fix(windows): vendor audiopus_sys + patch libopus for clang-cl SIMD
Some checks failed
Mirror to GitHub / mirror (push) Failing after 35s
Build Release Binaries / build-amd64 (push) Has been cancelled

cargo-xwin drives the Windows MSVC cross-compile via clang-cl, under
which CMake sets MSVC=1 — causing libopus 1.3.1's `if(NOT MSVC)` guards
to skip the per-file `-msse4.1` / `-mssse3` COMPILE_FLAGS that its x86
SIMD source files need. Clang-cl (unlike real cl.exe) still honors
Clang's target-feature system, so those files then fail to compile
with "always_inline function '_mm_cvtepi16_epi32' requires target
feature 'sse4.1'" errors across silk/NSQ_sse4_1.c, NSQ_del_dec_sse4_1.c,
and VQ_WMat_EC_sse4_1.c.

Earlier attempts to fix this downstream (cargo-xwin toolchain file,
override.cmake CMAKE_C_COMPILE_OBJECT <FLAGS> replace, CFLAGS env vars)
all failed because cargo-xwin rewrites override.cmake from scratch on
every `cargo xwin build` invocation and cmake-rs's -DCMAKE_C_FLAGS=
assembly happens before toolchain FORCE sets propagate.

Fixing it upstream at the source: vendor audiopus_sys 0.2.2 into
vendor/audiopus_sys, patch its bundled opus/CMakeLists.txt to introduce
an MSVC_CL var (true only when CMAKE_C_COMPILER_ID == "MSVC", i.e. real
cl.exe), and flip the eight `if(NOT MSVC)` SIMD guards to
`if(NOT MSVC_CL)`. Clang-cl then gets the GCC-style per-file flags and
the SSE4.1 sources build cleanly. Also flip the `if(MSVC)` global /arch
block at line 445 to `if(MSVC_CL)` so only cl.exe applies /arch:AVX and
clang-cl relies purely on per-file flags (no global/per-file mixing).

Wire via [patch.crates-io] in the workspace root Cargo.toml; the patch
is resolved relative to the workspace root as `vendor/audiopus_sys`.

Upstream context: xiph/opus#256, xiph/opus PR #257 (both stale).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Siavash Sameni
2026-04-10 14:12:59 +04:00
parent 53f57eea07
commit 0683dde5d3
423 changed files with 103000 additions and 2 deletions

View File

@@ -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;
}

View File

@@ -0,0 +1,184 @@
/***********************************************************************
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 __PREFILTER_FIX_MIPSR1_H__
#define __PREFILTER_FIX_MIPSR1_H__
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "main_FIX.h"
#include "stack_alloc.h"
#include "tuning_parameters.h"
#define OVERRIDE_silk_warped_LPC_analysis_filter_FIX
void silk_warped_LPC_analysis_filter_FIX(
opus_int32 state[], /* I/O State [order + 1] */
opus_int32 res_Q2[], /* O Residual signal [length] */
const opus_int16 coef_Q13[], /* I Coefficients [order] */
const opus_int16 input[], /* I Input signal [length] */
const opus_int16 lambda_Q16, /* I Warping factor */
const opus_int length, /* I Length of input signal */
const opus_int order, /* I Filter order (even) */
int arch
)
{
opus_int n, i;
opus_int32 acc_Q11, acc_Q22, tmp1, tmp2, tmp3, tmp4;
opus_int32 state_cur, state_next;
(void)arch;
/* Order must be even */
/* Length must be even */
silk_assert( ( order & 1 ) == 0 );
silk_assert( ( length & 1 ) == 0 );
for( n = 0; n < length; n+=2 ) {
/* Output of lowpass section */
tmp2 = silk_SMLAWB( state[ 0 ], state[ 1 ], lambda_Q16 );
state_cur = silk_LSHIFT( input[ n ], 14 );
/* Output of allpass section */
tmp1 = silk_SMLAWB( state[ 1 ], state[ 2 ] - tmp2, lambda_Q16 );
state_next = tmp2;
acc_Q11 = silk_RSHIFT( order, 1 );
acc_Q11 = silk_SMLAWB( acc_Q11, tmp2, coef_Q13[ 0 ] );
/* Output of lowpass section */
tmp4 = silk_SMLAWB( state_cur, state_next, lambda_Q16 );
state[ 0 ] = silk_LSHIFT( input[ n+1 ], 14 );
/* Output of allpass section */
tmp3 = silk_SMLAWB( state_next, tmp1 - tmp4, lambda_Q16 );
state[ 1 ] = tmp4;
acc_Q22 = silk_RSHIFT( order, 1 );
acc_Q22 = silk_SMLAWB( acc_Q22, tmp4, coef_Q13[ 0 ] );
/* Loop over allpass sections */
for( i = 2; i < order; i += 2 ) {
/* Output of allpass section */
tmp2 = silk_SMLAWB( state[ i ], state[ i + 1 ] - tmp1, lambda_Q16 );
state_cur = tmp1;
acc_Q11 = silk_SMLAWB( acc_Q11, tmp1, coef_Q13[ i - 1 ] );
/* Output of allpass section */
tmp1 = silk_SMLAWB( state[ i + 1 ], state[ i + 2 ] - tmp2, lambda_Q16 );
state_next = tmp2;
acc_Q11 = silk_SMLAWB( acc_Q11, tmp2, coef_Q13[ i ] );
/* Output of allpass section */
tmp4 = silk_SMLAWB( state_cur, state_next - tmp3, lambda_Q16 );
state[ i ] = tmp3;
acc_Q22 = silk_SMLAWB( acc_Q22, tmp3, coef_Q13[ i - 1 ] );
/* Output of allpass section */
tmp3 = silk_SMLAWB( state_next, tmp1 - tmp4, lambda_Q16 );
state[ i + 1 ] = tmp4;
acc_Q22 = silk_SMLAWB( acc_Q22, tmp4, coef_Q13[ i ] );
}
acc_Q11 = silk_SMLAWB( acc_Q11, tmp1, coef_Q13[ order - 1 ] );
res_Q2[ n ] = silk_LSHIFT( (opus_int32)input[ n ], 2 ) - silk_RSHIFT_ROUND( acc_Q11, 9 );
state[ order ] = tmp3;
acc_Q22 = silk_SMLAWB( acc_Q22, tmp3, coef_Q13[ order - 1 ] );
res_Q2[ n+1 ] = silk_LSHIFT( (opus_int32)input[ n+1 ], 2 ) - silk_RSHIFT_ROUND( acc_Q22, 9 );
}
}
/* Prefilter for finding Quantizer input signal */
#define OVERRIDE_silk_prefilt_FIX
static inline void silk_prefilt_FIX(
silk_prefilter_state_FIX *P, /* I/O state */
opus_int32 st_res_Q12[], /* I short term residual signal */
opus_int32 xw_Q3[], /* O prefiltered signal */
opus_int32 HarmShapeFIRPacked_Q12, /* I Harmonic shaping coeficients */
opus_int Tilt_Q14, /* I Tilt shaping coeficient */
opus_int32 LF_shp_Q14, /* I Low-frequancy shaping coeficients */
opus_int lag, /* I Lag for harmonic shaping */
opus_int length /* I Length of signals */
)
{
opus_int i, idx, LTP_shp_buf_idx;
opus_int32 n_LTP_Q12, n_Tilt_Q10, n_LF_Q10;
opus_int32 sLF_MA_shp_Q12, sLF_AR_shp_Q12;
opus_int16 *LTP_shp_buf;
/* To speed up use temp variables instead of using the struct */
LTP_shp_buf = P->sLTP_shp;
LTP_shp_buf_idx = P->sLTP_shp_buf_idx;
sLF_AR_shp_Q12 = P->sLF_AR_shp_Q12;
sLF_MA_shp_Q12 = P->sLF_MA_shp_Q12;
if( lag > 0 ) {
for( i = 0; i < length; i++ ) {
/* unrolled loop */
silk_assert( HARM_SHAPE_FIR_TAPS == 3 );
idx = lag + LTP_shp_buf_idx;
n_LTP_Q12 = silk_SMULBB( LTP_shp_buf[ ( idx - HARM_SHAPE_FIR_TAPS / 2 - 1) & LTP_MASK ], HarmShapeFIRPacked_Q12 );
n_LTP_Q12 = silk_SMLABT( n_LTP_Q12, LTP_shp_buf[ ( idx - HARM_SHAPE_FIR_TAPS / 2 ) & LTP_MASK ], HarmShapeFIRPacked_Q12 );
n_LTP_Q12 = silk_SMLABB( n_LTP_Q12, LTP_shp_buf[ ( idx - HARM_SHAPE_FIR_TAPS / 2 + 1) & LTP_MASK ], HarmShapeFIRPacked_Q12 );
n_Tilt_Q10 = silk_SMULWB( sLF_AR_shp_Q12, Tilt_Q14 );
n_LF_Q10 = silk_SMLAWB( silk_SMULWT( sLF_AR_shp_Q12, LF_shp_Q14 ), sLF_MA_shp_Q12, LF_shp_Q14 );
sLF_AR_shp_Q12 = silk_SUB32( st_res_Q12[ i ], silk_LSHIFT( n_Tilt_Q10, 2 ) );
sLF_MA_shp_Q12 = silk_SUB32( sLF_AR_shp_Q12, silk_LSHIFT( n_LF_Q10, 2 ) );
LTP_shp_buf_idx = ( LTP_shp_buf_idx - 1 ) & LTP_MASK;
LTP_shp_buf[ LTP_shp_buf_idx ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sLF_MA_shp_Q12, 12 ) );
xw_Q3[i] = silk_RSHIFT_ROUND( silk_SUB32( sLF_MA_shp_Q12, n_LTP_Q12 ), 9 );
}
}
else
{
for( i = 0; i < length; i++ ) {
n_LTP_Q12 = 0;
n_Tilt_Q10 = silk_SMULWB( sLF_AR_shp_Q12, Tilt_Q14 );
n_LF_Q10 = silk_SMLAWB( silk_SMULWT( sLF_AR_shp_Q12, LF_shp_Q14 ), sLF_MA_shp_Q12, LF_shp_Q14 );
sLF_AR_shp_Q12 = silk_SUB32( st_res_Q12[ i ], silk_LSHIFT( n_Tilt_Q10, 2 ) );
sLF_MA_shp_Q12 = silk_SUB32( sLF_AR_shp_Q12, silk_LSHIFT( n_LF_Q10, 2 ) );
LTP_shp_buf_idx = ( LTP_shp_buf_idx - 1 ) & LTP_MASK;
LTP_shp_buf[ LTP_shp_buf_idx ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sLF_MA_shp_Q12, 12 ) );
xw_Q3[i] = silk_RSHIFT_ROUND( sLF_MA_shp_Q12, 9 );
}
}
/* Copy temp variable back to state */
P->sLF_AR_shp_Q12 = sLF_AR_shp_Q12;
P->sLF_MA_shp_Q12 = sLF_MA_shp_Q12;
P->sLTP_shp_buf_idx = LTP_shp_buf_idx;
}
#endif /* __PREFILTER_FIX_MIPSR1_H__ */

View File

@@ -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__ */