wuggy/Dactyloidae
wuggy/Dactyloidae
1
0
Fork 0
mirror of https://repo.dactyloidae.xyz/Dactyloidae/UXP.git synced 2026-09-27 19:07:31 +09:00
Code Issues Projects Releases Packages Wiki Activity

Issue #1890 - Update libopus

Browse source 
This resolves #1890
This commit is contained in:
Moonchild 2022-05-04 17:03:37 +00:00 • committed by roytam1
commit a8b1099a1e
228 changed files with 10990 additions and 5670 deletions
Download patch file Download diff file Expand all files Collapse all files

71
media/libopus/silk/fixed/x86/burg_modified_FIX_sse.c → media/libopus/silk/fixed/x86/burg_modified_FIX_sse4_1.c
View file

@ -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
@ -42,7 +42,7 @@
#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */
#define QA 25
#define N_BITS_HEAD_ROOM 2
#define N_BITS_HEAD_ROOM 3
#define MIN_RSHIFTS -16
#define MAX_RSHIFTS (32 - QA)
@ -59,7 +59,7 @@ void silk_burg_modified_sse4_1(
int arch /* I Run-time architecture */
)
{
opus_int k, n, s, lz, rshifts, rshifts_extra, reached_max_gain;
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 ];
@ -68,30 +68,26 @@ void silk_burg_modified_sse4_1(
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);
silk_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE );
celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE );
/* Compute autocorrelations, added over subframes */
silk_sum_sqr_shift( &C0, &rshifts, x, nb_subfr * subfr_length );
if( rshifts > MAX_RSHIFTS ) {
C0 = silk_LSHIFT32( C0, rshifts - MAX_RSHIFTS );
silk_assert( C0 > 0 );
rshifts = MAX_RSHIFTS;
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 {
lz = silk_CLZ32( C0 ) - 1;
rshifts_extra = N_BITS_HEAD_ROOM - lz;
if( rshifts_extra > 0 ) {
rshifts_extra = silk_min( rshifts_extra, MAX_RSHIFTS - rshifts );
C0 = silk_RSHIFT32( C0, rshifts_extra );
} else {
rshifts_extra = silk_max( rshifts_extra, MIN_RSHIFTS - rshifts );
C0 = silk_LSHIFT32( C0, -rshifts_extra );
}
rshifts += rshifts_extra;
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 ) {
@ -99,7 +95,7 @@ void silk_burg_modified_sse4_1(
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 {
@ -203,8 +199,11 @@ void silk_burg_modified_sse4_1(
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 */
tmp1 = silk_MLA( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */
tmp2 = silk_MLA( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* 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 */
@ -350,7 +349,7 @@ void silk_burg_modified_sse4_1(
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++ ) {
@ -374,4 +373,28 @@ void silk_burg_modified_sse4_1(
*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
}

160
media/libopus/silk/fixed/x86/prefilter_FIX_sse.c
View file

@ -1,160 +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 <xmmintrin.h>
#include <emmintrin.h>
#include <smmintrin.h>
#include "main.h"
#include "celt/x86/x86cpu.h"
void silk_warped_LPC_analysis_filter_FIX_sse4_1(
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) */
)
{
opus_int n, i;
opus_int32 acc_Q11, tmp1, tmp2;
/* Order must be even */
silk_assert( ( order & 1 ) == 0 );
if (order == 10)
{
if (0 == lambda_Q16)
{
__m128i coef_Q13_3210, coef_Q13_7654;
__m128i coef_Q13_0123, coef_Q13_4567;
__m128i state_0123, state_4567;
__m128i xmm_product1, xmm_product2;
__m128i xmm_tempa, xmm_tempb;
register opus_int32 sum;
register opus_int32 state_8, state_9, state_a;
register opus_int64 coef_Q13_8, coef_Q13_9;
silk_assert( length > 0 );
coef_Q13_3210 = OP_CVTEPI16_EPI32_M64( &coef_Q13[ 0 ] );
coef_Q13_7654 = OP_CVTEPI16_EPI32_M64( &coef_Q13[ 4 ] );
coef_Q13_0123 = _mm_shuffle_epi32( coef_Q13_3210, _MM_SHUFFLE( 0, 1, 2, 3 ) );
coef_Q13_4567 = _mm_shuffle_epi32( coef_Q13_7654, _MM_SHUFFLE( 0, 1, 2, 3 ) );
coef_Q13_8 = (opus_int64) coef_Q13[ 8 ];
coef_Q13_9 = (opus_int64) coef_Q13[ 9 ];
state_0123 = _mm_loadu_si128( (__m128i *)(&state[ 0 ] ) );
state_4567 = _mm_loadu_si128( (__m128i *)(&state[ 4 ] ) );
state_0123 = _mm_shuffle_epi32( state_0123, _MM_SHUFFLE( 0, 1, 2, 3 ) );
state_4567 = _mm_shuffle_epi32( state_4567, _MM_SHUFFLE( 0, 1, 2, 3 ) );
state_8 = state[ 8 ];
state_9 = state[ 9 ];
state_a = 0;
for( n = 0; n < length; n++ )
{
xmm_product1 = _mm_mul_epi32( coef_Q13_0123, state_0123 ); /* 64-bit multiply, only 2 pairs */
xmm_product2 = _mm_mul_epi32( coef_Q13_4567, state_4567 );
xmm_tempa = _mm_shuffle_epi32( state_0123, _MM_SHUFFLE( 0, 1, 2, 3 ) );
xmm_tempb = _mm_shuffle_epi32( state_4567, _MM_SHUFFLE( 0, 1, 2, 3 ) );
xmm_product1 = _mm_srli_epi64( xmm_product1, 16 ); /* >> 16, zero extending works */
xmm_product2 = _mm_srli_epi64( xmm_product2, 16 );
xmm_tempa = _mm_mul_epi32( coef_Q13_3210, xmm_tempa );
xmm_tempb = _mm_mul_epi32( coef_Q13_7654, xmm_tempb );
xmm_tempa = _mm_srli_epi64( xmm_tempa, 16 );
xmm_tempb = _mm_srli_epi64( xmm_tempb, 16 );
xmm_tempa = _mm_add_epi32( xmm_tempa, xmm_product1 );
xmm_tempb = _mm_add_epi32( xmm_tempb, xmm_product2 );
xmm_tempa = _mm_add_epi32( xmm_tempa, xmm_tempb );
sum = (coef_Q13_8 * state_8) >> 16;
sum += (coef_Q13_9 * state_9) >> 16;
xmm_tempa = _mm_add_epi32( xmm_tempa, _mm_shuffle_epi32( xmm_tempa, _MM_SHUFFLE( 0, 0, 0, 2 ) ) );
sum += _mm_cvtsi128_si32( xmm_tempa);
res_Q2[ n ] = silk_LSHIFT( (opus_int32)input[ n ], 2 ) - silk_RSHIFT_ROUND( ( 5 + sum ), 9);
/* move right */
state_a = state_9;
state_9 = state_8;
state_8 = _mm_cvtsi128_si32( state_4567 );
state_4567 = _mm_alignr_epi8( state_0123, state_4567, 4 );
state_0123 = _mm_alignr_epi8( _mm_cvtsi32_si128( silk_LSHIFT( input[ n ], 14 ) ), state_0123, 4 );
}
_mm_storeu_si128( (__m128i *)( &state[ 0 ] ), _mm_shuffle_epi32( state_0123, _MM_SHUFFLE( 0, 1, 2, 3 ) ) );
_mm_storeu_si128( (__m128i *)( &state[ 4 ] ), _mm_shuffle_epi32( state_4567, _MM_SHUFFLE( 0, 1, 2, 3 ) ) );
state[ 8 ] = state_8;
state[ 9 ] = state_9;
state[ 10 ] = state_a;
return;
}
}
for( n = 0; n < length; n++ ) {
/* Output of lowpass section */
tmp2 = silk_SMLAWB( state[ 0 ], state[ 1 ], lambda_Q16 );
state[ 0 ] = silk_LSHIFT( input[ n ], 14 );
/* Output of allpass section */
tmp1 = silk_SMLAWB( state[ 1 ], state[ 2 ] - tmp2, lambda_Q16 );
state[ 1 ] = tmp2;
acc_Q11 = silk_RSHIFT( order, 1 );
acc_Q11 = silk_SMLAWB( acc_Q11, tmp2, 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[ i ] = 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[ i + 1 ] = tmp2;
acc_Q11 = silk_SMLAWB( acc_Q11, tmp2, coef_Q13[ i ] );
}
state[ order ] = tmp1;
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 );
}
}

40
media/libopus/silk/fixed/x86/vector_ops_FIX_sse.c → media/libopus/silk/fixed/x86/vector_ops_FIX_sse4_1.c
View file

@ -37,39 +37,36 @@
#include "SigProc_FIX.h"
#include "pitch.h"
opus_int64 silk_inner_prod16_aligned_64_sse4_1(
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, dataSize8;
opus_int i, dataSize4;
opus_int64 sum;
__m128i xmm_tempa;
__m128i inVec1_76543210, acc1;
__m128i inVec2_76543210, acc2;
__m128i xmm_prod_20, xmm_prod_31;
__m128i inVec1_3210, acc1;
__m128i inVec2_3210, acc2;
sum = 0;
dataSize8 = len & ~7;
dataSize4 = len & ~3;
acc1 = _mm_setzero_si128();
acc2 = _mm_setzero_si128();
for( i = 0; i < dataSize8; i += 8 ) {
inVec1_76543210 = _mm_loadu_si128( (__m128i *)(&inVec1[i + 0] ) );
inVec2_76543210 = _mm_loadu_si128( (__m128i *)(&inVec2[i + 0] ) );
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 );
/* only when all 4 operands are -32768 (0x8000), this results in wrap around */
inVec1_76543210 = _mm_madd_epi16( inVec1_76543210, inVec2_76543210 );
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 );
xmm_tempa = _mm_cvtepi32_epi64( inVec1_76543210 );
/* equal shift right 8 bytes */
inVec1_76543210 = _mm_shuffle_epi32( inVec1_76543210, _MM_SHUFFLE( 0, 0, 3, 2 ) );
inVec1_76543210 = _mm_cvtepi32_epi64( inVec1_76543210 );
acc1 = _mm_add_epi64( acc1, xmm_tempa );
acc2 = _mm_add_epi64( acc2, inVec1_76543210 );
acc1 = _mm_add_epi64( acc1, xmm_prod_20 );
acc2 = _mm_add_epi64( acc2, xmm_prod_31 );
}
acc1 = _mm_add_epi64( acc1, acc2 );
@ -84,5 +81,12 @@ opus_int64 silk_inner_prod16_aligned_64_sse4_1(
sum = silk_SMLABB( 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;
}