ref: 81624caf9ceb452fa8f909dcd8ad14511136da87
parent: 0563d71b255c2ef0cb65aab706ecbd44e0328c8d
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Sat Oct 7 13:45:39 EDT 2023
Silencing alignment warnings on x86 intrinsics Those intrinsics don't actually require alignment so we're OK
--- a/celt/x86/pitch_avx.c
+++ b/celt/x86/pitch_avx.c
@@ -56,7 +56,7 @@
if (i != len) { static const int mask[15] = {-1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 0, 0, 0, 0};__m256i m;
- m = _mm256_loadu_si256((__m256i*)(mask + 7+i-len));
+ m = _mm256_loadu_si256((__m256i*)(void*)(mask + 7+i-len));
x0 = _mm256_maskload_ps(x+i, m);
xsum0 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i , m), xsum0);
xsum1 = _mm256_fmadd_ps(x0, _mm256_maskload_ps(y+i+1, m), xsum1);
--- a/celt/x86/vq_sse2.c
+++ b/celt/x86/vq_sse2.c
@@ -75,7 +75,7 @@
sums = _mm_add_ps(sums, x4);
/* Clear y and iy in case we don't do the projection. */
_mm_storeu_ps(&y[j], _mm_setzero_ps());
- _mm_storeu_si128((__m128i*)&iy[j], _mm_setzero_si128());
+ _mm_storeu_si128((__m128i*)(void*)&iy[j], _mm_setzero_si128());
_mm_storeu_ps(&X[j], x4);
_mm_storeu_ps(&signy[j], s4);
}
@@ -116,7 +116,7 @@
rx4 = _mm_mul_ps(x4, rcp4);
iy4 = _mm_cvttps_epi32(rx4);
pulses_sum = _mm_add_epi32(pulses_sum, iy4);
- _mm_storeu_si128((__m128i*)&iy[j], iy4);
+ _mm_storeu_si128((__m128i*)(void*)&iy[j], iy4);
y4 = _mm_cvtepi32_ps(iy4);
xy4 = _mm_add_ps(xy4, _mm_mul_ps(x4, y4));
yy4 = _mm_add_ps(yy4, _mm_mul_ps(y4, y4));
@@ -205,10 +205,10 @@
{__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;
--- a/celt/x86/x86cpu.h
+++ b/celt/x86/x86cpu.h
@@ -68,6 +68,6 @@
(_mm_cvtepi8_epi32(_mm_cvtsi32_si128(OP_LOADU_EPI32(x))))
#define OP_CVTEPI16_EPI32_M64(x) \
- (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x))))
+ (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(void*)(x))))
#endif
--- a/dnn/vec_avx.h
+++ b/dnn/vec_avx.h
@@ -335,7 +335,7 @@
xi = _mm256_permute4x64_epi64(xi, 0xD8);
xi = _mm256_packus_epi16(xi, _mm256_setzero_si256());
xi = _mm256_permutevar8x32_epi32(xi, _mm256_setr_epi32(0,1, 0,0, 0,0, 0,0));
- _mm256_storeu_si256 ((__m256i *)&x[i], xi);
+ _mm256_storeu_si256 ((__m256i *)(void*)&x[i], xi);
}
}
@@ -794,20 +794,20 @@
{__m256i vxj;
__m256i vw;
- vxj = _mm256_set1_epi32(*(int*)&x[*idx++]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[*idx++]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[*idx++]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[*idx++]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[*idx++]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[*idx++]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[*idx++]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[*idx++]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
}
@@ -818,8 +818,8 @@
__m256i vw;
int pos;
pos = (*idx++);
- vxj = _mm256_set1_epi32(*(int*)&x[pos]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[pos]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
}
@@ -845,20 +845,20 @@
{__m256i vxj;
__m256i vw;
- vxj = _mm256_set1_epi32(*(int*)&x[j]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[j]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[j+4]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[j+4]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[j+8]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[j+8]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
- vxj = _mm256_set1_epi32(*(int*)&x[j+12]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[j+12]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
}
@@ -867,8 +867,8 @@
{__m256i vxj;
__m256i vw;
- vxj = _mm256_set1_epi32(*(int*)&x[j]);
- vw = _mm256_loadu_si256((const __m256i *)w);
+ vxj = _mm256_set1_epi32(*(int*)(void*)&x[j]);
+ vw = _mm256_loadu_si256((const __m256i *)(void*)w);
vy0 = opus_mm256_dpbusds_epi32(vy0, vxj, vw);
w += 32;
}
--- a/silk/x86/NSQ_del_dec_sse4_1.c
+++ b/silk/x86/NSQ_del_dec_sse4_1.c
@@ -428,7 +428,7 @@
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 );
@@ -483,7 +483,7 @@
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 */
@@ -497,7 +497,7 @@
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 );
@@ -512,7 +512,7 @@
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 );
@@ -525,7 +525,7 @@
tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp );
/* step 4 */
- psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -15 ] ) );
+ 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 );
@@ -830,7 +830,7 @@
xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC );
- _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 );
+ _mm_storeu_si128( (__m128i *)(void*)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 );
}
for( ; i < psEncC->subfr_length; i++ ) {@@ -862,7 +862,7 @@
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 ) );
@@ -874,7 +874,7 @@
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++ ) {--- a/silk/x86/NSQ_sse4_1.c
+++ b/silk/x86/NSQ_sse4_1.c
@@ -338,21 +338,21 @@
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 );
@@ -360,8 +360,8 @@
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 );
@@ -370,8 +370,8 @@
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 );
@@ -443,7 +443,7 @@
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] */
@@ -595,8 +595,8 @@
/* 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 ) ); */
{@@ -612,8 +612,8 @@
/* 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 ) );
@@ -644,7 +644,7 @@
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++)
@@ -698,7 +698,7 @@
xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC );
- _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 );
+ _mm_storeu_si128( (__m128i *)(void*)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 );
}
for( ; i < psEncC->subfr_length; i++ ) {@@ -729,7 +729,7 @@
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 ) );
@@ -741,7 +741,7 @@
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++ ) {--- a/silk/x86/VAD_sse4_1.c
+++ b/silk/x86/VAD_sse4_1.c
@@ -144,7 +144,7 @@
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 );
--- a/silk/x86/VQ_WMat_EC_sse4_1.c
+++ b/silk/x86/VQ_WMat_EC_sse4_1.c
@@ -65,7 +65,7 @@
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 *)(&XX_Q17[ 1 ] ) );
+ 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 */
--
⑨