, int *len) { + FILE *file; + void *data; + file = fopen(filename, "r"); + if (file == NULL) + { + perror("could not open blob file"); + *len = 0; + return NULL; + } + fseek(file, 0L, SEEK_END); + *len = ftell(file); + fseek(file, 0L, SEEK_SET); + if (*len <= 0) { + *len = 0; + return NULL; + } + data = malloc(*len); + if (!data) { + *len = 0; + return NULL; + } + *len = fread(data, 1, *len, file); + return data; +} +void free_blob(void *blob, int len) { + free(blob); + (void)len; +} +# endif +#endif + + +void print_usage( char* argv[] ) +{ + fprintf(stderr, "Usage: %s [-e] " + " [options] \n", argv[0]); + fprintf(stderr, " %s -d " + "[options] \n\n", argv[0]); + fprintf(stderr, "application: voip | audio | restricted-lowdelay\n" ); + fprintf(stderr, "options:\n" ); + fprintf(stderr, "-e : only runs the encoder (output the bit-stream)\n" ); + fprintf(stderr, "-d : only runs the decoder (reads the bit-stream as input)\n" ); + fprintf(stderr, "-cbr : enable constant bitrate; default: variable bitrate\n" ); + fprintf(stderr, "-cvbr : enable constrained variable bitrate; default: unconstrained\n" ); + fprintf(stderr, "-delayed-decision : use look-ahead for speech/music detection (experts only); default: disabled\n" ); + fprintf(stderr, "-bandwidth : audio bandwidth (from narrowband to fullband); default: sampling rate\n" ); + fprintf(stderr, "-framesize <2.5|5|10|20|40|60|80|100|120> : frame size in ms; default: 20 \n" ); + fprintf(stderr, "-max_payload : maximum payload size in bytes, default: 1024\n" ); + fprintf(stderr, "-complexity : encoder complexity, 0 (lowest) ... 10 (highest); default: 10\n" ); + fprintf(stderr, "-dec_complexity : decoder complexity, 0 (lowest) ... 10 (highest); default: 0\n" ); + fprintf(stderr, "-inbandfec : enable SILK inband FEC\n" ); + fprintf(stderr, "-forcemono : force mono encoding, even for stereo input\n" ); + fprintf(stderr, "-dtx : enable SILK DTX\n" ); + fprintf(stderr, "-loss : optimize for loss percentage and simulate packet loss, in percent (0-100); default: 0\n" ); +#ifdef ENABLE_LOSSGEN + fprintf(stderr, "-sim_loss : simulate realistic (bursty) packet loss from percentage, using generative model\n" ); +#endif + fprintf(stderr, "-lossfile : simulate packet loss, reading loss from file\n" ); + fprintf(stderr, "-dred : add Deep REDundancy (in units of 10-ms frames)\n" ); +} + +static void int_to_char(opus_uint32 i, unsigned char ch[4]) +{ + ch[0] = i>>24; + ch[1] = (i>>16)&0xFF; + ch[2] = (i>>8)&0xFF; + ch[3] = i&0xFF; +} + +static opus_uint32 char_to_int(unsigned char ch[4]) +{ + return ((opus_uint32)ch[0]<<24) | ((opus_uint32)ch[1]<<16) + | ((opus_uint32)ch[2]<< 8) | (opus_uint32)ch[3]; +} + +#define check_encoder_option(decode_only, opt) do {if (decode_only) {fprintf(stderr, "option %s is only for encoding\n", opt); goto failure;}} while(0) +#define check_decoder_option(encode_only, opt) do {if (encode_only) {fprintf(stderr, "option %s is only for decoding\n", opt); goto failure;}} while(0) + +static const int silk8_test[][4] = { + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960*3, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960*2, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 480, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960*3, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960*2, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_NARROWBAND, 480, 2} +}; + +static const int silk12_test[][4] = { + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960*3, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960*2, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 480, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960*3, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960*2, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 960, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_MEDIUMBAND, 480, 2} +}; + +static const int silk16_test[][4] = { + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960*3, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960*2, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 480, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960*3, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960*2, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_WIDEBAND, 480, 2} +}; + +static const int hybrid24_test[][4] = { + {MODE_SILK_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 960, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 480, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 960, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 480, 2} +}; + +static const int hybrid48_test[][4] = { + {MODE_SILK_ONLY, OPUS_BANDWIDTH_FULLBAND, 960, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_FULLBAND, 480, 1}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_FULLBAND, 960, 2}, + {MODE_SILK_ONLY, OPUS_BANDWIDTH_FULLBAND, 480, 2} +}; + +static const int celt_test[][4] = { + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 960, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 960, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960, 1}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 480, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 480, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 480, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 480, 1}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 240, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 240, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 240, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 240, 1}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 120, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 120, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 120, 1}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 120, 1}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 960, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 960, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 960, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 960, 2}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 480, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 480, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 480, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 480, 2}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 240, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 240, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 240, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 240, 2}, + + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 120, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_SUPERWIDEBAND, 120, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_WIDEBAND, 120, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_NARROWBAND, 120, 2}, + +}; + +static const int celt_hq_test[][4] = { + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 960, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 480, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 240, 2}, + {MODE_CELT_ONLY, OPUS_BANDWIDTH_FULLBAND, 120, 2}, +}; + +#if 0 /* This is a hack that replaces the normal encoder/decoder with the multistream version */ +#define OpusEncoder OpusMSEncoder +#define OpusDecoder OpusMSDecoder +#define opus_encode opus_multistream_encode +#define opus_decode opus_multistream_decode +#define opus_encoder_ctl opus_multistream_encoder_ctl +#define opus_decoder_ctl opus_multistream_decoder_ctl +#define opus_encoder_create ms_opus_encoder_create +#define opus_decoder_create ms_opus_decoder_create +#define opus_encoder_destroy opus_multistream_encoder_destroy +#define opus_decoder_destroy opus_multistream_decoder_destroy + +static OpusEncoder *ms_opus_encoder_create(opus_int32 Fs, int channels, int application, int *error) +{ + int streams, coupled_streams; + unsigned char mapping[256]; + return (OpusEncoder *)opus_multistream_surround_encoder_create(Fs, channels, 1, &streams, &coupled_streams, mapping, application, error); +} +static OpusDecoder *ms_opus_decoder_create(opus_int32 Fs, int channels, int *error) +{ + int streams; + int coupled_streams; + unsigned char mapping[256]={0,1}; + streams = 1; + coupled_streams = channels==2; + return (OpusDecoder *)opus_multistream_decoder_create(Fs, channels, streams, coupled_streams, mapping, error); +} +#endif + + +#ifdef ENABLE_OSCE_TRAINING_DATA +#define COMPLEXITY_MIN 0 +#define COMPLEXITY_MAX 10 + +#define PACKET_LOSS_PERC_MIN 0 +#define PACKET_LOSS_PERC_MAX 50 +#define PACKET_LOSS_PERC_STEP 5 + +#define CBR_BITRATE_LIMIT 8000 + +#define NUM_BITRATES 102 +static int bitrates[NUM_BITRATES] = { + 6000, 6060, 6120, 6180, 6240, 6300, 6360, 6420, 6480, + 6525, 6561, 6598, 6634, 6670, 6707, 6743, 6780, 6816, + 6853, 6889, 6926, 6962, 6999, 7042, 7085, 7128, 7171, + 7215, 7258, 7301, 7344, 7388, 7431, 7474, 7512, 7541, + 7570, 7599, 7628, 7657, 7686, 7715, 7744, 7773, 7802, + 7831, 7860, 7889, 7918, 7947, 7976, 8013, 8096, 8179, + 8262, 8344, 8427, 8511, 8605, 8699, 8792, 8886, 8980, + 9100, 9227, 9354, 9480, 9561, 9634, 9706, 9779, 9851, + 9924, 9996, 10161, 10330, 10499, 10698, 10898, 11124, 11378, + 11575, 11719, 11862, 12014, 12345, 12751, 13195, 13561, 13795, + 14069, 14671, 15403, 15790, 16371, 17399, 17968, 19382, 20468, + 22000, 32000, 64000 +}; + +static int randint(int min, int max, int step) +{ + double r = ((double) rand())/ (RAND_MAX + 1.); + int d; + + d = ((int) ((max + 1 - min) * r / step) * step) + min; + + return d; +} + +static void new_random_setting(OpusEncoder *enc) +{ + int bitrate_bps; + int complexity; + int packet_loss_perc; + int use_vbr; + + bitrate_bps = bitrates[randint(0, NUM_BITRATES - 1, 1)]; + complexity = randint(COMPLEXITY_MIN, COMPLEXITY_MAX, 1); + packet_loss_perc = randint(PACKET_LOSS_PERC_MIN, PACKET_LOSS_PERC_MAX, PACKET_LOSS_PERC_STEP); + use_vbr = bitrate_bps < CBR_BITRATE_LIMIT ? 1 : randint(0, 1, 1); + + if (1) + { + printf("changing settings to %d\t%d\t%d\t%d\n", bitrate_bps, complexity, packet_loss_perc, use_vbr); + } + + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrate_bps)); + opus_encoder_ctl(enc, OPUS_SET_COMPLEXITY(complexity)); + opus_encoder_ctl(enc, OPUS_SET_PACKET_LOSS_PERC(packet_loss_perc)); + opus_encoder_ctl(enc, OPUS_SET_VBR(use_vbr)); +} + +#endif + +int main(int argc, char *argv[]) +{ + int err; + char *inFile, *outFile; + FILE *fin=NULL; + FILE *fout=NULL; + OpusEncoder *enc=NULL; + OpusDecoder *dec=NULL; + OpusDRED *dred=NULL; + OpusDREDDecoder *dred_dec=NULL; + int args; + int len; + int frame_size, channels; + opus_int32 bitrate_bps=0; + unsigned char *data = NULL; + unsigned char *fbytes=NULL; + opus_int32 sampling_rate; + int use_vbr; + int max_payload_bytes; + int complexity; + int dec_complexity; + int use_inbandfec; + int use_dtx; + int forcechannels; + int cvbr = 0; + int packet_loss_perc; +#ifdef ENABLE_LOSSGEN + float lossgen_perc = -1.f; + LossGenState lossgen; +#endif + opus_int32 count=0, count_act=0; + int k; + opus_int32 skip=0; + int stop=0; + short *in=NULL; + short *out=NULL; + int application=OPUS_APPLICATION_AUDIO; + double bits=0.0, bits_max=0.0, bits_act=0.0, bits2=0.0, nrg; + double tot_samples=0; + opus_uint64 tot_in, tot_out; + int bandwidth=OPUS_AUTO; + const char *bandwidth_string; + int lost = 0, lost_prev = 1; + opus_uint32 enc_final_range; + opus_uint32 dec_final_range; + int encode_only=0, decode_only=0; + int max_frame_size = 48000*2; + size_t num_read; + int curr_read=0; + int sweep_bps = 0; + int random_framesize=0, newsize=0, delayed_celt=0; + int sweep_max=0, sweep_min=0; + int random_fec=0; + const int (*mode_list)[4]=NULL; + int nb_modes_in_list=0; + int curr_mode=0; + int curr_mode_count=0; + int mode_switch_time = 48000; + int nb_encoded=0; + int remaining=0; + int variable_duration=OPUS_FRAMESIZE_ARG; + int delayed_decision=0; + int ret = EXIT_FAILURE; + int lost_count=0; + FILE *packet_loss_file=NULL; + int dred_duration=0; +#ifdef ENABLE_OSCE_TRAINING_DATA + int silk_random_switching = 0; + int silk_frame_counter = 0; +#endif +#ifdef USE_WEIGHTS_FILE + int blob_len; + void *blob_data; + const char *filename = "weights_blob.bin"; + blob_data = load_blob(filename, &blob_len); +#endif + + if (argc < 5 ) + { + print_usage( argv ); + goto failure; + } + + tot_in=tot_out=0; + fprintf(stderr, "%s\n", opus_get_version_string()); + + args = 1; + if (strcmp(argv[args], "-e")==0) + { + encode_only = 1; + args++; + } else if (strcmp(argv[args], "-d")==0) + { + decode_only = 1; + args++; + } + if (!decode_only && argc < 7 ) + { + print_usage( argv ); + goto failure; + } + + if (!decode_only) + { + if (strcmp(argv[args], "voip")==0) + application = OPUS_APPLICATION_VOIP; + else if (strcmp(argv[args], "restricted-lowdelay")==0) + application = OPUS_APPLICATION_RESTRICTED_LOWDELAY; + else if (strcmp(argv[args], "audio")!=0) { + fprintf(stderr, "unknown application: %s\n", argv[args]); + print_usage(argv); + goto failure; + } + args++; + } + sampling_rate = (opus_int32)atol(argv[args]); + args++; + + if (sampling_rate != 8000 && sampling_rate != 12000 + && sampling_rate != 16000 && sampling_rate != 24000 + && sampling_rate != 48000) + { + fprintf(stderr, "Supported sampling rates are 8000, 12000, " + "16000, 24000 and 48000.\n"); + goto failure; + } + frame_size = sampling_rate/50; + + channels = atoi(argv[args]); + args++; + + if (channels < 1 || channels > 2) + { + fprintf(stderr, "Opus_demo supports only 1 or 2 channels.\n"); + goto failure; + } + + if (!decode_only) + { + bitrate_bps = (opus_int32)atol(argv[args]); + args++; + } + + /* defaults: */ + use_vbr = 1; + max_payload_bytes = MAX_PACKET; + complexity = 10; + dec_complexity = 0; + use_inbandfec = 0; + forcechannels = OPUS_AUTO; + use_dtx = 0; + packet_loss_perc = 0; + + while( args < argc - 2 ) { + /* process command line options */ + if( strcmp( argv[ args ], "-cbr" ) == 0 ) { + check_encoder_option(decode_only, "-cbr"); + use_vbr = 0; + args++; + } else if( strcmp( argv[ args ], "-bandwidth" ) == 0 ) { + check_encoder_option(decode_only, "-bandwidth"); + if (strcmp(argv[ args + 1 ], "NB")==0) + bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (strcmp(argv[ args + 1 ], "MB")==0) + bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (strcmp(argv[ args + 1 ], "WB")==0) + bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (strcmp(argv[ args + 1 ], "SWB")==0) + bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else if (strcmp(argv[ args + 1 ], "FB")==0) + bandwidth = OPUS_BANDWIDTH_FULLBAND; + else { + fprintf(stderr, "Unknown bandwidth %s. " + "Supported are NB, MB, WB, SWB, FB.\n", + argv[ args + 1 ]); + goto failure; + } + args += 2; + } else if( strcmp( argv[ args ], "-framesize" ) == 0 ) { + check_encoder_option(decode_only, "-framesize"); + if (strcmp(argv[ args + 1 ], "2.5")==0) + frame_size = sampling_rate/400; + else if (strcmp(argv[ args + 1 ], "5")==0) + frame_size = sampling_rate/200; + else if (strcmp(argv[ args + 1 ], "10")==0) + frame_size = sampling_rate/100; + else if (strcmp(argv[ args + 1 ], "20")==0) + frame_size = sampling_rate/50; + else if (strcmp(argv[ args + 1 ], "40")==0) + frame_size = sampling_rate/25; + else if (strcmp(argv[ args + 1 ], "60")==0) + frame_size = 3*sampling_rate/50; + else if (strcmp(argv[ args + 1 ], "80")==0) + frame_size = 4*sampling_rate/50; + else if (strcmp(argv[ args + 1 ], "100")==0) + frame_size = 5*sampling_rate/50; + else if (strcmp(argv[ args + 1 ], "120")==0) + frame_size = 6*sampling_rate/50; + else { + fprintf(stderr, "Unsupported frame size: %s ms. " + "Supported are 2.5, 5, 10, 20, 40, 60, 80, 100, 120.\n", + argv[ args + 1 ]); + goto failure; + } + args += 2; + } else if( strcmp( argv[ args ], "-max_payload" ) == 0 ) { + check_encoder_option(decode_only, "-max_payload"); + max_payload_bytes = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-complexity" ) == 0 ) { + check_encoder_option(decode_only, "-complexity"); + complexity = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-dec_complexity" ) == 0 ) { + check_decoder_option(encode_only, "-dec_complexity"); + dec_complexity = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-inbandfec" ) == 0 ) { + use_inbandfec = 1; + args++; + } else if( strcmp( argv[ args ], "-forcemono" ) == 0 ) { + check_encoder_option(decode_only, "-forcemono"); + forcechannels = 1; + args++; + } else if( strcmp( argv[ args ], "-cvbr" ) == 0 ) { + check_encoder_option(decode_only, "-cvbr"); + cvbr = 1; + args++; + } else if( strcmp( argv[ args ], "-delayed-decision" ) == 0 ) { + check_encoder_option(decode_only, "-delayed-decision"); + delayed_decision = 1; + args++; + } else if( strcmp( argv[ args ], "-dtx") == 0 ) { + check_encoder_option(decode_only, "-dtx"); + use_dtx = 1; + args++; + } else if( strcmp( argv[ args ], "-loss" ) == 0 ) { + packet_loss_perc = atoi( argv[ args + 1 ] ); + args += 2; +#ifdef ENABLE_LOSSGEN + } else if( strcmp( argv[ args ], "-sim_loss" ) == 0 ) { + lossgen_perc = atof( argv[ args + 1 ] ); + lossgen_init(&lossgen); + args += 2; +#endif + } else if( strcmp( argv[ args ], "-lossfile" ) == 0 ) { + packet_loss_file = fopen( argv[ args + 1 ], "r" ); + if (packet_loss_file == NULL) { + fprintf(stderr, "failed to open loss file %s\n", argv[ args + 1 ] ); + exit(1); + } + args += 2; + } else if( strcmp( argv[ args ], "-dred" ) == 0 ) { + dred_duration = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-sweep" ) == 0 ) { + check_encoder_option(decode_only, "-sweep"); + sweep_bps = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-random_framesize" ) == 0 ) { + check_encoder_option(decode_only, "-random_framesize"); + random_framesize = 1; + args++; + } else if( strcmp( argv[ args ], "-sweep_max" ) == 0 ) { + check_encoder_option(decode_only, "-sweep_max"); + sweep_max = atoi( argv[ args + 1 ] ); + args += 2; + } else if( strcmp( argv[ args ], "-random_fec" ) == 0 ) { + check_encoder_option(decode_only, "-random_fec"); + random_fec = 1; + args++; + } else if( strcmp( argv[ args ], "-silk8k_test" ) == 0 ) { + check_encoder_option(decode_only, "-silk8k_test"); + mode_list = silk8_test; + nb_modes_in_list = 8; + args++; + } else if( strcmp( argv[ args ], "-silk12k_test" ) == 0 ) { + check_encoder_option(decode_only, "-silk12k_test"); + mode_list = silk12_test; + nb_modes_in_list = 8; + args++; + } else if( strcmp( argv[ args ], "-silk16k_test" ) == 0 ) { + check_encoder_option(decode_only, "-silk16k_test"); + mode_list = silk16_test; + nb_modes_in_list = 8; + args++; + } else if( strcmp( argv[ args ], "-hybrid24k_test" ) == 0 ) { + check_encoder_option(decode_only, "-hybrid24k_test"); + mode_list = hybrid24_test; + nb_modes_in_list = 4; + args++; + } else if( strcmp( argv[ args ], "-hybrid48k_test" ) == 0 ) { + check_encoder_option(decode_only, "-hybrid48k_test"); + mode_list = hybrid48_test; + nb_modes_in_list = 4; + args++; + } else if( strcmp( argv[ args ], "-celt_test" ) == 0 ) { + check_encoder_option(decode_only, "-celt_test"); + mode_list = celt_test; + nb_modes_in_list = 32; + args++; + } else if( strcmp( argv[ args ], "-celt_hq_test" ) == 0 ) { + check_encoder_option(decode_only, "-celt_hq_test"); + mode_list = celt_hq_test; + nb_modes_in_list = 4; + args++; +#ifdef ENABLE_OSCE_TRAINING_DATA + } else if( strcmp( argv[ args ], "-silk_random_switching" ) == 0 ){ + silk_random_switching = atoi( argv[ args + 1 ] ); + printf("switching encoding parameters every %dth frame\n", silk_random_switching); + args += 2; +#endif + } else { + printf( "Error: unrecognized setting: %s\n\n", argv[ args ] ); + print_usage( argv ); + goto failure; + } + } + + if (sweep_max) + sweep_min = bitrate_bps; + + if (max_payload_bytes < 0 || max_payload_bytes > MAX_PACKET) + { + fprintf (stderr, "max_payload_bytes must be between 0 and %d\n", + MAX_PACKET); + goto failure; + } + + inFile = argv[argc-2]; + fin = fopen(inFile, "rb"); + if (!fin) + { + fprintf (stderr, "Could not open input file %s\n", argv[argc-2]); + goto failure; + } + if (mode_list) + { + int size; + fseek(fin, 0, SEEK_END); + size = ftell(fin); + fprintf(stderr, "File size is %d bytes\n", size); + fseek(fin, 0, SEEK_SET); + mode_switch_time = size/sizeof(short)/channels/nb_modes_in_list; + fprintf(stderr, "Switching mode every %d samples\n", mode_switch_time); + } + + outFile = argv[argc-1]; + fout = fopen(outFile, "wb+"); + if (!fout) + { + fprintf (stderr, "Could not open output file %s\n", argv[argc-1]); + goto failure; + } + + if (!decode_only) + { + enc = opus_encoder_create(sampling_rate, channels, application, &err); + if (err != OPUS_OK) + { + fprintf(stderr, "Cannot create encoder: %s\n", opus_strerror(err)); + goto failure; + } + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrate_bps)); + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(bandwidth)); + opus_encoder_ctl(enc, OPUS_SET_VBR(use_vbr)); + opus_encoder_ctl(enc, OPUS_SET_VBR_CONSTRAINT(cvbr)); + opus_encoder_ctl(enc, OPUS_SET_COMPLEXITY(complexity)); + opus_encoder_ctl(enc, OPUS_SET_INBAND_FEC(use_inbandfec)); + opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(forcechannels)); + opus_encoder_ctl(enc, OPUS_SET_DTX(use_dtx)); + opus_encoder_ctl(enc, OPUS_SET_PACKET_LOSS_PERC(packet_loss_perc)); + + opus_encoder_ctl(enc, OPUS_GET_LOOKAHEAD(&skip)); + opus_encoder_ctl(enc, OPUS_SET_LSB_DEPTH(16)); + opus_encoder_ctl(enc, OPUS_SET_EXPERT_FRAME_DURATION(variable_duration)); + if (dred_duration > 0) + { + opus_encoder_ctl(enc, OPUS_SET_DRED_DURATION(dred_duration)); + } + } + if (!encode_only) + { + dec = opus_decoder_create(sampling_rate, channels, &err); + if (err != OPUS_OK) + { + fprintf(stderr, "Cannot create decoder: %s\n", opus_strerror(err)); + goto failure; + } + opus_decoder_ctl(dec, OPUS_SET_COMPLEXITY(dec_complexity)); + } + switch(bandwidth) + { + case OPUS_BANDWIDTH_NARROWBAND: + bandwidth_string = "narrowband"; + break; + case OPUS_BANDWIDTH_MEDIUMBAND: + bandwidth_string = "mediumband"; + break; + case OPUS_BANDWIDTH_WIDEBAND: + bandwidth_string = "wideband"; + break; + case OPUS_BANDWIDTH_SUPERWIDEBAND: + bandwidth_string = "superwideband"; + break; + case OPUS_BANDWIDTH_FULLBAND: + bandwidth_string = "fullband"; + break; + case OPUS_AUTO: + bandwidth_string = "auto bandwidth"; + break; + default: + bandwidth_string = "unknown"; + break; + } + + if (decode_only) + fprintf(stderr, "Decoding with %ld Hz output (%d channels)\n", + (long)sampling_rate, channels); + else + fprintf(stderr, "Encoding %ld Hz input at %.3f kb/s " + "in %s with %d-sample frames.\n", + (long)sampling_rate, bitrate_bps*0.001, + bandwidth_string, frame_size); + + in = (short*)malloc(max_frame_size*channels*sizeof(short)); + out = (short*)malloc(max_frame_size*channels*sizeof(short)); + /* We need to allocate for 16-bit PCM data, but we store it as unsigned char. */ + fbytes = (unsigned char*)malloc(max_frame_size*channels*sizeof(short)); + data = (unsigned char*)calloc(max_payload_bytes,sizeof(unsigned char)); + if(delayed_decision) + { + if (frame_size==sampling_rate/400) + variable_duration = OPUS_FRAMESIZE_2_5_MS; + else if (frame_size==sampling_rate/200) + variable_duration = OPUS_FRAMESIZE_5_MS; + else if (frame_size==sampling_rate/100) + variable_duration = OPUS_FRAMESIZE_10_MS; + else if (frame_size==sampling_rate/50) + variable_duration = OPUS_FRAMESIZE_20_MS; + else if (frame_size==sampling_rate/25) + variable_duration = OPUS_FRAMESIZE_40_MS; + else if (frame_size==3*sampling_rate/50) + variable_duration = OPUS_FRAMESIZE_60_MS; + else if (frame_size==4*sampling_rate/50) + variable_duration = OPUS_FRAMESIZE_80_MS; + else if (frame_size==5*sampling_rate/50) + variable_duration = OPUS_FRAMESIZE_100_MS; + else + variable_duration = OPUS_FRAMESIZE_120_MS; + opus_encoder_ctl(enc, OPUS_SET_EXPERT_FRAME_DURATION(variable_duration)); + frame_size = 2*48000; + } + dred_dec = opus_dred_decoder_create(&err); + dred = opus_dred_alloc(&err); +#ifdef USE_WEIGHTS_FILE + if (enc) opus_encoder_ctl(enc, OPUS_SET_DNN_BLOB(blob_data, blob_len)); + if (dec) opus_decoder_ctl(dec, OPUS_SET_DNN_BLOB(blob_data, blob_len)); + if (dred_dec) opus_dred_decoder_ctl(dred_dec, OPUS_SET_DNN_BLOB(blob_data, blob_len)); +#endif + while (!stop) + { + if (delayed_celt) + { + frame_size = newsize; + delayed_celt = 0; + } else if (random_framesize && rand()%20==0) + { + newsize = rand()%6; + switch(newsize) + { + case 0: newsize=sampling_rate/400; break; + case 1: newsize=sampling_rate/200; break; + case 2: newsize=sampling_rate/100; break; + case 3: newsize=sampling_rate/50; break; + case 4: newsize=sampling_rate/25; break; + case 5: newsize=3*sampling_rate/50; break; + } + while (newsize < sampling_rate/25 && bitrate_bps-abs(sweep_bps) <= 3*12*sampling_rate/newsize) + newsize*=2; + if (newsize < sampling_rate/100 && frame_size >= sampling_rate/100) + { + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + delayed_celt=1; + } else { + frame_size = newsize; + } + } + if (random_fec && rand()%30==0) + { + opus_encoder_ctl(enc, OPUS_SET_INBAND_FEC(rand()%4==0)); + } + if (decode_only) + { + unsigned char ch[4]; + num_read = fread(ch, 1, 4, fin); + if (num_read!=4) + break; + len = char_to_int(ch); + if (len>max_payload_bytes || len<0) + { + fprintf(stderr, "Invalid payload length: %d\n",len); + break; + } + num_read = fread(ch, 1, 4, fin); + if (num_read!=4) + break; + enc_final_range = char_to_int(ch); + num_read = fread(data, 1, len, fin); + if (num_read!=(size_t)len) + { + fprintf(stderr, "Ran out of input, " + "expecting %d bytes got %d\n", + len,(int)num_read); + break; + } + } else { + int i; + if (mode_list!=NULL) + { + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(mode_list[curr_mode][1])); + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(mode_list[curr_mode][0])); + opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(mode_list[curr_mode][3])); + frame_size = mode_list[curr_mode][2]; + } +#ifdef ENABLE_OSCE_TRAINING_DATA + if (silk_random_switching) + { + silk_frame_counter += 1; + if (silk_frame_counter % silk_random_switching == 0) { + new_random_setting(enc); + } + } +#endif + num_read = fread(fbytes, sizeof(short)*channels, frame_size-remaining, fin); + curr_read = (int)num_read; + tot_in += curr_read; + for(i=0;i sweep_max) + sweep_bps = -sweep_bps; + else if (bitrate_bps < sweep_min) + sweep_bps = -sweep_bps; + } + /* safety */ + if (bitrate_bps<1000) + bitrate_bps = 1000; + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrate_bps)); + } + opus_encoder_ctl(enc, OPUS_GET_FINAL_RANGE(&enc_final_range)); + curr_mode_count += frame_size; + if (curr_mode_count > mode_switch_time && curr_mode < nb_modes_in_list-1) + { + curr_mode++; + curr_mode_count = 0; + } + } + +#if 0 /* This is for testing the padding code, do not enable by default */ + if (len<1275) + { + int new_len = len+rand()%(max_payload_bytes-len); + if ((err = opus_packet_pad(data, len, new_len)) != OPUS_OK) + { + fprintf(stderr, "padding failed: %s\n", opus_strerror(err)); + goto failure; + } + len = new_len; + } +#endif + if (encode_only) + { + unsigned char int_field[4]; + int_to_char(len, int_field); + if (fwrite(int_field, 1, 4, fout) != 4) { + fprintf(stderr, "Error writing.\n"); + goto failure; + } + int_to_char(enc_final_range, int_field); + if (fwrite(int_field, 1, 4, fout) != 4) { + fprintf(stderr, "Error writing.\n"); + goto failure; + } + if (fwrite(data, 1, len, fout) != (unsigned)len) { + fprintf(stderr, "Error writing.\n"); + goto failure; + } + tot_samples += nb_encoded; + } else { + int fr; + int run_decoder; + int dred_input=0; + int dred_end=0; + if (packet_loss_file != NULL) { + if ( fscanf(packet_loss_file, "%d", &lost) != 1) { + lost = 0; + } +#ifdef ENABLE_LOSSGEN + } else if (lossgen_perc >= 0) { + lost = sample_loss(&lossgen, lossgen_perc*.01f); +#endif + } else { + lost = (packet_loss_perc>0) && (rand()%100 < packet_loss_perc); + } + if (len == 0) lost = 1; + if (lost) + { + lost_count++; + run_decoder = 0; + } else { + run_decoder= 1; + } + if (run_decoder) + run_decoder += lost_count; + if (!lost && lost_count > 0) { + opus_int32 output_samples=0; + opus_decoder_ctl(dec, OPUS_GET_LAST_PACKET_DURATION(&output_samples)); + dred_input = lost_count*output_samples; + /* Only decode the amount we need to fill in the gap. */ + ret = opus_dred_parse(dred_dec, dred, data, len, IMIN(48000, IMAX(0, dred_input)), sampling_rate, &dred_end, 0); + dred_input = ret > 0 ? ret : 0; + } + /* FIXME: Figure out how to trigger the decoder when the last packet of the file is lost. */ + for (fr=0;fr 0) + output_samples = opus_decoder_dred_decode(dec, dred, (lost_count-fr)*output_samples, out, output_samples); + else + output_samples = opus_decode(dec, NULL, 0, out, output_samples, 0); + } else { + output_samples = max_frame_size; + output_samples = opus_decode(dec, data, len, out, output_samples, 0); + } + if (output_samples>0) + { + if (!decode_only && tot_out + output_samples > tot_in) + { + stop=1; + output_samples = (opus_int32)(tot_in - tot_out); + } + if (output_samples>skip) { + int i; + for(i=0;i<(output_samples-skip)*channels;i++) + { + short s; + s=out[i+(skip*channels)]; + fbytes[2*i]=s&0xFF; + fbytes[2*i+1]=(s>>8)&0xFF; + } + if (fwrite(fbytes, sizeof(short)*channels, output_samples-skip, fout) != (unsigned)(output_samples-skip)){ + fprintf(stderr, "Error writing.\n"); + goto failure; + } + tot_out += output_samples-skip; + } + if (output_samples= use_inbandfec ) { + /* count bits */ + bits += len*8; + bits_max = ( len*8 > bits_max ) ? len*8 : bits_max; + bits2 += len*len*64; + if (!decode_only) + { + nrg = 0.0; + for ( k = 0; k < frame_size * channels; k++ ) { + nrg += in[ k ] * (double)in[ k ]; + } + nrg /= frame_size * channels; + if( nrg > 1e5 ) { + bits_act += len*8; + count_act++; + } + } + } + count++; + } + + if(decode_only && count > 0) + frame_size = (int)(tot_samples / count); + count -= use_inbandfec; + if (tot_samples >= 1 && count > 0 && frame_size) + { + /* Print out bitrate statistics */ + double var; + fprintf (stderr, "average bitrate: %7.3f kb/s\n", + 1e-3*bits*sampling_rate/tot_samples); + fprintf (stderr, "maximum bitrate: %7.3f kb/s\n", + 1e-3*bits_max*sampling_rate/frame_size); + if (!decode_only) + fprintf (stderr, "active bitrate: %7.3f kb/s\n", + 1e-3*bits_act*sampling_rate/(1e-15+frame_size*(double)count_act)); + var = bits2/count - bits*bits/(count*(double)count); + if (var < 0) + var = 0; + fprintf (stderr, "bitrate standard deviation: %7.3f kb/s\n", + 1e-3*sqrt(var)*sampling_rate/frame_size); + } else { + fprintf(stderr, "bitrate statistics are undefined\n"); + } + silk_TimerSave("opus_timing.txt"); + ret = EXIT_SUCCESS; +failure: + opus_encoder_destroy(enc); + opus_decoder_destroy(dec); + opus_dred_free(dred); + opus_dred_decoder_destroy(dred_dec); + free(data); + if (fin) + fclose(fin); + if (fout) + fclose(fout); + free(in); + free(out); + free(fbytes); +#ifdef USE_WEIGHTS_FILE + free_blob(blob_data, blob_len); +#endif + return ret; +} diff --git a/code/opus-1.2.1/src/opus_encoder.c b/code/opus-1.5.2/src/opus_encoder.c similarity index 78% rename from code/opus-1.2.1/src/opus_encoder.c rename to code/opus-1.5.2/src/opus_encoder.c index 474e6c24..d18d582f 100644 --- a/code/opus-1.2.1/src/opus_encoder.c +++ b/code/opus-1.5.2/src/opus_encoder.c @@ -45,11 +45,19 @@ #include "analysis.h" #include "mathops.h" #include "tuning_parameters.h" + +#ifdef ENABLE_DRED +#include "dred_coding.h" +#endif + #ifdef FIXED_POINT #include "fixed/structs_FIX.h" #else #include "float/structs_FLP.h" #endif +#ifdef ENABLE_OSCE_TRAINING_DATA +#include +#endif #define MAX_ENCODER_BUFFER 480 @@ -67,6 +75,9 @@ struct OpusEncoder { int celt_enc_offset; int silk_enc_offset; silk_EncControlStruct silk_mode; +#ifdef ENABLE_DRED + DREDEnc dred_encoder; +#endif int application; int channels; int delay_compensation; @@ -87,6 +98,7 @@ struct OpusEncoder { int lfe; int arch; int use_dtx; /* general DTX for both SILK and CELT */ + int fec_config; #ifndef DISABLE_FLOAT_API TonalityAnalysisState analysis; #endif @@ -112,8 +124,16 @@ struct OpusEncoder { opus_val16 delay_buffer[MAX_ENCODER_BUFFER*2]; #ifndef DISABLE_FLOAT_API int detected_bandwidth; - int nb_no_activity_frames; + int nb_no_activity_ms_Q1; opus_val32 peak_signal_energy; +#endif +#ifdef ENABLE_DRED + int dred_duration; + int dred_q0; + int dred_dQ; + int dred_qmax; + int dred_target_chunks; + unsigned char activity_mem[DRED_MAX_FRAMES*4]; /* 2.5ms resolution*/ #endif int nonfinal_frame; /* current frame is not the final in a packet */ opus_uint32 rangeFinal; @@ -123,38 +143,38 @@ struct OpusEncoder { middle (memoriless) threshold. The second column is the hysteresis (difference with the middle) */ static const opus_int32 mono_voice_bandwidth_thresholds[8] = { - 10000, 1000, /* NB<->MB */ - 11000, 1000, /* MB<->WB */ + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ 13500, 1000, /* WB<->SWB */ 14000, 2000, /* SWB<->FB */ }; static const opus_int32 mono_music_bandwidth_thresholds[8] = { - 10000, 1000, /* NB<->MB */ - 11000, 1000, /* MB<->WB */ - 13500, 1000, /* WB<->SWB */ - 14000, 2000, /* SWB<->FB */ + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ }; static const opus_int32 stereo_voice_bandwidth_thresholds[8] = { - 10000, 1000, /* NB<->MB */ - 11000, 1000, /* MB<->WB */ + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ 13500, 1000, /* WB<->SWB */ 14000, 2000, /* SWB<->FB */ }; static const opus_int32 stereo_music_bandwidth_thresholds[8] = { - 10000, 1000, /* NB<->MB */ - 11000, 1000, /* MB<->WB */ - 13500, 1000, /* WB<->SWB */ - 14000, 2000, /* SWB<->FB */ + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ }; /* Threshold bit-rates for switching between mono and stereo */ -static const opus_int32 stereo_voice_threshold = 24000; -static const opus_int32 stereo_music_threshold = 24000; +static const opus_int32 stereo_voice_threshold = 19000; +static const opus_int32 stereo_music_threshold = 17000; /* Threshold bit-rate for switching between SILK/hybrid and CELT-only */ static const opus_int32 mode_thresholds[2][2] = { /* voice */ /* music */ - { 64000, 16000}, /* mono */ - { 36000, 16000}, /* stereo */ + { 64000, 10000}, /* mono */ + { 44000, 10000}, /* stereo */ }; static const opus_int32 fec_thresholds[] = { @@ -223,6 +243,7 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat st->silk_mode.packetLossPercentage = 0; st->silk_mode.complexity = 9; st->silk_mode.useInBandFEC = 0; + st->silk_mode.useDRED = 0; st->silk_mode.useDTX = 0; st->silk_mode.useCBR = 0; st->silk_mode.reducedDependency = 0; @@ -235,6 +256,11 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0)); celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(st->silk_mode.complexity)); +#ifdef ENABLE_DRED + /* Initialize DRED Encoder */ + dred_encoder_init( &st->dred_encoder, Fs, channels ); +#endif + st->use_vbr = 1; /* Makes constrained VBR the default (safer for real-time use) */ st->vbr_constraint = 1; @@ -385,20 +411,16 @@ static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *ou int c, i; int shift; - /* Approximates -round(log2(4.*cutoff_Hz/Fs)) */ - shift=celt_ilog2(Fs/(cutoff_Hz*3)); + /* Approximates -round(log2(6.3*cutoff_Hz/Fs)) */ + shift=celt_ilog2(Fs/(cutoff_Hz*4)); for (c=0;csilk_mode.useInBandFEC) { + dred_frac = MIN16(.7f, 3.f*st->silk_mode.packetLossPercentage/100.f); + bitrate_offset = 20000; + } else { + if (st->silk_mode.packetLossPercentage > 5) { + dred_frac = MIN16(.8f, .55f + st->silk_mode.packetLossPercentage/100.f); + } else { + dred_frac = 12*st->silk_mode.packetLossPercentage/100.f; + } + bitrate_offset = 12000; + } + /* Account for the fact that longer packets require less redundancy. */ + dred_frac = dred_frac/(dred_frac + (1-dred_frac)*(frame_size*50.f)/st->Fs); + /* Approximate fit based on a few experiments. Could probably be improved. */ + q0 = IMIN(15, IMAX(4, 51 - 3*EC_ILOG(IMAX(1, bitrate_bps-bitrate_offset)))); + dQ = bitrate_bps-bitrate_offset > 36000 ? 3 : 5; + qmax = 15; + target_dred_bitrate = IMAX(0, (int)(dred_frac*(bitrate_bps-bitrate_offset))); + if (st->dred_duration > 0) { + opus_int32 target_bits = target_dred_bitrate*frame_size/st->Fs; + max_dred_bits = estimate_dred_bitrate(q0, dQ, qmax, st->dred_duration, target_bits, &target_chunks); + } else { + max_dred_bits = 0; + target_chunks=0; + } + dred_bitrate = IMIN(target_dred_bitrate, max_dred_bits*st->Fs/frame_size); + /* If we can't afford enough bits, don't bother with DRED at all. */ + if (target_chunks < 2) + dred_bitrate = 0; + st->dred_q0 = q0; + st->dred_dQ = dQ; + st->dred_qmax = qmax; + st->dred_target_chunks = target_chunks; + return dred_bitrate; +} +#endif + static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int max_data_bytes) { if(!frame_size)frame_size=st->Fs/400; @@ -699,6 +772,12 @@ opus_val16 compute_stereo_width(const opus_val16 *pcm, int frame_size, opus_int3 xy += SHR32(pxy, 10); yy += SHR32(pyy, 10); } +#ifndef FIXED_POINT + if (!(xx < 1e9f) || celt_isnan(xx) || !(yy < 1e9f) || celt_isnan(yy)) + { + xy = xx = yy = 0; + } +#endif mem->XX += MULT16_32_Q15(short_alpha, xx-mem->XX); mem->XY += MULT16_32_Q15(short_alpha, xy-mem->XY); mem->YY += MULT16_32_Q15(short_alpha, yy-mem->YY); @@ -762,7 +841,7 @@ static int decide_fec(int useInBandFEC, int PacketLoss_perc, int last_fec, int m return 0; } -static int compute_silk_rate_for_hybrid(int rate, int bandwidth, int frame20ms, int vbr, int fec) { +static int compute_silk_rate_for_hybrid(int rate, int bandwidth, int frame20ms, int vbr, int fec, int channels) { int entry; int i; int N; @@ -779,6 +858,8 @@ static int compute_silk_rate_for_hybrid(int rate, int bandwidth, int frame20ms, {32000, 22000, 22000, 28000, 28000}, {64000, 38000, 38000, 50000, 50000} }; + /* Do the allocation per-channel. */ + rate /= channels; entry = 1 + frame20ms + 2*fec; N = sizeof(rate_table)/sizeof(rate_table[0]); for (i=1;i= 12000) + silk_rate -= 1000; return silk_rate; } @@ -816,7 +901,8 @@ static opus_int32 compute_equiv_rate(opus_int32 bitrate, int channels, opus_int32 equiv; equiv = bitrate; /* Take into account overhead from smaller frames. */ - equiv -= (40*channels+20)*(frame_rate - 50); + if (frame_rate > 50) + equiv -= (40*channels+20)*(frame_rate - 50); /* CBR is about a 8% penalty for both SILK and CELT. */ if (!vbr) equiv -= equiv/12; @@ -844,7 +930,7 @@ static opus_int32 compute_equiv_rate(opus_int32 bitrate, int channels, #ifndef DISABLE_FLOAT_API -static int is_digital_silence(const opus_val16* pcm, int frame_size, int channels, int lsb_depth) +int is_digital_silence(const opus_val16* pcm, int frame_size, int channels, int lsb_depth) { int silence = 0; opus_val32 sample_max = 0; @@ -878,7 +964,7 @@ static opus_val32 compute_frame_energy(const opus_val16 *pcm, int frame_size, in /* Compute the right shift required in the MAC to avoid an overflow */ max_shift = celt_ilog2(len); - shift = IMAX(0, (celt_ilog2(sample_max) << 1) + max_shift - 28); + shift = IMAX(0, (celt_ilog2(1+sample_max) << 1) + max_shift - 28); /* Compute the energy */ for (i=0; i NB_SPEECH_FRAMES_BEFORE_DTX*20*2) { - noise_energy = compute_frame_energy(pcm, frame_size, channels, arch); - is_sufficiently_quiet = peak_signal_energy >= (PSEUDO_SNR_THRESHOLD * noise_energy); - } - } - - if (is_silence || (is_noise && is_sufficiently_quiet)) - { - /* The number of consecutive DTX frames should be within the allowed bounds */ - (*nb_no_activity_frames)++; - - if (*nb_no_activity_frames > NB_SPEECH_FRAMES_BEFORE_DTX) - { - if (*nb_no_activity_frames <= (NB_SPEECH_FRAMES_BEFORE_DTX + MAX_CONSECUTIVE_DTX)) + if (*nb_no_activity_ms_Q1 <= (NB_SPEECH_FRAMES_BEFORE_DTX + MAX_CONSECUTIVE_DTX)*20*2) /* Valid frame for DTX! */ return 1; else - (*nb_no_activity_frames) = NB_SPEECH_FRAMES_BEFORE_DTX; + (*nb_no_activity_ms_Q1) = NB_SPEECH_FRAMES_BEFORE_DTX*20*2; } } else - (*nb_no_activity_frames) = 0; + (*nb_no_activity_ms_Q1) = 0; return 0; } #endif -static opus_int32 encode_multiframe_packet(OpusEncoder *st, - const opus_val16 *pcm, - int nb_frames, - int frame_size, - unsigned char *data, - opus_int32 out_data_bytes, - int to_celt, - int lsb_depth, - int float_api) -{ - int i; - int ret = 0; - VARDECL(unsigned char, tmp_data); - int bak_mode, bak_bandwidth, bak_channels, bak_to_mono; - VARDECL(OpusRepacketizer, rp); - int max_header_bytes; - opus_int32 bytes_per_frame; - opus_int32 cbr_bytes; - opus_int32 repacketize_len; - int tmp_len; - ALLOC_STACK; - - /* Worst cases: - * 2 frames: Code 2 with different compressed sizes - * >2 frames: Code 3 VBR */ - max_header_bytes = nb_frames == 2 ? 3 : (2+(nb_frames-1)*2); - - if (st->use_vbr || st->user_bitrate_bps==OPUS_BITRATE_MAX) - repacketize_len = out_data_bytes; - else { - cbr_bytes = 3*st->bitrate_bps/(3*8*st->Fs/(frame_size*nb_frames)); - repacketize_len = IMIN(cbr_bytes, out_data_bytes); - } - bytes_per_frame = IMIN(1276, 1+(repacketize_len-max_header_bytes)/nb_frames); - - ALLOC(tmp_data, nb_frames*bytes_per_frame, unsigned char); - ALLOC(rp, 1, OpusRepacketizer); - opus_repacketizer_init(rp); - - bak_mode = st->user_forced_mode; - bak_bandwidth = st->user_bandwidth; - bak_channels = st->force_channels; - - st->user_forced_mode = st->mode; - st->user_bandwidth = st->bandwidth; - st->force_channels = st->stream_channels; - - bak_to_mono = st->silk_mode.toMono; - if (bak_to_mono) - st->force_channels = 1; - else - st->prev_channels = st->stream_channels; - - for (i=0;isilk_mode.toMono = 0; - st->nonfinal_frame = i<(nb_frames-1); - - /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ - if (to_celt && i==nb_frames-1) - st->user_forced_mode = MODE_CELT_ONLY; - - tmp_len = opus_encode_native(st, pcm+i*(st->channels*frame_size), frame_size, - tmp_data+i*bytes_per_frame, bytes_per_frame, lsb_depth, NULL, 0, 0, 0, 0, - NULL, float_api); - - if (tmp_len<0) - { - RESTORE_STACK; - return OPUS_INTERNAL_ERROR; - } - - ret = opus_repacketizer_cat(rp, tmp_data+i*bytes_per_frame, tmp_len); - - if (ret<0) - { - RESTORE_STACK; - return OPUS_INTERNAL_ERROR; - } - } - - ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr); - - if (ret<0) - { - RESTORE_STACK; - return OPUS_INTERNAL_ERROR; - } - - /* Discard configs that were forced locally for the purpose of repacketization */ - st->user_forced_mode = bak_mode; - st->user_bandwidth = bak_bandwidth; - st->force_channels = bak_channels; - st->silk_mode.toMono = bak_to_mono; - - RESTORE_STACK; - return ret; -} - static int compute_redundancy_bytes(opus_int32 max_data_bytes, opus_int32 bitrate_bps, int frame_rate, int channels) { int redundancy_bytes_cap; @@ -1071,6 +1042,18 @@ static int compute_redundancy_bytes(opus_int32 max_data_bytes, opus_int32 bitrat return redundancy_bytes; } +static opus_int32 opus_encode_frame_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 max_data_bytes, + int float_api, int first_frame, +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + AnalysisInfo *analysis_info, int is_silence, +#endif + int redundancy, int celt_to_silk, int prefill, + opus_int32 equiv_rate, int to_celt); + opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, @@ -1080,28 +1063,17 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ CELTEncoder *celt_enc; int i; int ret=0; - opus_int32 nBytes; - ec_enc enc; - int bytes_target; int prefill=0; - int start_band = 0; int redundancy = 0; - int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */ int celt_to_silk = 0; - VARDECL(opus_val16, pcm_buf); - int nb_compr_bytes; int to_celt = 0; - opus_uint32 redundant_rng = 0; - int cutoff_Hz, hp_freq_smth1; int voice_est; /* Probability of voice in Q7 */ opus_int32 equiv_rate; - int delay_compensation; int frame_rate; opus_int32 max_rate; /* Max bitrate we're allowed to use */ int curr_bandwidth; - opus_val16 HB_gain; opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */ - int total_buffer; + opus_int32 cbr_bytes=-1; opus_val16 stereo_width; const CELTMode *celt_mode; #ifndef DISABLE_FLOAT_API @@ -1110,8 +1082,9 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ int analysis_read_subframe_bak=-1; int is_silence = 0; #endif - VARDECL(opus_val16, tmp_prefill); - +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps; +#endif ALLOC_STACK; max_data_bytes = IMIN(1276, out_data_bytes); @@ -1132,10 +1105,6 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ silk_enc = (char*)st+st->silk_enc_offset; celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); - if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) - delay_compensation = 0; - else - delay_compensation = st->delay_compensation; lsb_depth = IMIN(lsb_depth, st->lsb_depth); @@ -1148,21 +1117,19 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if (st->silk_mode.complexity >= 7 && st->Fs>=16000) #endif { - if (is_digital_silence(pcm, frame_size, st->channels, lsb_depth)) - { - is_silence = 1; - } else { - analysis_read_pos_bak = st->analysis.read_pos; - analysis_read_subframe_bak = st->analysis.read_subframe; - run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size, - c1, c2, analysis_channels, st->Fs, - lsb_depth, downmix, &analysis_info); - } + is_silence = is_digital_silence(pcm, frame_size, st->channels, lsb_depth); + analysis_read_pos_bak = st->analysis.read_pos; + analysis_read_subframe_bak = st->analysis.read_subframe; + run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size, + c1, c2, analysis_channels, st->Fs, + lsb_depth, downmix, &analysis_info); /* Track the peak signal energy */ if (!is_silence && analysis_info.activity_probability > DTX_ACTIVITY_THRESHOLD) st->peak_signal_energy = MAX32(MULT16_32_Q15(QCONST16(0.999f, 15), st->peak_signal_energy), compute_frame_energy(pcm, frame_size, st->channels, st->arch)); + } else if (st->analysis.initialized) { + tonality_analysis_reset(&st->analysis); } #else (void)analysis_pcm; @@ -1184,7 +1151,16 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ { int analysis_bandwidth; if (st->signal_type == OPUS_AUTO) - st->voice_ratio = (int)floor(.5+100*(1-analysis_info.music_prob)); + { + float prob; + if (st->prev_mode == 0) + prob = analysis_info.music_prob; + else if (st->prev_mode == MODE_CELT_ONLY) + prob = analysis_info.music_prob_max; + else + prob = analysis_info.music_prob_min; + st->voice_ratio = (int)floor(.5+100*(1-prob)); + } analysis_bandwidth = analysis_info.bandwidth; if (analysis_bandwidth<=12) @@ -1206,21 +1182,24 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ stereo_width = compute_stereo_width(pcm, frame_size, st->Fs, &st->width_mem); else stereo_width = 0; - total_buffer = delay_compensation; st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes); frame_rate = st->Fs/frame_size; if (!st->use_vbr) { - int cbrBytes; /* Multiply by 12 to make sure the division is exact. */ int frame_rate12 = 12*st->Fs/frame_size; /* We need to make sure that "int" values always fit in 16 bits. */ - cbrBytes = IMIN( (12*st->bitrate_bps/8 + frame_rate12/2)/frame_rate12, max_data_bytes); - st->bitrate_bps = cbrBytes*(opus_int32)frame_rate12*8/12; + cbr_bytes = IMIN( (12*st->bitrate_bps/8 + frame_rate12/2)/frame_rate12, max_data_bytes); + st->bitrate_bps = cbr_bytes*(opus_int32)frame_rate12*8/12; /* Make sure we provide at least one byte to avoid failing. */ - max_data_bytes = IMAX(1, cbrBytes); + max_data_bytes = IMAX(1, cbr_bytes); } +#ifdef ENABLE_DRED + /* Allocate some of the bits to DRED if needed. */ + dred_bitrate_bps = compute_dred_bitrate(st, st->bitrate_bps, frame_size); + st->bitrate_bps -= dred_bitrate_bps; +#endif if (max_data_bytes<3 || st->bitrate_bps < 3*frame_rate*8 || (frame_rate<50 && (max_data_bytes*frame_rate<300 || st->bitrate_bps < 2400))) { @@ -1314,6 +1293,8 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ st->stream_channels = st->force_channels; } else { #ifdef FUZZING + (void)stereo_music_threshold; + (void)stereo_voice_threshold; /* Random mono/stereo decision */ if (st->channels == 2 && (rand()&0x1F)==0) st->stream_channels = 3-st->stream_channels; @@ -1337,6 +1318,14 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ equiv_rate = compute_equiv_rate(st->bitrate_bps, st->stream_channels, st->Fs/frame_size, st->use_vbr, 0, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + /* Allow SILK DTX if DTX is enabled but the generalized DTX cannot be used, + e.g. because of the complexity setting or sample rate. */ +#ifndef DISABLE_FLOAT_API + st->silk_mode.useDTX = st->use_dtx && !(analysis_info.valid || is_silence); +#else + st->silk_mode.useDTX = st->use_dtx; +#endif + /* Mode selection depending on application and signal type */ if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) { @@ -1344,6 +1333,8 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ } else if (st->user_forced_mode == OPUS_AUTO) { #ifdef FUZZING + (void)stereo_width; + (void)mode_thresholds; /* Random mode switching */ if ((rand()&0xF)==0) { @@ -1381,17 +1372,12 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ st->mode = (equiv_rate >= threshold) ? MODE_CELT_ONLY: MODE_SILK_ONLY; - /* When FEC is enabled and there's enough packet loss, use SILK */ - if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4) + /* When FEC is enabled and there's enough packet loss, use SILK. + Unless the FEC is set to 2, in which case we don't switch to SILK if we're confident we have music. */ + if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4 && (st->fec_config != 2 || voice_est > 25)) st->mode = MODE_SILK_ONLY; /* When encoding voice and DTX is enabled but the generalized DTX cannot be used, - because of complexity and sampling frequency settings, switch to SILK DTX and - set the encoder to SILK mode */ -#ifndef DISABLE_FLOAT_API - st->silk_mode.useDTX = st->use_dtx && !(analysis_info.valid || is_silence); -#else - st->silk_mode.useDTX = st->use_dtx; -#endif + use SILK in order to make use of its DTX. */ if (st->silk_mode.useDTX && voice_est > 100) st->mode = MODE_SILK_ONLY; #endif @@ -1486,6 +1472,10 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if (equiv_rate >= threshold) break; } while (--bandwidth>OPUS_BANDWIDTH_NARROWBAND); + /* We don't use mediumband anymore, except when explicitly requested or during + mode transitions. */ + if (bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + bandwidth = OPUS_BANDWIDTH_WIDEBAND; st->bandwidth = st->auto_bandwidth = bandwidth; /* Prevents any transition to SWB/FB until the SILK layer has fully switched to WB mode and turned the variable LP filter off */ @@ -1563,6 +1553,15 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ { int enc_frame_size; int nb_frames; + VARDECL(unsigned char, tmp_data); + VARDECL(OpusRepacketizer, rp); + int max_header_bytes; + opus_int32 repacketize_len; + opus_int32 max_len_sum; + opus_int32 tot_size=0; + unsigned char *curr_data; + int tmp_len; + int dtx_count = 0; if (st->mode == MODE_SILK_ONLY) { @@ -1581,17 +1580,186 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ #ifndef DISABLE_FLOAT_API if (analysis_read_pos_bak!= -1) { + /* Reset analysis position to the beginning of the first frame so we + can use it one frame at a time. */ st->analysis.read_pos = analysis_read_pos_bak; st->analysis.read_subframe = analysis_read_subframe_bak; } #endif - ret = encode_multiframe_packet(st, pcm, nb_frames, enc_frame_size, data, - out_data_bytes, to_celt, lsb_depth, float_api); + /* Worst cases: + * 2 frames: Code 2 with different compressed sizes + * >2 frames: Code 3 VBR */ + max_header_bytes = nb_frames == 2 ? 3 : (2+(nb_frames-1)*2); + if (st->use_vbr || st->user_bitrate_bps==OPUS_BITRATE_MAX) + repacketize_len = out_data_bytes; + else { + celt_assert(cbr_bytes>=0); + repacketize_len = IMIN(cbr_bytes, out_data_bytes); + } + max_len_sum = nb_frames + repacketize_len - max_header_bytes; + + ALLOC(tmp_data, max_len_sum, unsigned char); + curr_data = tmp_data; + ALLOC(rp, 1, OpusRepacketizer); + opus_repacketizer_init(rp); + + + int bak_to_mono = st->silk_mode.toMono; + if (bak_to_mono) + st->force_channels = 1; + else + st->prev_channels = st->stream_channels; + + for (i=0;isilk_mode.toMono = 0; + st->nonfinal_frame = i<(nb_frames-1); + + /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ + frame_to_celt = to_celt && i==nb_frames-1; + frame_redundancy = redundancy && (frame_to_celt || (!to_celt && i==0)); + + curr_max = IMIN(3*st->bitrate_bps/(3*8*st->Fs/enc_frame_size), max_len_sum/nb_frames); +#ifdef ENABLE_DRED + curr_max = IMIN(curr_max, (max_len_sum-3*dred_bitrate_bps/(3*8*st->Fs/frame_size))/nb_frames); + if (first_frame) curr_max += 3*dred_bitrate_bps/(3*8*st->Fs/frame_size); +#endif + curr_max = IMIN(max_len_sum-tot_size, curr_max); +#ifndef DISABLE_FLOAT_API + if (analysis_read_pos_bak != -1) { + is_silence = is_digital_silence(pcm, frame_size, st->channels, lsb_depth); + /* Get analysis for current frame. */ + tonality_get_info(&st->analysis, &analysis_info, enc_frame_size); + } +#endif + + tmp_len = opus_encode_frame_native(st, pcm+i*(st->channels*enc_frame_size), enc_frame_size, curr_data, curr_max, float_api, first_frame, +#ifdef ENABLE_DRED + dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + &analysis_info, + is_silence, +#endif + frame_redundancy, celt_to_silk, prefill, + equiv_rate, frame_to_celt + ); + if (tmp_len<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } else if (tmp_len==1) { + dtx_count++; + } + ret = opus_repacketizer_cat(rp, curr_data, tmp_len); + + if (ret<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + tot_size += tmp_len; + curr_data += tmp_len; + } + ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr && (dtx_count != nb_frames), NULL, 0); + if (ret<0) + { + ret = OPUS_INTERNAL_ERROR; + } + st->silk_mode.toMono = bak_to_mono; RESTORE_STACK; return ret; + } else { + ret = opus_encode_frame_native(st, pcm, frame_size, data, max_data_bytes, float_api, 1, +#ifdef ENABLE_DRED + dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + &analysis_info, + is_silence, +#endif + redundancy, celt_to_silk, prefill, + equiv_rate, to_celt + ); + RESTORE_STACK; + return ret; } +} + +static opus_int32 opus_encode_frame_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 max_data_bytes, + int float_api, int first_frame, +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + AnalysisInfo *analysis_info, int is_silence, +#endif + int redundancy, int celt_to_silk, int prefill, + opus_int32 equiv_rate, int to_celt) +{ + void *silk_enc; + CELTEncoder *celt_enc; + const CELTMode *celt_mode; + int i; + int ret=0; + opus_int32 nBytes; + ec_enc enc; + int bytes_target; + int start_band = 0; + int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */ + int nb_compr_bytes; + opus_uint32 redundant_rng = 0; + int cutoff_Hz; + int hp_freq_smth1; + opus_val16 HB_gain; + int apply_padding; + int frame_rate; + int curr_bandwidth; + int delay_compensation; + int total_buffer; + opus_int activity = VAD_NO_DECISION; + VARDECL(opus_val16, pcm_buf); + VARDECL(opus_val16, tmp_prefill); + SAVE_STACK; + + st->rangeFinal = 0; + silk_enc = (char*)st+st->silk_enc_offset; + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); + curr_bandwidth = st->bandwidth; + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + total_buffer = delay_compensation; + + frame_rate = st->Fs/frame_size; + +#ifndef DISABLE_FLOAT_API + if (is_silence) + { + activity = !is_silence; + } else if (analysis_info->valid) + { + activity = analysis_info->activity_probability >= DTX_ACTIVITY_THRESHOLD; + if (!activity) + { + /* Mark as active if this noise frame is sufficiently loud */ + opus_val32 noise_energy = compute_frame_energy(pcm, frame_size, st->channels, st->arch); + activity = st->peak_signal_energy < (PSEUDO_SNR_THRESHOLD * noise_energy); + } + } +#endif /* For the first frame at a new SILK bandwidth */ if (st->silk_bw_switch) @@ -1599,7 +1767,8 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ redundancy = 1; celt_to_silk = 1; st->silk_bw_switch = 0; - prefill=1; + /* Do a prefill without resetting the sampling rate control. */ + prefill=2; } /* If we decided to go with CELT, make sure redundancy is off, no matter what @@ -1638,6 +1807,25 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if (st->application == OPUS_APPLICATION_VOIP) { hp_cutoff(pcm, cutoff_Hz, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs, st->arch); + +#ifdef ENABLE_OSCE_TRAINING_DATA + /* write out high pass filtered clean signal*/ + static FILE *fout =NULL; + if (fout == NULL) + { + fout = fopen("clean_hp.s16", "wb"); + } + + { + int idx; + opus_int16 tmp; + for (idx = 0; idx < frame_size; idx++) + { + tmp = (opus_int16) (32768 * pcm_buf[total_buffer + idx] + 0.5f); + fwrite(&tmp, sizeof(tmp), 1, fout); + } + } +#endif } else { dc_reject(pcm, 3, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); } @@ -1654,8 +1842,24 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ st->hp_mem[0] = st->hp_mem[1] = st->hp_mem[2] = st->hp_mem[3] = 0; } } +#else + (void)float_api; #endif +#ifdef ENABLE_DRED + if ( st->dred_duration > 0 && st->dred_encoder.loaded ) { + int frame_size_400Hz; + /* DRED Encoder */ + dred_compute_latents( &st->dred_encoder, &pcm_buf[total_buffer*st->channels], frame_size, total_buffer, st->arch ); + frame_size_400Hz = frame_size*400/st->Fs; + OPUS_MOVE(&st->activity_mem[frame_size_400Hz], st->activity_mem, 4*DRED_MAX_FRAMES-frame_size_400Hz); + for (i=0;iactivity_mem[i] = activity; + } else { + st->dred_encoder.latents_buffer_fill = 0; + OPUS_CLEAR(st->activity_mem, DRED_MAX_FRAMES); + } +#endif /* SILK processing */ HB_gain = Q15ONE; @@ -1674,7 +1878,8 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if( st->mode == MODE_HYBRID ) { /* Base rate for SILK */ st->silk_mode.bitRate = compute_silk_rate_for_hybrid(total_bitRate, - curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded); + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); if (!st->energy_masking) { /* Increasingly attenuate high band when it gets allocated fewer bits */ @@ -1736,7 +1941,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ } else if (curr_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { st->silk_mode.desiredInternalSampleRate = 12000; } else { - silk_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND ); + celt_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND ); st->silk_mode.desiredInternalSampleRate = 16000; } if( st->mode == MODE_HYBRID ) { @@ -1749,7 +1954,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ st->silk_mode.maxInternalSampleRate = 16000; if (st->mode == MODE_SILK_ONLY) { - opus_int32 effective_max_rate = max_rate; + opus_int32 effective_max_rate = frame_rate*max_data_bytes*8; if (frame_rate > 50) effective_max_rate = effective_max_rate*2/3; if (effective_max_rate < 8000) @@ -1779,9 +1984,19 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ } if (st->silk_mode.useCBR) { + /* When we're in CBR mode, but we have non-SILK data to encode, switch SILK to VBR with cap to + save on complexity. Any variations will be absorbed by CELT and/or DRED and we can still + produce a constant bitrate without wasting bits. */ +#ifdef ENABLE_DRED + if (st->mode == MODE_HYBRID || dred_bitrate_bps > 0) +#else if (st->mode == MODE_HYBRID) +#endif { - st->silk_mode.maxBits = IMIN(st->silk_mode.maxBits, st->silk_mode.bitRate * frame_size / st->Fs); + /* Allow SILK to steal up to 25% of the remaining bits */ + opus_int16 other_bits = IMAX(0, st->silk_mode.maxBits - st->silk_mode.bitRate * frame_size / st->Fs); + st->silk_mode.maxBits = IMAX(0, st->silk_mode.maxBits - other_bits*3/4); + st->silk_mode.useCBR = 0; } } else { /* Constrained VBR. */ @@ -1789,7 +2004,8 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ { /* Compute SILK bitrate corresponding to the max total bits available */ opus_int32 maxBitRate = compute_silk_rate_for_hybrid(st->silk_mode.maxBits*st->Fs / frame_size, - curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded); + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); st->silk_mode.maxBits = maxBitRate * frame_size / st->Fs; } } @@ -1814,7 +2030,9 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ for (i=0;iencoder_buffer*st->channels;i++) pcm_silk[i] = FLOAT2INT16(st->delay_buffer[i]); #endif - silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, 1 ); + silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, prefill, activity ); + /* Prevent a second switch in the real encode call. */ + st->silk_mode.opusCanSwitch = 0; } #ifdef FIXED_POINT @@ -1823,7 +2041,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ for (i=0;ichannels;i++) pcm_silk[i] = FLOAT2INT16(pcm_buf[total_buffer*st->channels + i]); #endif - ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0 ); + ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0, activity ); if( ret ) { /*fprintf (stderr, "SILK encode error: %d\n", ret);*/ /* Handle error */ @@ -1841,7 +2059,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND; } } else { - silk_assert( st->silk_mode.internalSampleRate == 16000 ); + celt_assert( st->silk_mode.internalSampleRate == 16000 ); } st->silk_mode.opusCanSwitch = st->silk_mode.switchReady && !st->nonfinal_frame; @@ -1891,26 +2109,10 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if (st->mode != MODE_SILK_ONLY) { opus_val32 celt_pred=2; - celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); /* We may still decide to disable prediction later */ if (st->silk_mode.reducedDependency) celt_pred = 0; celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(celt_pred)); - - if (st->mode == MODE_HYBRID) - { - if( st->use_vbr ) { - celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); - celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); - } - } else { - if (st->use_vbr) - { - celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); - celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); - celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); - } - } } ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_val16); @@ -1936,7 +2138,14 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ } st->prev_HB_gain = HB_gain; if (st->mode != MODE_HYBRID || st->stream_channels==1) - st->silk_mode.stereoWidth_Q14 = IMIN((1<<14),2*IMAX(0,equiv_rate-24000)); + { + if (equiv_rate > 32000) + st->silk_mode.stereoWidth_Q14 = 16384; + else if (equiv_rate < 16000) + st->silk_mode.stereoWidth_Q14 = 0; + else + st->silk_mode.stereoWidth_Q14 = 16384 - 2048*(opus_int32)(32000-equiv_rate)/(equiv_rate-14000); + } if( !st->energy_masking && st->channels == 2 ) { /* Apply stereo width reduction (at low bitrates) */ if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) { @@ -1997,21 +2206,33 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ ec_enc_done(&enc); nb_compr_bytes = ret; } else { - nb_compr_bytes = (max_data_bytes-1)-redundancy_bytes; - ec_enc_shrink(&enc, nb_compr_bytes); + nb_compr_bytes = (max_data_bytes-1)-redundancy_bytes; +#ifdef ENABLE_DRED + if (st->dred_duration > 0) + { + int max_celt_bytes; + opus_int32 dred_bytes = dred_bitrate_bps/(frame_rate*8); + /* Allow CELT to steal up to 25% of the remaining bits. */ + max_celt_bytes = nb_compr_bytes - dred_bytes*3/4; + /* But try to give CELT at least 5 bytes to prevent a mismatch with + the redundancy signaling. */ + max_celt_bytes = IMAX((ec_tell(&enc)+7)/8 + 5, max_celt_bytes); + /* Subject to the original max. */ + nb_compr_bytes = IMIN(nb_compr_bytes, max_celt_bytes); + } +#endif + ec_enc_shrink(&enc, nb_compr_bytes); } #ifndef DISABLE_FLOAT_API if (redundancy || st->mode != MODE_SILK_ONLY) - celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(&analysis_info)); + celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(analysis_info)); #endif if (st->mode == MODE_HYBRID) { SILKInfo info; info.signalType = st->silk_mode.signalType; info.offset = st->silk_mode.offset; celt_encoder_ctl(celt_enc, CELT_SET_SILK_INFO(&info)); - } else { - celt_encoder_ctl(celt_enc, CELT_SET_SILK_INFO((SILKInfo*)NULL)); } /* 5 ms redundant frame for CELT->SILK */ @@ -2035,6 +2256,34 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ if (st->mode != MODE_SILK_ONLY) { + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(st->use_vbr)); + if (st->mode == MODE_HYBRID) + { + if( st->use_vbr ) { + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); + } + } else { + if (st->use_vbr) + { + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); + } + } +#ifdef ENABLE_DRED + /* When Using DRED CBR, we can actually make the CELT part VBR and have DRED pick up the slack. */ + if (!st->use_vbr && st->dred_duration > 0) + { + opus_int32 celt_bitrate = st->bitrate_bps; + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); + if (st->mode == MODE_HYBRID) { + celt_bitrate -= st->silk_mode.bitRate; + } + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(celt_bitrate)); + } +#endif if (st->mode != st->prev_mode && st->prev_mode > 0) { unsigned char dummy[2]; @@ -2047,10 +2296,6 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ /* If false, we already busted the budget and we'll end up with a "PLC frame" */ if (ec_tell(&enc) <= 8*nb_compr_bytes) { - /* Set the bitrate again if it was overridden in the redundancy code above*/ - if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && st->use_vbr) - celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); - celt_encoder_ctl(celt_enc, OPUS_SET_VBR(st->use_vbr)); ret = celt_encode_with_ec(celt_enc, pcm_buf, frame_size, NULL, nb_compr_bytes, &enc); if (ret < 0) { @@ -2058,10 +2303,10 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ return OPUS_INTERNAL_ERROR; } /* Put CELT->SILK redundancy data in the right place. */ - if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && st->use_vbr) + if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && nb_compr_bytes != ret) { OPUS_MOVE(data+ret, data+nb_compr_bytes, redundancy_bytes); - nb_compr_bytes = nb_compr_bytes+redundancy_bytes; + nb_compr_bytes = ret+redundancy_bytes; } } } @@ -2118,16 +2363,17 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ /* DTX decision */ #ifndef DISABLE_FLOAT_API - if (st->use_dtx && (analysis_info.valid || is_silence)) + if (st->use_dtx && (analysis_info->valid || is_silence)) { - if (decide_dtx_mode(analysis_info.activity_probability, &st->nb_no_activity_frames, - st->peak_signal_energy, pcm, frame_size, st->channels, is_silence, st->arch)) + if (decide_dtx_mode(activity, &st->nb_no_activity_ms_Q1, 2*1000*frame_size/st->Fs)) { st->rangeFinal = 0; data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); RESTORE_STACK; return 1; } + } else { + st->nb_no_activity_ms_Q1 = 0; } #endif @@ -2155,7 +2401,51 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ } /* Count ToC and redundancy */ ret += 1+redundancy_bytes; - if (!st->use_vbr) + apply_padding = !st->use_vbr; +#ifdef ENABLE_DRED + if (st->dred_duration > 0 && st->dred_encoder.loaded && first_frame) { + opus_extension_data extension; + unsigned char buf[DRED_MAX_DATA_SIZE]; + int dred_chunks; + int dred_bytes_left; + dred_chunks = IMIN((st->dred_duration+5)/4, DRED_NUM_REDUNDANCY_FRAMES/2); + if (st->use_vbr) dred_chunks = IMIN(dred_chunks, st->dred_target_chunks); + /* Remaining space for DRED, accounting for cost the 3 extra bytes for code 3, padding length, and extension number. */ + dred_bytes_left = IMIN(DRED_MAX_DATA_SIZE, max_data_bytes-ret-3); + /* Account for the extra bytes required to signal large padding length. */ + dred_bytes_left -= (dred_bytes_left+1+DRED_EXPERIMENTAL_BYTES)/255; + /* Check whether we actually have something to encode. */ + if (dred_chunks >= 1 && dred_bytes_left >= DRED_MIN_BYTES+DRED_EXPERIMENTAL_BYTES) { + int dred_bytes; +#ifdef DRED_EXPERIMENTAL_VERSION + /* Add temporary extension type and version. + These bytes will be removed once extension is finalized. */ + buf[0] = 'D'; + buf[1] = DRED_EXPERIMENTAL_VERSION; +#endif + dred_bytes = dred_encode_silk_frame(&st->dred_encoder, buf+DRED_EXPERIMENTAL_BYTES, dred_chunks, dred_bytes_left-DRED_EXPERIMENTAL_BYTES, + st->dred_q0, st->dred_dQ, st->dred_qmax, st->activity_mem, st->arch); + if (dred_bytes > 0) { + dred_bytes += DRED_EXPERIMENTAL_BYTES; + celt_assert(dred_bytes <= dred_bytes_left); + extension.id = DRED_EXTENSION_ID; + extension.frame = 0; + extension.data = buf; + extension.len = dred_bytes; + ret = opus_packet_pad_impl(data, ret, max_data_bytes, !st->use_vbr, &extension, 1); + if (ret < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + apply_padding = 0; + } + } + } +#else + (void)first_frame; /* Avoids a warning about first_frame being unused. */ +#endif + if (apply_padding) { if (opus_packet_pad(data, ret, max_data_bytes) != OPUS_OK) { @@ -2422,11 +2712,12 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_SET_INBAND_FEC_REQUEST: { opus_int32 value = va_arg(ap, opus_int32); - if(value<0 || value>1) + if(value<0 || value>2) { goto bad_arg; } - st->silk_mode.useInBandFEC = value; + st->fec_config = value; + st->silk_mode.useInBandFEC = (value != 0); } break; case OPUS_GET_INBAND_FEC_REQUEST: @@ -2436,7 +2727,7 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { goto bad_arg; } - *value = st->silk_mode.useInBandFEC; + *value = st->fec_config; } break; case OPUS_SET_PACKET_LOSS_PERC_REQUEST: @@ -2605,7 +2896,6 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) goto bad_arg; } st->variable_duration = value; - celt_encoder_ctl(celt_enc, OPUS_SET_EXPERT_FRAME_DURATION(value)); } break; case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: @@ -2654,6 +2944,29 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) celt_encoder_ctl(celt_enc, OPUS_GET_PHASE_INVERSION_DISABLED(value)); } break; +#ifdef ENABLE_DRED + case OPUS_SET_DRED_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>DRED_MAX_FRAMES) + { + goto bad_arg; + } + st->dred_duration = value; + st->silk_mode.useDRED = !!value; + } + break; + case OPUS_GET_DRED_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->dred_duration; + } + break; +#endif case OPUS_RESET_STATE: { void *silk_enc; @@ -2669,6 +2982,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); silk_InitEncoder( silk_enc, st->arch, &dummy ); +#ifdef ENABLE_DRED + /* Initialize DRED Encoder */ + dred_encoder_reset( &st->dred_encoder ); +#endif st->stream_channels = st->channels; st->hybrid_stereo_width_Q14 = 1 << 14; st->prev_HB_gain = Q15ONE; @@ -2702,7 +3019,48 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) ret = celt_encoder_ctl(celt_enc, OPUS_SET_ENERGY_MASK(value)); } break; - + case OPUS_GET_IN_DTX_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + if (st->silk_mode.useDTX && (st->prev_mode == MODE_SILK_ONLY || st->prev_mode == MODE_HYBRID)) { + /* DTX determined by Silk. */ + silk_encoder *silk_enc = (silk_encoder*)(void *)((char*)st+st->silk_enc_offset); + *value = silk_enc->state_Fxx[0].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + /* Stereo: check second channel unless only the middle channel was encoded. */ + if(*value == 1 && st->silk_mode.nChannelsInternal == 2 && silk_enc->prev_decode_only_middle == 0) { + *value = silk_enc->state_Fxx[1].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + } + } +#ifndef DISABLE_FLOAT_API + else if (st->use_dtx) { + /* DTX determined by Opus. */ + *value = st->nb_no_activity_ms_Q1 >= NB_SPEECH_FRAMES_BEFORE_DTX*20*2; + } +#endif + else { + *value = 0; + } + } + break; +#ifdef USE_WEIGHTS_FILE + case OPUS_SET_DNN_BLOB_REQUEST: + { + const unsigned char *data = va_arg(ap, const unsigned char *); + opus_int32 len = va_arg(ap, opus_int32); + if(len<0 || data == NULL) + { + goto bad_arg; + } +#ifdef ENABLE_DRED + ret = dred_encoder_load_model(&st->dred_encoder, data, len); +#endif + } + break; +#endif case CELT_GET_MODE_REQUEST: { const CELTMode ** value = va_arg(ap, const CELTMode**); diff --git a/code/opus-1.2.1/src/opus_multistream.c b/code/opus-1.5.2/src/opus_multistream.c similarity index 100% rename from code/opus-1.2.1/src/opus_multistream.c rename to code/opus-1.5.2/src/opus_multistream.c diff --git a/code/opus-1.2.1/src/opus_multistream_decoder.c b/code/opus-1.5.2/src/opus_multistream_decoder.c similarity index 91% rename from code/opus-1.2.1/src/opus_multistream_decoder.c rename to code/opus-1.5.2/src/opus_multistream_decoder.c index e421726f..4ae877a7 100644 --- a/code/opus-1.2.1/src/opus_multistream_decoder.c +++ b/code/opus-1.5.2/src/opus_multistream_decoder.c @@ -37,16 +37,19 @@ #include "float_cast.h" #include "os_support.h" -struct OpusMSDecoder { - ChannelLayout layout; - /* Decoder states go here */ -}; - - - - /* DECODER */ +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +static void validate_ms_decoder(OpusMSDecoder *st) +{ + validate_layout(&st->layout); +} +#define VALIDATE_MS_DECODER(st) validate_ms_decoder(st) +#else +#define VALIDATE_MS_DECODER(st) +#endif + + opus_int32 opus_multistream_decoder_get_size(int nb_streams, int nb_coupled_streams) { int coupled_size; @@ -143,15 +146,6 @@ OpusMSDecoder *opus_multistream_decoder_create( return st; } -typedef void (*opus_copy_channel_out_func)( - void *dst, - int dst_stride, - int dst_channel, - const opus_val16 *src, - int src_stride, - int frame_size -); - static int opus_multistream_packet_validate(const unsigned char *data, opus_int32 len, int nb_streams, opus_int32 Fs) { @@ -168,7 +162,7 @@ static int opus_multistream_packet_validate(const unsigned char *data, if (len<=0) return OPUS_INVALID_PACKET; count = opus_packet_parse_impl(data, len, s!=nb_streams-1, &toc, NULL, - size, NULL, &packet_offset); + size, NULL, &packet_offset, NULL, NULL); if (count<0) return count; tmp_samples = opus_packet_get_nb_samples(data, packet_offset, Fs); @@ -181,7 +175,7 @@ static int opus_multistream_packet_validate(const unsigned char *data, return samples; } -static int opus_multistream_decode_native( +int opus_multistream_decode_native( OpusMSDecoder *st, const unsigned char *data, opus_int32 len, @@ -189,7 +183,8 @@ static int opus_multistream_decode_native( opus_copy_channel_out_func copy_channel_out, int frame_size, int decode_fec, - int soft_clip + int soft_clip, + void *user_data ) { opus_int32 Fs; @@ -201,8 +196,14 @@ static int opus_multistream_decode_native( VARDECL(opus_val16, buf); ALLOC_STACK; + VALIDATE_MS_DECODER(st); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } /* Limit frame_size to avoid excessive stack allocations. */ - opus_multistream_decoder_ctl(st, OPUS_GET_SAMPLE_RATE(&Fs)); + MUST_SUCCEED(opus_multistream_decoder_ctl(st, OPUS_GET_SAMPLE_RATE(&Fs))); frame_size = IMIN(frame_size, Fs/25*3); ALLOC(buf, 2*frame_size, opus_val16); ptr = (char*)st + align(sizeof(OpusMSDecoder)); @@ -249,9 +250,12 @@ static int opus_multistream_decode_native( return OPUS_INTERNAL_ERROR; } packet_offset = 0; - ret = opus_decode_native(dec, data, len, buf, frame_size, decode_fec, s!=st->layout.nb_streams-1, &packet_offset, soft_clip); - data += packet_offset; - len -= packet_offset; + ret = opus_decode_native(dec, data, len, buf, frame_size, decode_fec, s!=st->layout.nb_streams-1, &packet_offset, soft_clip, NULL, 0); + if (!do_plc) + { + data += packet_offset; + len -= packet_offset; + } if (ret <= 0) { RESTORE_STACK; @@ -266,7 +270,7 @@ static int opus_multistream_decode_native( while ( (chan = get_left_channel(&st->layout, s, prev)) != -1) { (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf, 2, frame_size); + buf, 2, frame_size, user_data); prev = chan; } prev = -1; @@ -274,7 +278,7 @@ static int opus_multistream_decode_native( while ( (chan = get_right_channel(&st->layout, s, prev)) != -1) { (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf+1, 2, frame_size); + buf+1, 2, frame_size, user_data); prev = chan; } } else { @@ -284,7 +288,7 @@ static int opus_multistream_decode_native( while ( (chan = get_mono_channel(&st->layout, s, prev)) != -1) { (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf, 1, frame_size); + buf, 1, frame_size, user_data); prev = chan; } } @@ -295,7 +299,7 @@ static int opus_multistream_decode_native( if (st->layout.mapping[c] == 255) { (*copy_channel_out)(pcm, st->layout.nb_channels, c, - NULL, 0, frame_size); + NULL, 0, frame_size, user_data); } } RESTORE_STACK; @@ -309,11 +313,13 @@ static void opus_copy_channel_out_float( int dst_channel, const opus_val16 *src, int src_stride, - int frame_size + int frame_size, + void *user_data ) { float *float_dst; opus_int32 i; + (void)user_data; float_dst = (float*)dst; if (src != NULL) { @@ -338,11 +344,13 @@ static void opus_copy_channel_out_short( int dst_channel, const opus_val16 *src, int src_stride, - int frame_size + int frame_size, + void *user_data ) { opus_int16 *short_dst; opus_int32 i; + (void)user_data; short_dst = (opus_int16*)dst; if (src != NULL) { @@ -373,7 +381,7 @@ int opus_multistream_decode( ) { return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_short, frame_size, decode_fec, 0); + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 0, NULL); } #ifndef DISABLE_FLOAT_API @@ -381,7 +389,7 @@ int opus_multistream_decode_float(OpusMSDecoder *st, const unsigned char *data, opus_int32 len, float *pcm, int frame_size, int decode_fec) { return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0); + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0, NULL); } #endif @@ -391,32 +399,30 @@ int opus_multistream_decode(OpusMSDecoder *st, const unsigned char *data, opus_int32 len, opus_int16 *pcm, int frame_size, int decode_fec) { return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_short, frame_size, decode_fec, 1); + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 1, NULL); } int opus_multistream_decode_float( OpusMSDecoder *st, const unsigned char *data, opus_int32 len, - float *pcm, + opus_val16 *pcm, int frame_size, int decode_fec ) { return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0); + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0, NULL); } #endif -int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) +int opus_multistream_decoder_ctl_va_list(OpusMSDecoder *st, int request, + va_list ap) { - va_list ap; int coupled_size, mono_size; char *ptr; int ret = OPUS_OK; - va_start(ap, request); - coupled_size = opus_decoder_get_size(2); mono_size = opus_decoder_get_size(1); ptr = (char*)st + align(sizeof(OpusMSDecoder)); @@ -484,7 +490,7 @@ int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) OpusDecoder **value; stream_id = va_arg(ap, opus_int32); if (stream_id<0 || stream_id >= st->layout.nb_streams) - ret = OPUS_BAD_ARG; + goto bad_arg; value = va_arg(ap, OpusDecoder**); if (!value) { @@ -525,14 +531,20 @@ int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) ret = OPUS_UNIMPLEMENTED; break; } - - va_end(ap); return ret; bad_arg: - va_end(ap); return OPUS_BAD_ARG; } +int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) +{ + int ret; + va_list ap; + va_start(ap, request); + ret = opus_multistream_decoder_ctl_va_list(st, request, ap); + va_end(ap); + return ret; +} void opus_multistream_decoder_destroy(OpusMSDecoder *st) { diff --git a/code/opus-1.2.1/src/opus_multistream_encoder.c b/code/opus-1.5.2/src/opus_multistream_encoder.c similarity index 90% rename from code/opus-1.2.1/src/opus_multistream_encoder.c rename to code/opus-1.5.2/src/opus_multistream_encoder.c index 032fc003..1725ade7 100644 --- a/code/opus-1.2.1/src/opus_multistream_encoder.c +++ b/code/opus-1.5.2/src/opus_multistream_encoder.c @@ -61,37 +61,6 @@ static const VorbisLayout vorbis_mappings[8] = { {5, 3, {0, 6, 1, 2, 3, 4, 5, 7}}, /* 8: 7.1 surround */ }; -typedef void (*opus_copy_channel_in_func)( - opus_val16 *dst, - int dst_stride, - const void *src, - int src_stride, - int src_channel, - int frame_size -); - -typedef enum { - MAPPING_TYPE_NONE, - MAPPING_TYPE_SURROUND -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS - , /* Do not include comma at end of enumerator list */ - MAPPING_TYPE_AMBISONICS -#endif -} MappingType; - -struct OpusMSEncoder { - ChannelLayout layout; - int arch; - int lfe_stream; - int application; - int variable_duration; - MappingType mapping_type; - opus_int32 bitrate_bps; - /* Encoder states go here */ - /* then opus_val32 window_mem[channels*120]; */ - /* then opus_val32 preemph_mem[channels]; */ -}; - static opus_val32 *ms_get_preemph_mem(OpusMSEncoder *st) { int s; @@ -132,19 +101,20 @@ static opus_val32 *ms_get_window_mem(OpusMSEncoder *st) return (opus_val32*)(void*)ptr; } -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS static int validate_ambisonics(int nb_channels, int *nb_streams, int *nb_coupled_streams) { int order_plus_one; int acn_channels; int nondiegetic_channels; + if (nb_channels < 1 || nb_channels > 227) + return 0; + order_plus_one = isqrt32(nb_channels); acn_channels = order_plus_one * order_plus_one; nondiegetic_channels = nb_channels - acn_channels; - if (order_plus_one < 1 || order_plus_one > 15 || - (nondiegetic_channels != 0 && nondiegetic_channels != 2)) + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) return 0; if (nb_streams) @@ -153,7 +123,6 @@ static int validate_ambisonics(int nb_channels, int *nb_streams, int *nb_coupled *nb_coupled_streams = nondiegetic_channels != 0; return 1; } -#endif static int validate_encoder_layout(const ChannelLayout *layout) { @@ -296,7 +265,7 @@ void surround_analysis(const CELTMode *celt_mode, const void *pcm, opus_val16 *b int nb_frames = frame_size/freq_size; celt_assert(nb_frames*freq_size == frame_size); OPUS_COPY(in, mem+c*overlap, overlap); - (*copy_channel_in)(x, 1, pcm, channels, c, len); + (*copy_channel_in)(x, 1, pcm, channels, c, len, NULL); celt_preemphasis(x, in+overlap, frame_size, 1, upsample, celt_mode->preemph, preemph_mem+c, 0); #ifndef FIXED_POINT { @@ -443,12 +412,10 @@ opus_int32 opus_multistream_surround_encoder_get_size(int channels, int mapping_ { nb_streams=channels; nb_coupled_streams=0; -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS - } else if (mapping_family==254) + } else if (mapping_family==2) { if (!validate_ambisonics(channels, &nb_streams, &nb_coupled_streams)) return 0; -#endif } else return 0; size = opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); @@ -476,7 +443,8 @@ static int opus_multistream_encoder_init_impl( char *ptr; if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams)) + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) return OPUS_BAD_ARG; st->arch = opus_select_arch(); @@ -492,14 +460,11 @@ static int opus_multistream_encoder_init_impl( st->layout.mapping[i] = mapping[i]; if (!validate_layout(&st->layout)) return OPUS_BAD_ARG; - if (mapping_type == MAPPING_TYPE_SURROUND && - !validate_encoder_layout(&st->layout)) + if (!validate_encoder_layout(&st->layout)) return OPUS_BAD_ARG; -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS if (mapping_type == MAPPING_TYPE_AMBISONICS && !validate_ambisonics(st->layout.nb_channels, NULL, NULL)) return OPUS_BAD_ARG; -#endif ptr = (char*)st + align(sizeof(OpusMSEncoder)); coupled_size = opus_encoder_get_size(2); mono_size = opus_encoder_get_size(1); @@ -591,8 +556,7 @@ int opus_multistream_surround_encoder_init( *coupled_streams=0; for(i=0;i2 && mapping_family==1) { mapping_type = MAPPING_TYPE_SURROUND; -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS - } else if (mapping_family==254) + } else if (mapping_family==2) { mapping_type = MAPPING_TYPE_AMBISONICS; -#endif } else { mapping_type = MAPPING_TYPE_NONE; @@ -634,7 +595,8 @@ OpusMSEncoder *opus_multistream_encoder_create( int ret; OpusMSEncoder *st; if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams)) + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) { if (error) *error = OPUS_BAD_ARG; @@ -772,7 +734,6 @@ static void surround_rate_allocation( } } -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS static void ambisonics_rate_allocation( OpusMSEncoder *st, opus_int32 *rate, @@ -781,24 +742,15 @@ static void ambisonics_rate_allocation( ) { int i; - int total_rate; - int directional_rate; - int nondirectional_rate; - int leftover_bits; + opus_int32 total_rate; + opus_int32 per_stream_rate; - /* Each nondirectional channel gets (rate_ratio_num / rate_ratio_den) times - * as many bits as all other ambisonics channels. - */ - const int rate_ratio_num = 4; - const int rate_ratio_den = 3; const int nb_channels = st->layout.nb_streams + st->layout.nb_coupled_streams; - const int nb_nondirectional_channels = st->layout.nb_coupled_streams * 2 + 1; - const int nb_directional_channels = st->layout.nb_streams - 1; if (st->bitrate_bps==OPUS_AUTO) { total_rate = (st->layout.nb_coupled_streams + st->layout.nb_streams) * - (Fs+60*Fs/frame_size) + st->layout.nb_streams * 15000; + (Fs+60*Fs/frame_size) + st->layout.nb_streams * (opus_int32)15000; } else if (st->bitrate_bps==OPUS_BITRATE_MAX) { total_rate = nb_channels * 320000; @@ -807,49 +759,14 @@ static void ambisonics_rate_allocation( total_rate = st->bitrate_bps; } - /* Let y be the directional rate, m be the num of nondirectional channels - * m = (s + 1) - * and let p, q be integers such that the nondirectional rate is - * m_rate = (p / q) * y - * Also let T be the total bitrate to allocate. Then - * T = (n - m) * y + m * m_rate - * Solving for y, - * y = (q * T) / (m * (p - q) + n * q) - */ - directional_rate = - total_rate * rate_ratio_den - / (nb_nondirectional_channels * (rate_ratio_num - rate_ratio_den) - + nb_channels * rate_ratio_den); - - /* Calculate the nondirectional rate. - * m_rate = y * (p / q) - */ - nondirectional_rate = directional_rate * rate_ratio_num / rate_ratio_den; - - /* Calculate the leftover from truncation error. - * leftover = T - y * (n - m) - m_rate * m - * Place leftover bits in omnidirectional channel. - */ - leftover_bits = total_rate - - directional_rate * nb_directional_channels - - nondirectional_rate * nb_nondirectional_channels; - - /* Calculate rates for each channel */ + /* Allocate equal number of bits to Ambisonic (uncoupled) and non-diegetic + * (coupled) streams */ + per_stream_rate = total_rate / st->layout.nb_streams; for (i = 0; i < st->layout.nb_streams; i++) { - if (i < st->layout.nb_coupled_streams) - { - rate[i] = nondirectional_rate * 2; - } else if (i == st->layout.nb_coupled_streams) - { - rate[i] = nondirectional_rate + leftover_bits; - } else - { - rate[i] = directional_rate; - } + rate[i] = per_stream_rate; } } -#endif /* ENABLE_EXPERIMENTAL_AMBISONICS */ static opus_int32 rate_allocation( OpusMSEncoder *st, @@ -865,11 +782,9 @@ static opus_int32 rate_allocation( ptr = (char*)st + align(sizeof(OpusMSEncoder)); opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { ambisonics_rate_allocation(st, rate, frame_size, Fs); } else -#endif { surround_rate_allocation(st, rate, frame_size, Fs); } @@ -884,7 +799,7 @@ static opus_int32 rate_allocation( /* Max size in case the encoder decides to return six frames (6 x 20 ms = 120 ms) */ #define MS_FRAME_TMP (6*1275+12) -static int opus_multistream_encode_native +int opus_multistream_encode_native ( OpusMSEncoder *st, opus_copy_channel_in_func copy_channel_in, @@ -894,7 +809,8 @@ static int opus_multistream_encode_native opus_int32 max_data_bytes, int lsb_depth, downmix_func downmix, - int float_api + int float_api, + void *user_data ) { opus_int32 Fs; @@ -1001,11 +917,9 @@ static int opus_multistream_encode_native opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(2)); } } -#ifdef ENABLE_EXPERIMENTAL_AMBISONICS else if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); } -#endif } ptr = (char*)st + align(sizeof(OpusMSEncoder)); @@ -1028,9 +942,9 @@ static int opus_multistream_encode_native left = get_left_channel(&st->layout, s, -1); right = get_right_channel(&st->layout, s, -1); (*copy_channel_in)(buf, 2, - pcm, st->layout.nb_channels, left, frame_size); + pcm, st->layout.nb_channels, left, frame_size, user_data); (*copy_channel_in)(buf+1, 2, - pcm, st->layout.nb_channels, right, frame_size); + pcm, st->layout.nb_channels, right, frame_size, user_data); ptr += align(coupled_size); if (st->mapping_type == MAPPING_TYPE_SURROUND) { @@ -1046,7 +960,7 @@ static int opus_multistream_encode_native int i; int chan = get_mono_channel(&st->layout, s, -1); (*copy_channel_in)(buf, 1, - pcm, st->layout.nb_channels, chan, frame_size); + pcm, st->layout.nb_channels, chan, frame_size, user_data); ptr += align(mono_size); if (st->mapping_type == MAPPING_TYPE_SURROUND) { @@ -1089,7 +1003,7 @@ static int opus_multistream_encode_native return OPUS_INTERNAL_ERROR; } len = opus_repacketizer_out_range_impl(&rp, 0, opus_repacketizer_get_nb_frames(&rp), - data, max_data_bytes-tot_size, s != st->layout.nb_streams-1, !vbr && s == st->layout.nb_streams-1); + data, max_data_bytes-tot_size, s != st->layout.nb_streams-1, !vbr && s == st->layout.nb_streams-1, NULL, 0); data += len; tot_size += len; } @@ -1105,11 +1019,13 @@ static void opus_copy_channel_in_float( const void *src, int src_stride, int src_channel, - int frame_size + int frame_size, + void *user_data ) { const float *float_src; opus_int32 i; + (void)user_data; float_src = (const float *)src; for (i=0;i= st->layout.nb_streams) - ret = OPUS_BAD_ARG; + goto bad_arg; value = va_arg(ap, OpusEncoder**); if (!value) { @@ -1392,14 +1308,21 @@ int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) ret = OPUS_UNIMPLEMENTED; break; } - - va_end(ap); return ret; bad_arg: - va_end(ap); return OPUS_BAD_ARG; } +int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) +{ + int ret; + va_list ap; + va_start(ap, request); + ret = opus_multistream_encoder_ctl_va_list(st, request, ap); + va_end(ap); + return ret; +} + void opus_multistream_encoder_destroy(OpusMSEncoder *st) { opus_free(st); diff --git a/code/opus-1.2.1/src/opus_private.h b/code/opus-1.5.2/src/opus_private.h similarity index 63% rename from code/opus-1.2.1/src/opus_private.h rename to code/opus-1.5.2/src/opus_private.h index a731cc55..279f5f95 100644 --- a/code/opus-1.2.1/src/opus_private.h +++ b/code/opus-1.5.2/src/opus_private.h @@ -33,6 +33,7 @@ #include "opus.h" #include "celt.h" +#include /* va_list */ #include /* offsetof */ struct OpusRepacketizer { @@ -41,8 +42,17 @@ struct OpusRepacketizer { const unsigned char *frames[48]; opus_int16 len[48]; int framesize; + const unsigned char *paddings[48]; + opus_int32 padding_len[48]; }; +typedef struct { + int id; + int frame; + const unsigned char *data; + opus_int32 len; +} opus_extension_data; + typedef struct ChannelLayout { int nb_channels; int nb_streams; @@ -50,12 +60,59 @@ typedef struct ChannelLayout { unsigned char mapping[256]; } ChannelLayout; +typedef enum { + MAPPING_TYPE_NONE, + MAPPING_TYPE_SURROUND, + MAPPING_TYPE_AMBISONICS +} MappingType; + +struct OpusMSEncoder { + ChannelLayout layout; + int arch; + int lfe_stream; + int application; + int variable_duration; + MappingType mapping_type; + opus_int32 bitrate_bps; + /* Encoder states go here */ + /* then opus_val32 window_mem[channels*120]; */ + /* then opus_val32 preemph_mem[channels]; */ +}; + +struct OpusMSDecoder { + ChannelLayout layout; + /* Decoder states go here */ +}; + +int opus_multistream_encoder_ctl_va_list(struct OpusMSEncoder *st, int request, + va_list ap); +int opus_multistream_decoder_ctl_va_list(struct OpusMSDecoder *st, int request, + va_list ap); + int validate_layout(const ChannelLayout *layout); int get_left_channel(const ChannelLayout *layout, int stream_id, int prev); int get_right_channel(const ChannelLayout *layout, int stream_id, int prev); int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev); +typedef void (*opus_copy_channel_in_func)( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +); +typedef void (*opus_copy_channel_out_func)( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data +); #define MODE_SILK_ONLY 1000 #define MODE_HYBRID 1001 @@ -87,6 +144,7 @@ int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev); typedef void (*downmix_func)(const void *, opus_val32 *, int, int, int, int, int); void downmix_float(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); void downmix_int(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); +int is_digital_silence(const opus_val16* pcm, int frame_size, int channels, int lsb_depth); int encode_size(int size, unsigned char *data); @@ -99,7 +157,7 @@ opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_ int opus_decode_native(OpusDecoder *st, const unsigned char *data, opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec, int self_delimited, - opus_int32 *packet_offset, int soft_clip); + opus_int32 *packet_offset, int soft_clip, const OpusDRED *dred, opus_int32 dred_offset); /* Make sure everything is properly aligned. */ static OPUS_INLINE int align(int i) @@ -113,14 +171,53 @@ static OPUS_INLINE int align(int i) return ((i + alignment - 1) / alignment) * alignment; } +/* More than that is ridiculous for now (3 * max frames per packet)*/ +opus_int32 skip_extension(const unsigned char **data, opus_int32 len, opus_int32 *header_size); + int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, int self_delimited, unsigned char *out_toc, const unsigned char *frames[48], opus_int16 size[48], - int *payload_offset, opus_int32 *packet_offset); + int *payload_offset, opus_int32 *packet_offset, + const unsigned char **padding, opus_int32 *padding_len); opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, - unsigned char *data, opus_int32 maxlen, int self_delimited, int pad); + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad, + const opus_extension_data *extensions, int nb_extensions); int pad_frame(unsigned char *data, opus_int32 len, opus_int32 new_len); +int opus_multistream_encode_native +( + struct OpusMSEncoder *st, + opus_copy_channel_in_func copy_channel_in, + const void *pcm, + int analysis_frame_size, + unsigned char *data, + opus_int32 max_data_bytes, + int lsb_depth, + downmix_func downmix, + int float_api, + void *user_data +); + +int opus_multistream_decode_native( + struct OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + void *pcm, + opus_copy_channel_out_func copy_channel_out, + int frame_size, + int decode_fec, + int soft_clip, + void *user_data +); + +opus_int32 opus_packet_extensions_parse(const unsigned char *data, opus_int32 len, opus_extension_data *extensions, opus_int32 *nb_extensions); + +opus_int32 opus_packet_extensions_generate(unsigned char *data, opus_int32 len, const opus_extension_data *extensions, opus_int32 nb_extensions, int pad); + +opus_int32 opus_packet_extensions_count(const unsigned char *data, opus_int32 len); + +opus_int32 opus_packet_pad_impl(unsigned char *data, opus_int32 len, opus_int32 new_len, int pad, const opus_extension_data *extensions, int nb_extensions); + #endif /* OPUS_PRIVATE_H */ diff --git a/code/opus-1.5.2/src/opus_projection_decoder.c b/code/opus-1.5.2/src/opus_projection_decoder.c new file mode 100644 index 00000000..c2e07d5b --- /dev/null +++ b/code/opus-1.5.2/src/opus_projection_decoder.c @@ -0,0 +1,258 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "os_support.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "opus_projection.h" +#include "opus_multistream.h" +#include "mapping_matrix.h" +#include "stack_alloc.h" + +struct OpusProjectionDecoder +{ + opus_int32 demixing_matrix_size_in_bytes; + /* Encoder states go here */ +}; + +#if !defined(DISABLE_FLOAT_API) +static void opus_projection_copy_channel_out_float( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data) +{ + float *float_dst; + const MappingMatrix *matrix; + float_dst = (float *)dst; + matrix = (const MappingMatrix *)user_data; + + if (dst_channel == 0) + OPUS_CLEAR(float_dst, frame_size * dst_stride); + + if (src != NULL) + mapping_matrix_multiply_channel_out_float(matrix, src, dst_channel, + src_stride, float_dst, dst_stride, frame_size); +} +#endif + +static void opus_projection_copy_channel_out_short( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data) +{ + opus_int16 *short_dst; + const MappingMatrix *matrix; + short_dst = (opus_int16 *)dst; + matrix = (const MappingMatrix *)user_data; + if (dst_channel == 0) + OPUS_CLEAR(short_dst, frame_size * dst_stride); + + if (src != NULL) + mapping_matrix_multiply_channel_out_short(matrix, src, dst_channel, + src_stride, short_dst, dst_stride, frame_size); +} + +static MappingMatrix *get_dec_demixing_matrix(OpusProjectionDecoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix*)(void*)((char*)st + + align(sizeof(OpusProjectionDecoder))); +} + +static OpusMSDecoder *get_multistream_decoder(OpusProjectionDecoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (OpusMSDecoder*)(void*)((char*)st + + align(sizeof(OpusProjectionDecoder) + + st->demixing_matrix_size_in_bytes)); +} + +opus_int32 opus_projection_decoder_get_size(int channels, int streams, + int coupled_streams) +{ + opus_int32 matrix_size; + opus_int32 decoder_size; + + matrix_size = + mapping_matrix_get_size(streams + coupled_streams, channels); + if (!matrix_size) + return 0; + + decoder_size = opus_multistream_decoder_get_size(streams, coupled_streams); + if (!decoder_size) + return 0; + + return align(sizeof(OpusProjectionDecoder)) + matrix_size + decoder_size; +} + +int opus_projection_decoder_init(OpusProjectionDecoder *st, opus_int32 Fs, + int channels, int streams, int coupled_streams, + unsigned char *demixing_matrix, opus_int32 demixing_matrix_size) +{ + int nb_input_streams; + opus_int32 expected_matrix_size; + int i, ret; + unsigned char mapping[255]; + VARDECL(opus_int16, buf); + ALLOC_STACK; + + /* Verify supplied matrix size. */ + nb_input_streams = streams + coupled_streams; + expected_matrix_size = nb_input_streams * channels * sizeof(opus_int16); + if (expected_matrix_size != demixing_matrix_size) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Convert demixing matrix input into internal format. */ + ALLOC(buf, nb_input_streams * channels, opus_int16); + for (i = 0; i < nb_input_streams * channels; i++) + { + int s = demixing_matrix[2*i + 1] << 8 | demixing_matrix[2*i]; + s = ((s & 0xFFFF) ^ 0x8000) - 0x8000; + buf[i] = (opus_int16)s; + } + + /* Assign demixing matrix. */ + st->demixing_matrix_size_in_bytes = + mapping_matrix_get_size(channels, nb_input_streams); + if (!st->demixing_matrix_size_in_bytes) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + mapping_matrix_init(get_dec_demixing_matrix(st), channels, nb_input_streams, 0, + buf, demixing_matrix_size); + + /* Set trivial mapping so each input channel pairs with a matrix column. */ + for (i = 0; i < channels; i++) + mapping[i] = i; + + ret = opus_multistream_decoder_init( + get_multistream_decoder(st), Fs, channels, streams, coupled_streams, mapping); + RESTORE_STACK; + return ret; +} + +OpusProjectionDecoder *opus_projection_decoder_create( + opus_int32 Fs, int channels, int streams, int coupled_streams, + unsigned char *demixing_matrix, opus_int32 demixing_matrix_size, int *error) +{ + int size; + int ret; + OpusProjectionDecoder *st; + + /* Allocate space for the projection decoder. */ + size = opus_projection_decoder_get_size(channels, streams, coupled_streams); + if (!size) { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + st = (OpusProjectionDecoder *)opus_alloc(size); + if (!st) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + + /* Initialize projection decoder with provided settings. */ + ret = opus_projection_decoder_init(st, Fs, channels, streams, coupled_streams, + demixing_matrix, demixing_matrix_size); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +#ifdef FIXED_POINT +int opus_projection_decode(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, + int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_short, frame_size, decode_fec, 0, + get_dec_demixing_matrix(st)); +} +#else +int opus_projection_decode(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, + int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_short, frame_size, decode_fec, 1, + get_dec_demixing_matrix(st)); +} +#endif + +#ifndef DISABLE_FLOAT_API +int opus_projection_decode_float(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, float *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_float, frame_size, decode_fec, 0, + get_dec_demixing_matrix(st)); +} +#endif + +int opus_projection_decoder_ctl(OpusProjectionDecoder *st, int request, ...) +{ + va_list ap; + int ret = OPUS_OK; + + va_start(ap, request); + ret = opus_multistream_decoder_ctl_va_list(get_multistream_decoder(st), + request, ap); + va_end(ap); + return ret; +} + +void opus_projection_decoder_destroy(OpusProjectionDecoder *st) +{ + opus_free(st); +} + diff --git a/code/opus-1.5.2/src/opus_projection_encoder.c b/code/opus-1.5.2/src/opus_projection_encoder.c new file mode 100644 index 00000000..92813ad0 --- /dev/null +++ b/code/opus-1.5.2/src/opus_projection_encoder.c @@ -0,0 +1,510 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "os_support.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "opus_projection.h" +#include "opus_multistream.h" +#include "stack_alloc.h" +#include "mapping_matrix.h" + +struct OpusProjectionEncoder +{ + opus_int32 mixing_matrix_size_in_bytes; + opus_int32 demixing_matrix_size_in_bytes; + /* Encoder states go here */ +}; + +#if !defined(DISABLE_FLOAT_API) +static void opus_projection_copy_channel_in_float( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_float((const MappingMatrix*)user_data, + (const float*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} +#endif + +static void opus_projection_copy_channel_in_short( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_short((const MappingMatrix*)user_data, + (const opus_int16*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} + +static int get_order_plus_one_from_channels(int channels, int *order_plus_one) +{ + int order_plus_one_; + int acn_channels; + int nondiegetic_channels; + + /* Allowed numbers of channels: + * (1 + n)^2 + 2j, for n = 0...14 and j = 0 or 1. + */ + if (channels < 1 || channels > 227) + return OPUS_BAD_ARG; + + order_plus_one_ = isqrt32(channels); + acn_channels = order_plus_one_ * order_plus_one_; + nondiegetic_channels = channels - acn_channels; + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) + return OPUS_BAD_ARG; + + if (order_plus_one) + *order_plus_one = order_plus_one_; + return OPUS_OK; +} + +static int get_streams_from_channels(int channels, int mapping_family, + int *streams, int *coupled_streams, + int *order_plus_one) +{ + if (mapping_family == 3) + { + if (get_order_plus_one_from_channels(channels, order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + if (streams) + *streams = (channels + 1) / 2; + if (coupled_streams) + *coupled_streams = channels / 2; + return OPUS_OK; + } + return OPUS_BAD_ARG; +} + +static MappingMatrix *get_mixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder))); +} + +static MappingMatrix *get_enc_demixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes)); +} + +static OpusMSEncoder *get_multistream_encoder(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (OpusMSEncoder *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes + + st->demixing_matrix_size_in_bytes)); +} + +opus_int32 opus_projection_ambisonics_encoder_get_size(int channels, + int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + int order_plus_one; + int mixing_matrix_rows, mixing_matrix_cols; + int demixing_matrix_rows, demixing_matrix_cols; + opus_int32 mixing_matrix_size, demixing_matrix_size; + opus_int32 encoder_size; + int ret; + + ret = get_streams_from_channels(channels, mapping_family, &nb_streams, + &nb_coupled_streams, &order_plus_one); + if (ret != OPUS_OK) + return 0; + + if (order_plus_one == 2) + { + mixing_matrix_rows = mapping_matrix_foa_mixing.rows; + mixing_matrix_cols = mapping_matrix_foa_mixing.cols; + demixing_matrix_rows = mapping_matrix_foa_demixing.rows; + demixing_matrix_cols = mapping_matrix_foa_demixing.cols; + } + else if (order_plus_one == 3) + { + mixing_matrix_rows = mapping_matrix_soa_mixing.rows; + mixing_matrix_cols = mapping_matrix_soa_mixing.cols; + demixing_matrix_rows = mapping_matrix_soa_demixing.rows; + demixing_matrix_cols = mapping_matrix_soa_demixing.cols; + } + else if (order_plus_one == 4) + { + mixing_matrix_rows = mapping_matrix_toa_mixing.rows; + mixing_matrix_cols = mapping_matrix_toa_mixing.cols; + demixing_matrix_rows = mapping_matrix_toa_demixing.rows; + demixing_matrix_cols = mapping_matrix_toa_demixing.cols; + } + else if (order_plus_one == 5) + { + mixing_matrix_rows = mapping_matrix_fourthoa_mixing.rows; + mixing_matrix_cols = mapping_matrix_fourthoa_mixing.cols; + demixing_matrix_rows = mapping_matrix_fourthoa_demixing.rows; + demixing_matrix_cols = mapping_matrix_fourthoa_demixing.cols; + } + else if (order_plus_one == 6) + { + mixing_matrix_rows = mapping_matrix_fifthoa_mixing.rows; + mixing_matrix_cols = mapping_matrix_fifthoa_mixing.cols; + demixing_matrix_rows = mapping_matrix_fifthoa_demixing.rows; + demixing_matrix_cols = mapping_matrix_fifthoa_demixing.cols; + } + else + return 0; + + mixing_matrix_size = + mapping_matrix_get_size(mixing_matrix_rows, mixing_matrix_cols); + if (!mixing_matrix_size) + return 0; + + demixing_matrix_size = + mapping_matrix_get_size(demixing_matrix_rows, demixing_matrix_cols); + if (!demixing_matrix_size) + return 0; + + encoder_size = + opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (!encoder_size) + return 0; + + return align(sizeof(OpusProjectionEncoder)) + + mixing_matrix_size + demixing_matrix_size + encoder_size; +} + +int opus_projection_ambisonics_encoder_init(OpusProjectionEncoder *st, opus_int32 Fs, + int channels, int mapping_family, + int *streams, int *coupled_streams, + int application) +{ + MappingMatrix *mixing_matrix; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int i; + int ret; + int order_plus_one; + unsigned char mapping[255]; + + if (streams == NULL || coupled_streams == NULL) { + return OPUS_BAD_ARG; + } + + if (get_streams_from_channels(channels, mapping_family, streams, + coupled_streams, &order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + + if (mapping_family == 3) + { + /* Assign mixing matrix based on available pre-computed matrices. */ + mixing_matrix = get_mixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_foa_mixing.rows, + mapping_matrix_foa_mixing.cols, mapping_matrix_foa_mixing.gain, + mapping_matrix_foa_mixing_data, + sizeof(mapping_matrix_foa_mixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_soa_mixing.rows, + mapping_matrix_soa_mixing.cols, mapping_matrix_soa_mixing.gain, + mapping_matrix_soa_mixing_data, + sizeof(mapping_matrix_soa_mixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_toa_mixing.rows, + mapping_matrix_toa_mixing.cols, mapping_matrix_toa_mixing.gain, + mapping_matrix_toa_mixing_data, + sizeof(mapping_matrix_toa_mixing_data)); + } + else if (order_plus_one == 5) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_fourthoa_mixing.rows, + mapping_matrix_fourthoa_mixing.cols, mapping_matrix_fourthoa_mixing.gain, + mapping_matrix_fourthoa_mixing_data, + sizeof(mapping_matrix_fourthoa_mixing_data)); + } + else if (order_plus_one == 6) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_fifthoa_mixing.rows, + mapping_matrix_fifthoa_mixing.cols, mapping_matrix_fifthoa_mixing.gain, + mapping_matrix_fifthoa_mixing_data, + sizeof(mapping_matrix_fifthoa_mixing_data)); + } + else + return OPUS_BAD_ARG; + + st->mixing_matrix_size_in_bytes = mapping_matrix_get_size( + mixing_matrix->rows, mixing_matrix->cols); + if (!st->mixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + + /* Assign demixing matrix based on available pre-computed matrices. */ + demixing_matrix = get_enc_demixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_foa_demixing.rows, + mapping_matrix_foa_demixing.cols, mapping_matrix_foa_demixing.gain, + mapping_matrix_foa_demixing_data, + sizeof(mapping_matrix_foa_demixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_soa_demixing.rows, + mapping_matrix_soa_demixing.cols, mapping_matrix_soa_demixing.gain, + mapping_matrix_soa_demixing_data, + sizeof(mapping_matrix_soa_demixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_toa_demixing.rows, + mapping_matrix_toa_demixing.cols, mapping_matrix_toa_demixing.gain, + mapping_matrix_toa_demixing_data, + sizeof(mapping_matrix_toa_demixing_data)); + } + else if (order_plus_one == 5) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_fourthoa_demixing.rows, + mapping_matrix_fourthoa_demixing.cols, mapping_matrix_fourthoa_demixing.gain, + mapping_matrix_fourthoa_demixing_data, + sizeof(mapping_matrix_fourthoa_demixing_data)); + } + else if (order_plus_one == 6) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_fifthoa_demixing.rows, + mapping_matrix_fifthoa_demixing.cols, mapping_matrix_fifthoa_demixing.gain, + mapping_matrix_fifthoa_demixing_data, + sizeof(mapping_matrix_fifthoa_demixing_data)); + } + else + return OPUS_BAD_ARG; + + st->demixing_matrix_size_in_bytes = mapping_matrix_get_size( + demixing_matrix->rows, demixing_matrix->cols); + if (!st->demixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + } + else + return OPUS_UNIMPLEMENTED; + + /* Ensure matrices are large enough for desired coding scheme. */ + if (*streams + *coupled_streams > mixing_matrix->rows || + channels > mixing_matrix->cols || + channels > demixing_matrix->rows || + *streams + *coupled_streams > demixing_matrix->cols) + return OPUS_BAD_ARG; + + /* Set trivial mapping so each input channel pairs with a matrix column. */ + for (i = 0; i < channels; i++) + mapping[i] = i; + + /* Initialize multistream encoder with provided settings. */ + ms_encoder = get_multistream_encoder(st); + ret = opus_multistream_encoder_init(ms_encoder, Fs, channels, *streams, + *coupled_streams, mapping, application); + return ret; +} + +OpusProjectionEncoder *opus_projection_ambisonics_encoder_create( + opus_int32 Fs, int channels, int mapping_family, int *streams, + int *coupled_streams, int application, int *error) +{ + int size; + int ret; + OpusProjectionEncoder *st; + + /* Allocate space for the projection encoder. */ + size = opus_projection_ambisonics_encoder_get_size(channels, mapping_family); + if (!size) { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + st = (OpusProjectionEncoder *)opus_alloc(size); + if (!st) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + + /* Initialize projection encoder with provided settings. */ + ret = opus_projection_ambisonics_encoder_init(st, Fs, channels, + mapping_family, streams, coupled_streams, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +int opus_projection_encode(OpusProjectionEncoder *st, const opus_int16 *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_short, pcm, frame_size, data, + max_data_bytes, 16, downmix_int, 0, get_mixing_matrix(st)); +} + +#ifndef DISABLE_FLOAT_API +#ifdef FIXED_POINT +int opus_projection_encode_float(OpusProjectionEncoder *st, const float *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_float, pcm, frame_size, data, + max_data_bytes, 16, downmix_float, 1, get_mixing_matrix(st)); +} +#else +int opus_projection_encode_float(OpusProjectionEncoder *st, const float *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_float, pcm, frame_size, data, + max_data_bytes, 24, downmix_float, 1, get_mixing_matrix(st)); +} +#endif +#endif + +void opus_projection_encoder_destroy(OpusProjectionEncoder *st) +{ + opus_free(st); +} + +int opus_projection_encoder_ctl(OpusProjectionEncoder *st, int request, ...) +{ + va_list ap; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int ret = OPUS_OK; + + ms_encoder = get_multistream_encoder(st); + demixing_matrix = get_enc_demixing_matrix(st); + + va_start(ap, request); + switch(request) + { + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = + ms_encoder->layout.nb_channels * (ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams) * sizeof(opus_int16); + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = demixing_matrix->gain; + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST: + { + int i, j, k, l; + int nb_input_streams; + int nb_output_streams; + unsigned char *external_char; + opus_int16 *internal_short; + opus_int32 external_size; + opus_int32 internal_size; + + /* (I/O is in relation to the decoder's perspective). */ + nb_input_streams = ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams; + nb_output_streams = ms_encoder->layout.nb_channels; + + external_char = va_arg(ap, unsigned char *); + external_size = va_arg(ap, opus_int32); + if (!external_char) + { + goto bad_arg; + } + internal_short = mapping_matrix_get_data(demixing_matrix); + internal_size = nb_input_streams * nb_output_streams * sizeof(opus_int16); + if (external_size != internal_size) + { + goto bad_arg; + } + + /* Copy demixing matrix subset to output destination. */ + l = 0; + for (i = 0; i < nb_input_streams; i++) { + for (j = 0; j < nb_output_streams; j++) { + k = demixing_matrix->rows * i + j; + external_char[2*l] = (unsigned char)internal_short[k]; + external_char[2*l+1] = (unsigned char)(internal_short[k] >> 8); + l++; + } + } + } + break; + default: + { + ret = opus_multistream_encoder_ctl_va_list(ms_encoder, request, ap); + } + break; + } + va_end(ap); + return ret; + +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + diff --git a/code/opus-1.2.1/src/repacketizer.c b/code/opus-1.5.2/src/repacketizer.c similarity index 68% rename from code/opus-1.2.1/src/repacketizer.c rename to code/opus-1.5.2/src/repacketizer.c index c80ee7f0..79798b02 100644 --- a/code/opus-1.2.1/src/repacketizer.c +++ b/code/opus-1.5.2/src/repacketizer.c @@ -32,6 +32,7 @@ #include "opus.h" #include "opus_private.h" #include "os_support.h" +#include "stack_alloc.h" int opus_repacketizer_get_size(void) @@ -82,10 +83,19 @@ static int opus_repacketizer_cat_impl(OpusRepacketizer *rp, const unsigned char return OPUS_INVALID_PACKET; } - ret=opus_packet_parse_impl(data, len, self_delimited, &tmp_toc, &rp->frames[rp->nb_frames], &rp->len[rp->nb_frames], NULL, NULL); + ret=opus_packet_parse_impl(data, len, self_delimited, &tmp_toc, &rp->frames[rp->nb_frames], &rp->len[rp->nb_frames], + NULL, NULL, &rp->paddings[rp->nb_frames], &rp->padding_len[rp->nb_frames]); if(ret<1)return ret; - rp->nb_frames += curr_nb_frames; + /* set padding length to zero for all but the first frame */ + while (curr_nb_frames > 1) + { + rp->nb_frames++; + rp->padding_len[rp->nb_frames] = 0; + rp->paddings[rp->nb_frames] = NULL; + curr_nb_frames--; + } + rp->nb_frames++; return OPUS_OK; } @@ -100,17 +110,23 @@ int opus_repacketizer_get_nb_frames(OpusRepacketizer *rp) } opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, - unsigned char *data, opus_int32 maxlen, int self_delimited, int pad) + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad, const opus_extension_data *extensions, int nb_extensions) { int i, count; opus_int32 tot_size; opus_int16 *len; const unsigned char **frames; unsigned char * ptr; + int ones_begin=0, ones_end=0; + int ext_begin=0, ext_len=0; + int ext_count, total_ext_count; + VARDECL(opus_extension_data, all_extensions); + SAVE_STACK; if (begin<0 || begin>=end || end>rp->nb_frames) { /*fprintf(stderr, "%d %d %d\n", begin, end, rp->nb_frames);*/ + RESTORE_STACK; return OPUS_BAD_ARG; } count = end-begin; @@ -122,13 +138,51 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int else tot_size = 0; + /* figure out total number of extensions */ + total_ext_count = nb_extensions; + for (i=begin;ipaddings[i], rp->padding_len[i]); + if (n > 0) total_ext_count += n; + } + ALLOC(all_extensions, total_ext_count ? total_ext_count : ALLOC_NONE, opus_extension_data); + /* copy over any extensions that were passed in */ + for (ext_count=0;ext_countpaddings[i], rp->padding_len[i], + &all_extensions[ext_count], &frame_ext_count); + if (ret<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + /* renumber the extension frame numbers */ + for (j=0;j maxlen) + { + RESTORE_STACK; return OPUS_BUFFER_TOO_SMALL; + } *ptr++ = rp->toc&0xFC; } else if (count==2) { @@ -137,18 +191,24 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int /* Code 1 */ tot_size += 2*len[0]+1; if (tot_size > maxlen) + { + RESTORE_STACK; return OPUS_BUFFER_TOO_SMALL; + } *ptr++ = (rp->toc&0xFC) | 0x1; } else { /* Code 2 */ tot_size += len[0]+len[1]+2+(len[0]>=252); if (tot_size > maxlen) + { + RESTORE_STACK; return OPUS_BUFFER_TOO_SMALL; + } *ptr++ = (rp->toc&0xFC) | 0x2; ptr += encode_size(len[0], ptr); } } - if (count > 2 || (pad && tot_size < maxlen)) + if (count > 2 || (pad && tot_size < maxlen) || ext_count > 0) { /* Code 3 */ int vbr; @@ -177,22 +237,45 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int tot_size += len[count-1]; if (tot_size > maxlen) + { + RESTORE_STACK; return OPUS_BUFFER_TOO_SMALL; + } *ptr++ = (rp->toc&0xFC) | 0x3; *ptr++ = count | 0x80; } else { tot_size += count*len[0]+2; if (tot_size > maxlen) + { + RESTORE_STACK; return OPUS_BUFFER_TOO_SMALL; + } *ptr++ = (rp->toc&0xFC) | 0x3; *ptr++ = count; } pad_amount = pad ? (maxlen-tot_size) : 0; + if (ext_count>0) + { + /* figure out how much space we need for the extensions */ + ext_len = opus_packet_extensions_generate(NULL, maxlen-tot_size, all_extensions, ext_count, 0); + if (ext_len < 0) return ext_len; + if (!pad) + pad_amount = ext_len + ext_len/254 + 1; + } if (pad_amount != 0) { int nb_255s; data[1] |= 0x40; nb_255s = (pad_amount-1)/255; + if (tot_size + ext_len + nb_255s + 1 > maxlen) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + ext_begin = tot_size+pad_amount-ext_len; + /* Prepend 0x01 padding */ + ones_begin = tot_size+nb_255s+1; + ones_end = tot_size+pad_amount-ext_len; for (i=0;i 0) { + int ret = opus_packet_extensions_generate(&data[ext_begin], ext_len, all_extensions, ext_count, 0); + celt_assert(ret == ext_len); + } + for (i=ones_begin;inb_frames, data, maxlen, 0, 0); + return opus_repacketizer_out_range_impl(rp, 0, rp->nb_frames, data, maxlen, 0, 0, NULL, 0); } -int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len) +opus_int32 opus_packet_pad_impl(unsigned char *data, opus_int32 len, opus_int32 new_len, int pad, const opus_extension_data *extensions, int nb_extensions) { OpusRepacketizer rp; opus_int32 ret; + VARDECL(unsigned char, copy); + SAVE_STACK; if (len < 1) return OPUS_BAD_ARG; if (len==new_len) return OPUS_OK; else if (len > new_len) return OPUS_BAD_ARG; + ALLOC(copy, len, unsigned char); opus_repacketizer_init(&rp); /* Moving payload to the end of the packet so we can do in-place padding */ - OPUS_MOVE(data+new_len-len, data, len); - ret = opus_repacketizer_cat(&rp, data+new_len-len, len); + OPUS_COPY(copy, data, len); + ret = opus_repacketizer_cat(&rp, copy, len); if (ret != OPUS_OK) return ret; - ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, new_len, 0, 1); + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, new_len, 0, pad, extensions, nb_extensions); + RESTORE_STACK; + return ret; +} + +int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len) +{ + opus_int32 ret; + ALLOC_STACK; + ret = opus_packet_pad_impl(data, len, new_len, 1, NULL, 0); + RESTORE_STACK; if (ret > 0) return OPUS_OK; else @@ -263,13 +367,19 @@ opus_int32 opus_packet_unpad(unsigned char *data, opus_int32 len) { OpusRepacketizer rp; opus_int32 ret; + int i; if (len < 1) return OPUS_BAD_ARG; opus_repacketizer_init(&rp); ret = opus_repacketizer_cat(&rp, data, len); if (ret < 0) return ret; - ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, len, 0, 0); + /* Discard all padding and extensions. */ + for (i=0;i 0 && ret <= len); return ret; } @@ -296,7 +406,7 @@ int opus_multistream_packet_pad(unsigned char *data, opus_int32 len, opus_int32 if (len<=0) return OPUS_INVALID_PACKET; count = opus_packet_parse_impl(data, len, 1, &toc, NULL, - size, NULL, &packet_offset); + size, NULL, &packet_offset, NULL, NULL); if (count<0) return count; data += packet_offset; @@ -323,18 +433,24 @@ opus_int32 opus_multistream_packet_unpad(unsigned char *data, opus_int32 len, in for (s=0;s +#include +#include + +#define MAX_PACKETOUT 32000 + +void usage(char *argv0) +{ + fprintf(stderr, "usage: %s [options] input_file output_file\n", argv0); +} + +static void int_to_char(opus_uint32 i, unsigned char ch[4]) +{ + ch[0] = i>>24; + ch[1] = (i>>16)&0xFF; + ch[2] = (i>>8)&0xFF; + ch[3] = i&0xFF; +} + +static opus_uint32 char_to_int(unsigned char ch[4]) +{ + return ((opus_uint32)ch[0]<<24) | ((opus_uint32)ch[1]<<16) + | ((opus_uint32)ch[2]<< 8) | (opus_uint32)ch[3]; +} + +int main(int argc, char *argv[]) +{ + int i, eof=0; + FILE *fin, *fout; + unsigned char packets[48][1500]; + int len[48]; + int rng[48]; + OpusRepacketizer *rp; + unsigned char output_packet[MAX_PACKETOUT]; + int merge = 1, split=0; + + if (argc < 3) + { + usage(argv[0]); + return EXIT_FAILURE; + } + for (i=1;i48) + { + fprintf(stderr, "-merge parameter must be less than 48.\n"); + return EXIT_FAILURE; + } + i++; + } else if (strcmp(argv[i], "-split")==0) + split = 1; + else + { + fprintf(stderr, "Unknown option: %s\n", argv[i]); + usage(argv[0]); + return EXIT_FAILURE; + } + } + fin = fopen(argv[argc-2], "r"); + if(fin==NULL) + { + fprintf(stderr, "Error opening input file: %s\n", argv[argc-2]); + return EXIT_FAILURE; + } + fout = fopen(argv[argc-1], "w"); + if(fout==NULL) + { + fprintf(stderr, "Error opening output file: %s\n", argv[argc-1]); + fclose(fin); + return EXIT_FAILURE; + } + + rp = opus_repacketizer_create(); + while (!eof) + { + int err; + int nb_packets=merge; + opus_repacketizer_init(rp); + for (i=0;i1500 || len[i]<0) + { + if (feof(fin)) + { + eof = 1; + } else { + fprintf(stderr, "Invalid payload length\n"); + fclose(fin); + fclose(fout); + return EXIT_FAILURE; + } + break; + } + if (fread(ch, 1, 4, fin)!=4) + { + if (feof(fin)) + { + eof = 1; + } else { + fprintf(stderr, "Error reading.\n"); + fclose(fin); + fclose(fout); + return EXIT_FAILURE; + } + break; + } + rng[i] = char_to_int(ch); + if (fread(packets[i], len[i], 1, fin)!=1) { + if (feof(fin)) + { + eof = 1; + } else { + fprintf(stderr, "Error reading packet of %u bytes.\n", len[i]); + fclose(fin); + fclose(fout); + return EXIT_FAILURE; + } + break; + } + err = opus_repacketizer_cat(rp, packets[i], len[i]); + if (err!=OPUS_OK) + { + fprintf(stderr, "opus_repacketizer_cat() failed: %s\n", opus_strerror(err)); + break; + } + } + nb_packets = i; + + if (eof) + break; + + if (!split) + { + err = opus_repacketizer_out(rp, output_packet, MAX_PACKETOUT); + if (err>0) { + unsigned char int_field[4]; + int_to_char(err, int_field); + if(fwrite(int_field, 1, 4, fout)!=4){ + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + int_to_char(rng[nb_packets-1], int_field); + if (fwrite(int_field, 1, 4, fout)!=4) { + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + if (fwrite(output_packet, 1, err, fout)!=(unsigned)err) { + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + /*fprintf(stderr, "out len = %d\n", err);*/ + } else { + fprintf(stderr, "opus_repacketizer_out() failed: %s\n", opus_strerror(err)); + } + } else { + int nb_frames = opus_repacketizer_get_nb_frames(rp); + for (i=0;i0) { + unsigned char int_field[4]; + int_to_char(err, int_field); + if (fwrite(int_field, 1, 4, fout)!=4) { + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + if (i==nb_frames-1) + int_to_char(rng[nb_packets-1], int_field); + else + int_to_char(0, int_field); + if (fwrite(int_field, 1, 4, fout)!=4) { + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + if (fwrite(output_packet, 1, err, fout)!=(unsigned)err) { + fprintf(stderr, "Error writing.\n"); + return EXIT_FAILURE; + } + /*fprintf(stderr, "out len = %d\n", err);*/ + } else { + fprintf(stderr, "opus_repacketizer_out() failed: %s\n", opus_strerror(err)); + } + + } + } + } + + fclose(fin); + fclose(fout); + return EXIT_SUCCESS; +} diff --git a/code/opusfile-0.9/include/opusfile.h b/code/opusfile-0.12/include/opusfile.h similarity index 99% rename from code/opusfile-0.9/include/opusfile.h rename to code/opusfile-0.12/include/opusfile.h index e3a3dc83..7c1c89e0 100644 --- a/code/opusfile-0.9/include/opusfile.h +++ b/code/opusfile-0.12/include/opusfile.h @@ -6,7 +6,7 @@ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 1994-2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ******************************************************************** @@ -28,7 +28,7 @@ reference libogg and - libopus + libopus libraries. libopusfile provides several sets of built-in routines for @@ -58,7 +58,7 @@ it is stored in the header to allow you to resample to it after decoding (the libopusfile API does not currently provide a resampler, but the - the + the Speex resampler is a good choice if you need one). In general, if you are playing back the audio, you should leave it at 48 kHz, provided your audio hardware supports it. @@ -68,7 +68,7 @@ Opus files can contain anywhere from 1 to 255 channels of audio. The channel mappings for up to 8 channels are the same as the - Vorbis + Vorbis mappings. A special stereo API can convert everything to 2 channels, making it simple to support multichannel files in an application which only has stereo @@ -284,7 +284,7 @@ struct OpusHead{ A particular tag may occur more than once, and order is significant. The character set encoding for the strings is always UTF-8, but the tag names are limited to ASCII, and treated as case-insensitive. - See the Vorbis + See the Vorbis comment header specification for details. In filling in this structure, libopusfile will null-terminate the @@ -698,7 +698,7 @@ void opus_picture_tag_clear(OpusPictureTag *_pic) OP_ARG_NONNULL(1); #define OP_HTTP_PROXY_PASS_REQUEST (6720) #define OP_GET_SERVER_INFO_REQUEST (6784) -#define OP_URL_OPT(_request) ((_request)+(char *)0) +#define OP_URL_OPT(_request) ((char *)(_request)) /*These macros trigger compilation errors or warnings if the wrong types are provided to one of the URL options.*/ @@ -1890,7 +1890,7 @@ void op_set_dither_enabled(OggOpusFile *_of,int _enabled) OP_ARG_NONNULL(1); signed native-endian 16-bit values at 48 kHz with a nominal range of [-32768,32767). Multiple channels are interleaved using the - Vorbis + Vorbis channel ordering. This must have room for at least \a _buf_size values. \param _buf_size The number of values that can be stored in \a _pcm. @@ -1972,7 +1972,7 @@ OP_WARN_UNUSED_RESULT int op_read(OggOpusFile *_of, signed floats at 48 kHz with a nominal range of [-1.0,1.0]. Multiple channels are interleaved using the - Vorbis + Vorbis channel ordering. This must have room for at least \a _buf_size floats. \param _buf_size The number of floats that can be stored in \a _pcm. diff --git a/code/opusfile-0.9/src/http.c b/code/opusfile-0.12/src/http.c similarity index 97% rename from code/opusfile-0.9/src/http.c rename to code/opusfile-0.12/src/http.c index acd0e8cd..bd085627 100644 --- a/code/opusfile-0.9/src/http.c +++ b/code/opusfile-0.12/src/http.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012-2020 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ #ifdef HAVE_CONFIG_H @@ -334,6 +334,9 @@ int SSL_CTX_set_default_verify_paths_win32(SSL_CTX *_ssl_ctx); # else /*Normal Berkeley sockets.*/ +# ifndef BSD_COMP +# define BSD_COMP 1 /* for FIONREAD on Solaris/Illumos */ +# endif # include # include # include @@ -355,9 +358,22 @@ typedef int op_sock; # define op_reset_errno() (errno=0) # endif -# include + +# ifdef OP_HAVE_CLOCK_GETTIME +# include +typedef struct timespec op_time; +# else +# include +typedef struct timeb op_time; +# endif + # include +# if (defined(LIBRESSL_VERSION_NUMBER)&&OPENSSL_VERSION_NUMBER==0x20000000L) +# undef OPENSSL_VERSION_NUMBER +# define OPENSSL_VERSION_NUMBER 0x1000115fL +# endif + /*The maximum number of simultaneous connections. RFC 2616 says this SHOULD NOT be more than 2, but everyone on the modern web ignores that (e.g., IE 8 bumped theirs up from 2 to 6, Firefox uses 15). @@ -548,7 +564,7 @@ static int op_parse_url_impl(OpusParsedURL *_dst,const char *_src){ /*Discard the component, if any. This doesn't get sent to the server. Some day we should add support for Media Fragment URIs - .*/ + .*/ if(*path_end=='#')uri_end=path_end+1+strspn(path_end+1,OP_URL_QUERY_FRAG); else uri_end=path_end; /*If there's anything left, this was not a valid URL.*/ @@ -799,7 +815,7 @@ struct OpusHTTPConn{ /*The next connection in either the LRU or free list.*/ OpusHTTPConn *next; /*The last time we blocked for reading from this connection.*/ - struct timeb read_time; + op_time read_time; /*The number of bytes we've read since the last time we blocked.*/ opus_int64 read_bytes; /*The estimated throughput of this connection, in bytes/s.*/ @@ -849,7 +865,7 @@ struct OpusHTTPStream{ struct sockaddr_in6 v6; } addr; /*The last time we re-resolved the host.*/ - struct timeb resolve_time; + op_time resolve_time; /*A buffer used to build HTTP requests.*/ OpusStringBuf request; /*A buffer used to build proxy CONNECT requests.*/ @@ -1004,8 +1020,31 @@ static int op_http_conn_estimate_available(OpusHTTPConn *_conn){ return available; } -static opus_int32 op_time_diff_ms(const struct timeb *_end, - const struct timeb *_start){ +static void op_time_get(op_time *now){ +# ifdef OP_HAVE_CLOCK_GETTIME + /*Prefer a monotonic clock that continues to increment during suspend.*/ +# ifdef CLOCK_BOOTTIME + if(clock_gettime(CLOCK_BOOTTIME,now)!=0) +# endif +# ifdef CLOCK_MONOTONIC + if(clock_gettime(CLOCK_MONOTONIC,now)!=0) +# endif + OP_ALWAYS_TRUE(!clock_gettime(CLOCK_REALTIME,now)); +# else + ftime(now); +# endif +} + +static opus_int32 op_time_diff_ms(const op_time *_end, const op_time *_start){ +# ifdef OP_HAVE_CLOCK_GETTIME + opus_int64 dtime; + dtime=_end->tv_sec-(opus_int64)_start->tv_sec; + OP_ASSERT(_end->tv_nsec<1000000000); + OP_ASSERT(_start->tv_nsec<1000000000); + if(OP_UNLIKELY(dtime>(OP_INT32_MAX-1000)/1000))return OP_INT32_MAX; + if(OP_UNLIKELY(dtime<(OP_INT32_MIN+1000)/1000))return OP_INT32_MIN; + return (opus_int32)dtime*1000+(_end->tv_nsec-_start->tv_nsec)/1000000; +# else opus_int64 dtime; dtime=_end->time-(opus_int64)_start->time; OP_ASSERT(_end->millitm<1000); @@ -1013,17 +1052,18 @@ static opus_int32 op_time_diff_ms(const struct timeb *_end, if(OP_UNLIKELY(dtime>(OP_INT32_MAX-1000)/1000))return OP_INT32_MAX; if(OP_UNLIKELY(dtime<(OP_INT32_MIN+1000)/1000))return OP_INT32_MIN; return (opus_int32)dtime*1000+_end->millitm-_start->millitm; +# endif } /*Update the read rate estimate for this connection.*/ static void op_http_conn_read_rate_update(OpusHTTPConn *_conn){ - struct timeb read_time; + op_time read_time; opus_int32 read_delta_ms; opus_int64 read_delta_bytes; opus_int64 read_rate; read_delta_bytes=_conn->read_bytes; if(read_delta_bytes<=0)return; - ftime(&read_time); + op_time_get(&read_time); read_delta_ms=op_time_diff_ms(&read_time,&_conn->read_time); read_rate=_conn->read_rate; read_delta_ms=OP_MAX(read_delta_ms,1); @@ -1517,8 +1557,8 @@ static long op_bio_retry_ctrl(BIO *_b,int _cmd,long _num,void *_ptr){ case BIO_CTRL_RESET: case BIO_C_RESET_READ_REQUEST:{ BIO_clear_retry_flags(_b); - /*Fall through.*/ } + /*Fall through.*/ case BIO_CTRL_EOF: case BIO_CTRL_SET: case BIO_CTRL_SET_CLOSE: @@ -1530,7 +1570,7 @@ static long op_bio_retry_ctrl(BIO *_b,int _cmd,long _num,void *_ptr){ return ret; } -# if OPENSSL_VERSION_NUMBER<0x10100000L +# if (OPENSSL_VERSION_NUMBER<0x10100000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) # define BIO_set_data(_b,_ptr) ((_b)->ptr=(_ptr)) # define BIO_set_init(_b,_init) ((_b)->init=(_init)) # define ASN1_STRING_get0_data ASN1_STRING_data @@ -1538,7 +1578,7 @@ static long op_bio_retry_ctrl(BIO *_b,int _cmd,long _num,void *_ptr){ static int op_bio_retry_new(BIO *_b){ BIO_set_init(_b,1); -# if OPENSSL_VERSION_NUMBER<0x10100000L +# if (OPENSSL_VERSION_NUMBER<0x10100000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) _b->num=0; # endif BIO_set_data(_b,NULL); @@ -1549,7 +1589,7 @@ static int op_bio_retry_free(BIO *_b){ return _b!=NULL; } -# if OPENSSL_VERSION_NUMBER<0x10100000L +# if (OPENSSL_VERSION_NUMBER<0x10100000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) /*This is not const because OpenSSL doesn't allow it, even though it won't write to it.*/ static BIO_METHOD op_bio_retry_method={ @@ -1570,7 +1610,7 @@ static BIO_METHOD op_bio_retry_method={ proxying https URL requests.*/ static int op_http_conn_establish_tunnel(OpusHTTPStream *_stream, OpusHTTPConn *_conn,op_sock _fd,SSL *_ssl_conn,BIO *_ssl_bio){ -# if OPENSSL_VERSION_NUMBER>=0x10100000L +# if (OPENSSL_VERSION_NUMBER>=0x10100000L||LIBRESSL_VERSION_NUMBER>=0x2070000fL) BIO_METHOD *bio_retry_method; # endif BIO *retry_bio; @@ -1583,7 +1623,7 @@ static int op_http_conn_establish_tunnel(OpusHTTPStream *_stream, ret=op_http_conn_write_fully(_conn, _stream->proxy_connect.buf,_stream->proxy_connect.nbuf); if(OP_UNLIKELY(ret<0))return ret; -# if OPENSSL_VERSION_NUMBER>=0x10100000L +# if (OPENSSL_VERSION_NUMBER>=0x10100000L||LIBRESSL_VERSION_NUMBER>=0x2070000fL) bio_retry_method=BIO_meth_new(BIO_TYPE_NULL,"retry"); if(bio_retry_method==NULL)return OP_EFAULT; BIO_meth_set_write(bio_retry_method,op_bio_retry_write); @@ -1606,7 +1646,7 @@ static int op_http_conn_establish_tunnel(OpusHTTPStream *_stream, /*This shouldn't succeed, since we can't read yet.*/ OP_ALWAYS_TRUE(SSL_connect(_ssl_conn)<0); SSL_set_bio(_ssl_conn,_ssl_bio,_ssl_bio); -# if OPENSSL_VERSION_NUMBER>=0x10100000L +# if (OPENSSL_VERSION_NUMBER>=0x10100000L||LIBRESSL_VERSION_NUMBER>=0x2070000fL) BIO_meth_free(bio_retry_method); # endif /*Only now do we disable write coalescing, to allow the CONNECT @@ -1635,7 +1675,7 @@ static struct addrinfo *op_inet_pton(const char *_host){ return NULL; } -# if OPENSSL_VERSION_NUMBER<0x10002000L +# if (OPENSSL_VERSION_NUMBER<0x10002000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) /*Match a host name against a host with a possible wildcard pattern according to the rules of RFC 6125 Section 6.4.3. Return: 0 if the pattern doesn't match, and a non-zero value if it does.*/ @@ -1893,7 +1933,7 @@ static int op_http_conn_start_tls(OpusHTTPStream *_stream,OpusHTTPConn *_conn, SSL_set_tlsext_host_name(_ssl_conn,_stream->url.host); # endif skip_certificate_check=_stream->skip_certificate_check; -# if OPENSSL_VERSION_NUMBER>=0x10002000L +# if (OPENSSL_VERSION_NUMBER>=0x10002000L||LIBRESSL_VERSION_NUMBER>=0x2070000fL) /*As of version 1.0.2, OpenSSL can finally do hostname checks automatically. Of course, they make it much more complicated than it needs to be.*/ if(!skip_certificate_check){ @@ -1956,13 +1996,13 @@ static int op_http_conn_start_tls(OpusHTTPStream *_stream,OpusHTTPConn *_conn, if(OP_UNLIKELY(ret<=0))return OP_FALSE; ssl_session=_stream->ssl_session; if(ssl_session==NULL -# if OPENSSL_VERSION_NUMBER<0x10002000L +# if (OPENSSL_VERSION_NUMBER<0x10002000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) ||!skip_certificate_check # endif ){ ret=op_do_ssl_step(_ssl_conn,_fd,SSL_do_handshake); if(OP_UNLIKELY(ret<=0))return OP_FALSE; -# if OPENSSL_VERSION_NUMBER<0x10002000L +# if (OPENSSL_VERSION_NUMBER<0x10002000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) /*OpenSSL before version 1.0.2 does not do automatic hostname verification, despite the fact that we just passed it the hostname above in the call to SSL_set_tlsext_host_name(). @@ -2015,7 +2055,7 @@ static int op_sock_connect_next(op_sock _fd, # define OP_NPROTOS (2) static int op_http_connect_impl(OpusHTTPStream *_stream,OpusHTTPConn *_conn, - struct addrinfo *_addrs,struct timeb *_start_time){ + struct addrinfo *_addrs,op_time *_start_time){ struct addrinfo *addr; struct addrinfo *addrs[OP_NPROTOS]; struct pollfd fds[OP_NPROTOS]; @@ -2045,7 +2085,7 @@ static int op_http_connect_impl(OpusHTTPStream *_stream,OpusHTTPConn *_conn, _stream->free_head=_conn->next; _conn->next=_stream->lru_head; _stream->lru_head=_conn; - ftime(_start_time); + op_time_get(_start_time); *&_conn->read_time=*_start_time; _conn->read_bytes=0; _conn->read_rate=0; @@ -2147,14 +2187,14 @@ static int op_http_connect_impl(OpusHTTPStream *_stream,OpusHTTPConn *_conn, } static int op_http_connect(OpusHTTPStream *_stream,OpusHTTPConn *_conn, - struct addrinfo *_addrs,struct timeb *_start_time){ - struct timeb resolve_time; + struct addrinfo *_addrs,op_time *_start_time){ + op_time resolve_time; struct addrinfo *new_addrs; int ret; /*Re-resolve the host if we need to (RFC 6555 says we MUST do so occasionally).*/ new_addrs=NULL; - ftime(&resolve_time); + op_time_get(&resolve_time); if(_addrs!=&_stream->addr_info||op_time_diff_ms(&resolve_time, &_stream->resolve_time)>=OP_RESOLVE_CACHE_TIMEOUT_MS){ new_addrs=op_resolve(_stream->connect_host,_stream->connect_port); @@ -2305,8 +2345,8 @@ static int op_http_stream_open(OpusHTTPStream *_stream,const char *_url, addrs=NULL; for(nredirs=0;nredirsurl)&&_stream->ssl_ctx==NULL){ SSL_CTX *ssl_ctx; -# if OPENSSL_VERSION_NUMBER<0x10100000L +# if (OPENSSL_VERSION_NUMBER<0x10100000L&&LIBRESSL_VERSION_NUMBER<0x2070000fL) # if !defined(OPENSSL_NO_LOCKING) /*The documentation says SSL_library_init() is not reentrant. We don't want to add our own depenencies on a threading library, and it @@ -2440,7 +2480,7 @@ static int op_http_stream_open(OpusHTTPStream *_stream,const char *_url, if(OP_UNLIKELY(ret<0))return ret; ret=op_http_conn_read_response(_stream->conns+0,&_stream->response); if(OP_UNLIKELY(ret<0))return ret; - ftime(&end_time); + op_time_get(&end_time); next=op_http_parse_status_line(&v1_1_compat,&status_code, _stream->response.buf); if(OP_UNLIKELY(next==NULL))return OP_FALSE; @@ -2852,8 +2892,8 @@ static int op_http_conn_handle_response(OpusHTTPStream *_stream, converted into a request for the rest.*/ static int op_http_conn_open_pos(OpusHTTPStream *_stream, OpusHTTPConn *_conn,opus_int64 _pos,opus_int32 _chunk_size){ - struct timeb start_time; - struct timeb end_time; + op_time start_time; + op_time end_time; opus_int32 connect_rate; opus_int32 connect_time; int ret; @@ -2863,7 +2903,7 @@ static int op_http_conn_open_pos(OpusHTTPStream *_stream, if(OP_UNLIKELY(ret<0))return ret; ret=op_http_conn_handle_response(_stream,_conn); if(OP_UNLIKELY(ret!=0))return OP_FALSE; - ftime(&end_time); + op_time_get(&end_time); _stream->cur_conni=(int)(_conn-_stream->conns); OP_ASSERT(_stream->cur_conni>=0&&_stream->cur_connicontent_length||_offsetOP_INT64_MAX-content_length){ + return -1; + } + pos=content_length+_offset; }break; default:return -1; } @@ -3156,7 +3198,7 @@ static int op_http_stream_seek(void *_stream,opus_int64 _offset,int _whence){ op_http_conn_read_rate_update(stream->conns+ci); *&seek_time=*&stream->conns[ci].read_time; } - else ftime(&seek_time); + else op_time_get(&seek_time); /*If we seeked past the end of the stream, just disable the active connection.*/ if(pos>=content_length){ diff --git a/code/opusfile-0.9/src/info.c b/code/opusfile-0.12/src/info.c similarity index 98% rename from code/opusfile-0.9/src/info.c rename to code/opusfile-0.12/src/info.c index 2a4f737e..000c74b3 100644 --- a/code/opusfile-0.9/src/info.c +++ b/code/opusfile-0.12/src/info.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012-2020 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ #ifdef HAVE_CONFIG_H @@ -96,6 +96,9 @@ void opus_tags_clear(OpusTags *_tags){ int ci; ncomments=_tags->comments; if(_tags->user_comments!=NULL)ncomments++; + else{ + OP_ASSERT(ncomments==0); + } for(ci=ncomments;ci-->0;)_ogg_free(_tags->user_comments[ci]); _ogg_free(_tags->user_comments); _ogg_free(_tags->comment_lengths); @@ -116,7 +119,6 @@ static int op_tags_ensure_capacity(OpusTags *_tags,size_t _ncomments){ Trimming requires cleaning up the allocated strings in the old space, and is best handled separately if it's ever needed.*/ OP_ASSERT(_ncomments>=(size_t)cur_ncomments); - comment_lengths=_tags->comment_lengths; comment_lengths=(int *)_ogg_realloc(_tags->comment_lengths,size); if(OP_UNLIKELY(comment_lengths==NULL))return OP_EFAULT; if(_tags->comment_lengths==NULL){ @@ -282,7 +284,7 @@ int opus_tags_copy(OpusTags *_dst,const OpusTags *_src){ ret=opus_tags_copy_impl(&dst,_src); if(OP_UNLIKELY(ret<0))opus_tags_clear(&dst); else *_dst=*&dst; - return 0; + return ret; } int opus_tags_add(OpusTags *_tags,const char *_tag,const char *_value){ @@ -454,8 +456,7 @@ int opus_tags_get_track_gain(const OpusTags *_tags,int *_gain_q8){ } static int op_is_jpeg(const unsigned char *_buf,size_t _buf_sz){ - return _buf_sz>=11&&memcmp(_buf,"\xFF\xD8\xFF\xE0",4)==0 - &&(_buf[4]<<8|_buf[5])>=16&&memcmp(_buf+6,"JFIF",5)==0; + return _buf_sz>=3&&memcmp(_buf,"\xFF\xD8\xFF",3)==0; } /*Tries to extract the width, height, bits per pixel, and palette size of a @@ -592,6 +593,7 @@ static int opus_picture_tag_parse_impl(OpusPictureTag *_pic,const char *_tag, int colors_set; size_t i; /*Decode the BASE64 data.*/ + OP_ASSERT(_base64_sz>=11); for(i=0;i<_base64_sz;i++){ opus_uint32 value; int j; diff --git a/code/opusfile-0.9/src/internal.c b/code/opusfile-0.12/src/internal.c similarity index 90% rename from code/opusfile-0.9/src/internal.c rename to code/opusfile-0.12/src/internal.c index 96c80def..81c9c379 100644 --- a/code/opusfile-0.9/src/internal.c +++ b/code/opusfile-0.12/src/internal.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012-2020 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ #ifdef HAVE_CONFIG_H diff --git a/code/opusfile-0.9/src/internal.h b/code/opusfile-0.12/src/internal.h similarity index 98% rename from code/opusfile-0.9/src/internal.h rename to code/opusfile-0.12/src/internal.h index 9ac17e02..0c2d2bb3 100644 --- a/code/opusfile-0.9/src/internal.h +++ b/code/opusfile-0.12/src/internal.h @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012-2020 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ #if !defined(_opusfile_internal_h) diff --git a/code/opusfile-0.9/src/opusfile.c b/code/opusfile-0.12/src/opusfile.c similarity index 98% rename from code/opusfile-0.9/src/opusfile.c rename to code/opusfile-0.12/src/opusfile.c index 5ab222ce..642c7847 100644 --- a/code/opusfile-0.9/src/opusfile.c +++ b/code/opusfile-0.12/src/opusfile.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 1994-2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 1994-2020 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ******************************************************************** @@ -931,7 +931,7 @@ static int op_find_initial_pcm_offset(OggOpusFile *_of, prev_packet_gp=pcm_start; for(pi=0;pioy,(long)_initial_bytes); } /*Can we seek? - Stevens suggests the seek test is portable.*/ + Stevens suggests the seek test is portable. + It's actually not for files on win32, but we address that by fixing it in + our callback implementation (see stream.c).*/ seekable=_cb->seek!=NULL&&(*_cb->seek)(_stream,0,SEEK_CUR)!=-1; /*If seek is implemented, tell must also be implemented.*/ if(seekable){ @@ -1732,7 +1734,7 @@ opus_int64 op_raw_total(const OggOpusFile *_of,int _li){ ogg_int64_t op_pcm_total(const OggOpusFile *_of,int _li){ OggOpusLink *links; ogg_int64_t pcm_total; - ogg_int64_t diff=0; + ogg_int64_t diff; int nlinks; nlinks=_of->nlinks; if(OP_UNLIKELY(_of->ready_state(OP_INT64_MAX-(_samples>>1))/(48000*8))){ ogg_int64_t den; @@ -1789,7 +1792,6 @@ static opus_int32 op_calc_bitrate(opus_int64 _bytes,ogg_int64_t _samples){ den=_samples/(48000*8); return (opus_int32)((_bytes+(den>>1))/den); } - if(OP_UNLIKELY(_samples<=0))return OP_INT32_MAX; /*This can't actually overflow in normal operation: even with a pre-skip of 545 2.5 ms frames with 8 streams running at 1282*8+1 bytes per packet (1275 byte frames + Opus framing overhead + Ogg lacing values), that all @@ -1835,7 +1837,7 @@ static int op_get_link_from_serialno(const OggOpusFile *_of,int _cur_link, nlinks=_of->nlinks; li_lo=0; /*Start off by guessing we're just a multiplexed page in the current link.*/ - li_hi=_cur_link+1=links[_cur_link].offset)li_lo=_cur_link; @@ -1988,9 +1990,12 @@ static int op_fetch_and_process_page(OggOpusFile *_of, Drain the packets from the page anyway. If we don't, they'll still be there when we fetch the next page. Then, when we go to pull out packets, we might get more than 255, - which would overrun our packet buffer.*/ - total_duration=op_collect_audio_packets(_of,durations); - OP_ASSERT(total_duration>=0); + which would overrun our packet buffer. + We repeat this call until we get any actual packets, since we might + have buffered multiple out-of-sequence pages with no packets on + them.*/ + do total_duration=op_collect_audio_packets(_of,durations); + while(total_duration<0); if(!_ignore_holes){ /*Report the hole to the caller after we finish timestamping the packets.*/ @@ -2078,7 +2083,11 @@ static int op_fetch_and_process_page(OggOpusFile *_of, &&OP_LIKELY(diff0){ /*If we trimmed the entire packet, stop (the spec says encoders @@ -2175,7 +2184,7 @@ int op_raw_seek(OggOpusFile *_of,opus_int64 _pos){ static ogg_int64_t op_get_granulepos(const OggOpusFile *_of, ogg_int64_t _pcm_offset,int *_li){ const OggOpusLink *links; - ogg_int64_t duration=0; + ogg_int64_t duration; ogg_int64_t pcm_start; opus_int32 pre_skip; int nlinks; @@ -2254,7 +2263,7 @@ static int op_pcm_seek_page(OggOpusFile *_of, ogg_int64_t _target_gp,int _li){ const OggOpusLink *link; ogg_page og; - ogg_int64_t pcm_pre_skip=0; + ogg_int64_t pcm_pre_skip; ogg_int64_t pcm_start; ogg_int64_t pcm_end; ogg_int64_t best_gp; @@ -2342,7 +2351,7 @@ static int op_pcm_seek_page(OggOpusFile *_of, } } else{ - ogg_int64_t prev_page_gp=0; + ogg_int64_t prev_page_gp; /*We might get lucky and already have the packet with the target buffered. Worth checking. @@ -2597,15 +2606,14 @@ int op_pcm_seek(OggOpusFile *_of,ogg_int64_t _pcm_offset){ ogg_int64_t gp; gp=_of->prev_packet_gp; if(OP_LIKELY(gp!=-1)){ - int nbuffered; + ogg_int64_t discard_count; + int nbuffered; nbuffered=OP_MAX(_of->od_buffer_size-_of->od_buffer_pos,0); OP_ALWAYS_TRUE(!op_granpos_add(&gp,gp,-nbuffered)); /*We do _not_ add cur_discard_count to gp. Otherwise the total amount to discard could grow without bound, and it would be better just to do a full seek.*/ - if(OP_LIKELY(!op_granpos_diff(&diff,gp,pcm_start))){ - ogg_int64_t discard_count; - discard_count=_pcm_offset-diff; + if(OP_LIKELY(!op_granpos_diff(&discard_count,target_gp,gp))){ /*We use a threshold of 90 ms instead of 80, since 80 ms is the _minimum_ we would have discarded after a full seek. Assuming 20 ms frames (the default), we'd discard 90 ms on average.*/ @@ -2646,11 +2654,12 @@ int op_pcm_seek(OggOpusFile *_of,ogg_int64_t _pcm_offset){ ret=op_fetch_and_process_page(_of,NULL,-1,0,1); if(OP_UNLIKELY(ret<0))return OP_EBADLINK; } - OP_ALWAYS_TRUE(!op_granpos_diff(&diff,prev_packet_gp,pcm_start)); - /*We skipped too far. + /*We skipped too far, or couldn't get within 2 billion samples of the target. Either the timestamps were illegal or there was a hole in the data.*/ - if(diff>skip)return OP_EBADLINK; - OP_ASSERT(_pcm_offset-diffskip + ||_pcm_offset-diff>=OP_INT32_MAX){ + return OP_EBADLINK; + } /*TODO: If there are further holes/illegal timestamps, we still won't decode to the correct sample. However, at least op_pcm_tell() will report the correct value immediately @@ -2813,10 +2822,17 @@ static int op_read_native(OggOpusFile *_of, /*If we have buffered samples, return them.*/ if(nsamples>0){ if(nsamples*nchannels>_buf_size)nsamples=_buf_size/nchannels; - memcpy(_pcm,_of->od_buffer+nchannels*od_buffer_pos, - sizeof(*_pcm)*nchannels*nsamples); - od_buffer_pos+=nsamples; - _of->od_buffer_pos=od_buffer_pos; + OP_ASSERT(_pcm!=NULL||nsamples<=0); + /*Check nsamples again so we don't pass NULL to memcpy() if _buf_size + is zero. + That would technically be undefined behavior, even if the number of + bytes to copy were zero.*/ + if(nsamples>0){ + memcpy(_pcm,_of->od_buffer+nchannels*od_buffer_pos, + sizeof(*_pcm)*nchannels*nsamples); + od_buffer_pos+=nsamples; + _of->od_buffer_pos=od_buffer_pos; + } if(_li!=NULL)*_li=_of->cur_link; return nsamples; } @@ -2870,6 +2886,7 @@ static int op_read_native(OggOpusFile *_of, _of->samples_tracked+=trimmed_duration-od_buffer_pos; } else{ + OP_ASSERT(_pcm!=NULL); /*Otherwise decode directly into the user's buffer.*/ ret=op_decode(_of,_pcm,pop,duration,nchannels); if(OP_UNLIKELY(ret<0))return ret; @@ -3162,7 +3179,7 @@ static opus_uint32 op_rand(opus_uint32 _seed){ # define OP_GAIN (32753.0F) -# define OP_PRNG_GAIN (1.0F/0xFFFFFFFF) +# define OP_PRNG_GAIN (1.0F/(float)0xFFFFFFFF) /*48 kHz noise shaping filter, sd=2.34.*/ diff --git a/code/opusfile-0.9/src/stream.c b/code/opusfile-0.12/src/stream.c similarity index 81% rename from code/opusfile-0.9/src/stream.c rename to code/opusfile-0.12/src/stream.c index 6a85197a..6559e1db 100644 --- a/code/opusfile-0.9/src/stream.c +++ b/code/opusfile-0.12/src/stream.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 1994-2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 1994-2018 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ******************************************************************** @@ -63,7 +63,7 @@ static int op_fread(void *_stream,unsigned char *_ptr,int _buf_size){ static int op_fseek(void *_stream,opus_int64 _offset,int _whence){ #if defined(_WIN32) - /*_fseeki64() is not exposed until MSCVCRT80. + /*_fseeki64() is not exposed until MSVCRT80. This is the default starting with MSVC 2005 (_MSC_VER>=1400), but we want to allow linking against older MSVCRT versions for compatibility back to XP without installing extra runtime libraries. @@ -72,7 +72,7 @@ static int op_fseek(void *_stream,opus_int64 _offset,int _whence){ other libraries (that don't use MSVCRT80 from MSVC 2005 by default). i686-w64-mingw32 does have fseeko() and respects _FILE_OFFSET_BITS, but I don't know how to detect that at compile time. - We could just use fseeko64() (which is available in both), but its + We could just use fseeko64() (which is available in both), but it's implemented using fgetpos()/fsetpos() just like this code, except without the overflow checking, so we prefer our version.*/ opus_int64 pos; @@ -105,7 +105,7 @@ static int op_fseek(void *_stream,opus_int64 _offset,int _whence){ static opus_int64 op_ftell(void *_stream){ #if defined(_WIN32) - /*_ftelli64() is not exposed until MSCVCRT80, and ftello()/ftello64() have + /*_ftelli64() is not exposed until MSVCRT80, and ftello()/ftello64() have the same problems as fseeko()/fseeko64() in MingW. See above for a more detailed explanation.*/ opus_int64 pos; @@ -227,6 +227,57 @@ static wchar_t *op_utf8_to_utf16(const char *_src){ return dst; } +/*fsetpos() internally dispatches to the win32 API call SetFilePointer(). + According to SetFilePointer()'s documentation [0], the behavior is + undefined if you do not call it on "a file stored on a seeking device". + However, none of the MSVCRT seeking functions verify what kind of file is + being used before calling it (which I believe is a bug, since they are + supposed to fail and return an error, but it is a bug that has been there + for multiple decades now). + In practice, SetFilePointer() appears to succeed for things like stdin, + even when you are not just piping in a regular file, which prevents the use + of this API to determine whether it is possible to seek in a file at all. + Therefore, we take the approach recommended by the SetFilePointer() + documentation and confirm the type of file using GetFileType() first. + We do this once, when the file is opened, and return the corresponding + callback in order to avoid an extra win32 API call on every seek in the + common case. + Hopefully the return value of GetFileType() cannot actually change for the + lifetime of a file handle. + [0] https://docs.microsoft.com/en-us/windows/desktop/api/fileapi/nf-fileapi-setfilepointer +*/ +static int op_fseek_fail(void *_stream,opus_int64 _offset,int _whence){ + (void)_stream; + (void)_offset; + (void)_whence; + return -1; +} + +static const OpusFileCallbacks OP_UNSEEKABLE_FILE_CALLBACKS={ + op_fread, + op_fseek_fail, + op_ftell, + (op_close_func)fclose +}; + +# define WIN32_LEAN_AND_MEAN +# define WIN32_EXTRA_LEAN +# include + +static const OpusFileCallbacks *op_get_file_callbacks(FILE *_fp){ + intptr_t h_file; + h_file=_get_osfhandle(_fileno(_fp)); + if(h_file!=-1 + &&(GetFileType((HANDLE)h_file)&~FILE_TYPE_REMOTE)==FILE_TYPE_DISK){ + return &OP_FILE_CALLBACKS; + } + return &OP_UNSEEKABLE_FILE_CALLBACKS; +} +#else +static const OpusFileCallbacks *op_get_file_callbacks(FILE *_fp){ + (void)_fp; + return &OP_FILE_CALLBACKS; +} #endif void *op_fopen(OpusFileCallbacks *_cb,const char *_path,const char *_mode){ @@ -247,14 +298,14 @@ void *op_fopen(OpusFileCallbacks *_cb,const char *_path,const char *_mode){ _ogg_free(wpath); } #endif - if(fp!=NULL)*_cb=*&OP_FILE_CALLBACKS; + if(fp!=NULL)*_cb=*op_get_file_callbacks(fp); return fp; } void *op_fdopen(OpusFileCallbacks *_cb,int _fd,const char *_mode){ FILE *fp; fp=fdopen(_fd,_mode); - if(fp!=NULL)*_cb=*&OP_FILE_CALLBACKS; + if(fp!=NULL)*_cb=*op_get_file_callbacks(fp); return fp; } @@ -277,7 +328,7 @@ void *op_freopen(OpusFileCallbacks *_cb,const char *_path,const char *_mode, _ogg_free(wpath); } #endif - if(fp!=NULL)*_cb=*&OP_FILE_CALLBACKS; + if(fp!=NULL)*_cb=*op_get_file_callbacks(fp); return fp; } @@ -322,8 +373,8 @@ static int op_mem_seek(void *_stream,opus_int64 _offset,int _whence){ size=stream->size; OP_ASSERT(size>=0); /*Check for overflow:*/ - if(_offset>size||_offsetOP_MEM_DIFF_MAX-size)return -1; + pos=(ptrdiff_t)(size+_offset); }break; default:return -1; } diff --git a/code/opusfile-0.9/src/wincerts.c b/code/opusfile-0.12/src/wincerts.c similarity index 96% rename from code/opusfile-0.9/src/wincerts.c rename to code/opusfile-0.12/src/wincerts.c index 7f915bd3..409a4e09 100644 --- a/code/opusfile-0.9/src/wincerts.c +++ b/code/opusfile-0.12/src/wincerts.c @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2013 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2013-2016 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ @@ -14,7 +14,7 @@ in their bugtracker for over two years that implements this, without any action, so I'm giving up and re-implementing it locally. - [1] */ + [1] */ #ifdef HAVE_CONFIG_H #include "config.h" diff --git a/code/opusfile-0.9/src/winerrno.h b/code/opusfile-0.12/src/winerrno.h similarity index 96% rename from code/opusfile-0.9/src/winerrno.h rename to code/opusfile-0.12/src/winerrno.h index 32a90b4e..a5177428 100644 --- a/code/opusfile-0.9/src/winerrno.h +++ b/code/opusfile-0.12/src/winerrno.h @@ -5,8 +5,8 @@ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * * * - * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012 * - * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * + * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012-2013 * + * by the Xiph.Org Foundation and contributors https://xiph.org/ * * * ********************************************************************/ #if !defined(_opusfile_winerrno_h)