generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
15f9bea73b
vpx/vp8/encoder/x86/x86_csystemdependent.c

353 lines
14 KiB
C
Raw Normal View History

Initial WebM release
2010-05-18 17:58:33 +02:00
/*
Use WebM in copyright notice for consistency Changes 'The VP8 project' to 'The WebM project', for consistency with other webmproject.org repositories. Fixes issue #97. Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
2010-09-09 14:16:39 +02:00
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
Initial WebM release
2010-05-18 17:58:33 +02:00
*
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
* Use of this source code is governed by a BSD-style license
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
* in the file PATENTS. All contributing project authors may
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
* be found in the AUTHORS file in the root of the source tree.
Initial WebM release
2010-05-18 17:58:33 +02:00
*/
#include "vpx_ports/config.h"
#include "vpx_ports/x86.h"
#include "variance.h"
#include "onyx_int.h"
#if HAVE_MMX
void vp8_short_fdct8x4_mmx(short *input, short *output, int pitch)
{
FDCT optimizations. Fixed up the fdct for mmx and 8x4 sse2 to match them most recent changes. Change-Id: Ibee2d6c536fe14dcf75cd6eb1c73f4848a56d719
2010-10-21 19:53:15 +02:00
vp8_short_fdct4x4_mmx(input, output, pitch);
vp8_short_fdct4x4_mmx(input + 4, output + 16, pitch);
Initial WebM release
2010-05-18 17:58:33 +02:00
}
int vp8_fast_quantize_b_impl_mmx(short *coeff_ptr, short *zbin_ptr,
short *qcoeff_ptr, short *dequant_ptr,
short *scan_mask, short *round_ptr,
short *quant_ptr, short *dqcoeff_ptr);
void vp8_fast_quantize_b_mmx(BLOCK *b, BLOCKD *d)
{
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *scan_mask = vp8_default_zig_zag_mask;//d->scan_order_mask_ptr;
short *coeff_ptr = b->coeff;
short *zbin_ptr = b->zbin;
short *round_ptr = b->round;
Use the fast quantizer for inter mode selection Use the fast quantizer for inter mode selection and the regular quantizer for the rest of the encode for good quality, speed 1. Both performance and quality were improved. The quality gains will make up for the quality loss mentioned in I9dc089007ca08129fb6c11fe7692777ebb8647b0. Change-Id: Ia90bc9cf326a7c65d60d31fa32f6465ab6984d21
2010-12-28 20:51:46 +01:00
short *quant_ptr = b->quant_fast;
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *qcoeff_ptr = d->qcoeff;
Initial WebM release
2010-05-18 17:58:33 +02:00
short *dqcoeff_ptr = d->dqcoeff;
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *dequant_ptr = d->dequant;
Initial WebM release
2010-05-18 17:58:33 +02:00
d->eob = vp8_fast_quantize_b_impl_mmx(
coeff_ptr,
zbin_ptr,
qcoeff_ptr,
dequant_ptr,
scan_mask,
round_ptr,
quant_ptr,
dqcoeff_ptr
);
}
int vp8_mbblock_error_mmx_impl(short *coeff_ptr, short *dcoef_ptr, int dc);
int vp8_mbblock_error_mmx(MACROBLOCK *mb, int dc)
{
short *coeff_ptr = mb->block[0].coeff;
short *dcoef_ptr = mb->e_mbd.block[0].dqcoeff;
return vp8_mbblock_error_mmx_impl(coeff_ptr, dcoef_ptr, dc);
}
int vp8_mbuverror_mmx_impl(short *s_ptr, short *d_ptr);
int vp8_mbuverror_mmx(MACROBLOCK *mb)
{
short *s_ptr = &mb->coeff[256];
short *d_ptr = &mb->e_mbd.dqcoeff[256];
return vp8_mbuverror_mmx_impl(s_ptr, d_ptr);
}
void vp8_subtract_b_mmx_impl(unsigned char *z, int src_stride,
short *diff, unsigned char *predictor,
int pitch);
void vp8_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch)
{
unsigned char *z = *(be->base_src) + be->src;
unsigned int src_stride = be->src_stride;
short *diff = &be->src_diff[0];
unsigned char *predictor = &bd->predictor[0];
vp8_subtract_b_mmx_impl(z, src_stride, diff, predictor, pitch);
}
#endif
#if HAVE_SSE2
Added vp8_fast_quantize_b_sse2 Moved vp8_fast_quantize_b_sse from quantize_mmx.asm into quantize_sse2.asm and renamed. Updated the assembly code to match the C version. Change-Id: I1766d9e1ca60e173f65badc0ca0c160c2b51b200
2010-10-07 17:43:19 +02:00
int vp8_fast_quantize_b_impl_sse2(short *coeff_ptr,
Initial WebM release
2010-05-18 17:58:33 +02:00
short *qcoeff_ptr, short *dequant_ptr,
vp8 fast quantizer sse2 optimizations for eob. Changed the end of block computation to use pmaxw. Removed additional pushing and popping of registers that was not needed. Change-Id: I08cb9b424513cd8a2c7ad8cea53b4e2adc66ef98
2010-12-10 00:00:30 +01:00
const short *inv_scan_order, short *round_ptr,
Initial WebM release
2010-05-18 17:58:33 +02:00
short *quant_ptr, short *dqcoeff_ptr);
Added vp8_fast_quantize_b_sse2 Moved vp8_fast_quantize_b_sse from quantize_mmx.asm into quantize_sse2.asm and renamed. Updated the assembly code to match the C version. Change-Id: I1766d9e1ca60e173f65badc0ca0c160c2b51b200
2010-10-07 17:43:19 +02:00
void vp8_fast_quantize_b_sse2(BLOCK *b, BLOCKD *d)
Initial WebM release
2010-05-18 17:58:33 +02:00
{
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *scan_mask = vp8_default_zig_zag_mask;//d->scan_order_mask_ptr;
short *coeff_ptr = b->coeff;
short *round_ptr = b->round;
Use the fast quantizer for inter mode selection Use the fast quantizer for inter mode selection and the regular quantizer for the rest of the encode for good quality, speed 1. Both performance and quality were improved. The quality gains will make up for the quality loss mentioned in I9dc089007ca08129fb6c11fe7692777ebb8647b0. Change-Id: Ia90bc9cf326a7c65d60d31fa32f6465ab6984d21
2010-12-28 20:51:46 +01:00
short *quant_ptr = b->quant_fast;
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *qcoeff_ptr = d->qcoeff;
Initial WebM release
2010-05-18 17:58:33 +02:00
short *dqcoeff_ptr = d->dqcoeff;
Convert [4][4] matrices to [16] arrays. Most of the code that actually uses these matrices indexes them as if they were a single contiguous array, and coverity produces reports about the resulting accesses that overflow the static bounds of the first row. This is perfectly legal in C, but converting them to actual [16] arrays should eliminate the report, and removes a good deal of extraneous indexing and address operators from the code. Change-Id: Ibda479e2232b3e51f9edf3b355b8640520fdbf23
2010-10-22 02:04:30 +02:00
short *dequant_ptr = d->dequant;
Initial WebM release
2010-05-18 17:58:33 +02:00
Fix compiler warning about vp8_fast_quantize_b_impl_ssse2. Typo had function defined as _ssse2 and prototyped as _sse2. Change-Id: If9f19da1a83cff40774a90cf936d601c0bf1b7fe
2010-10-14 02:08:13 +02:00
d->eob = vp8_fast_quantize_b_impl_sse2(
Initial WebM release
2010-05-18 17:58:33 +02:00
coeff_ptr,
qcoeff_ptr,
dequant_ptr,
vp8 fast quantizer sse2 optimizations for eob. Changed the end of block computation to use pmaxw. Removed additional pushing and popping of registers that was not needed. Change-Id: I08cb9b424513cd8a2c7ad8cea53b4e2adc66ef98
2010-12-10 00:00:30 +01:00
vp8_default_inv_zig_zag,
Initial WebM release
2010-05-18 17:58:33 +02:00
round_ptr,
quant_ptr,
dqcoeff_ptr
);
}
Added vp8_fast_quantize_b_sse2 Moved vp8_fast_quantize_b_sse from quantize_mmx.asm into quantize_sse2.asm and renamed. Updated the assembly code to match the C version. Change-Id: I1766d9e1ca60e173f65badc0ca0c160c2b51b200
2010-10-07 17:43:19 +02:00
sse2 version of vp8_regular_quantize_b Added sse2 version of vp8_regular_quantize_b which improved encode performance(for the clip used) by ~10% for 32 bit builds and ~3% for 64 bit builds. Also updated SHADOW_ARGS_TO_STACK to allow for more than 9 arguments. Change-Id: I62f78eabc8040b39f3ffdf21be175811e96b39af
2010-06-14 20:07:56 +02:00
int vp8_regular_quantize_b_impl_sse2(short *coeff_ptr, short *zbin_ptr,
update sse2 regular quantizer about ~5% gain on 32bit. disabled for 64bit unset executable bit on ssse3 version (cosmetic) Change-Id: I1a5860839eb294ce4261f819caea2dcfa78e57ca
2011-01-11 15:41:57 +01:00
short *qcoeff_ptr,short *dequant_ptr,
const int *default_zig_zag, short *round_ptr,
short *quant_ptr, short *dqcoeff_ptr,
unsigned short zbin_oq_value,
short *zbin_boost_ptr,
short *quant_shift_ptr);
sse2 version of vp8_regular_quantize_b Added sse2 version of vp8_regular_quantize_b which improved encode performance(for the clip used) by ~10% for 32 bit builds and ~3% for 64 bit builds. Also updated SHADOW_ARGS_TO_STACK to allow for more than 9 arguments. Change-Id: I62f78eabc8040b39f3ffdf21be175811e96b39af
2010-06-14 20:07:56 +02:00
void vp8_regular_quantize_b_sse2(BLOCK *b,BLOCKD *d)
{
update sse2 regular quantizer about ~5% gain on 32bit. disabled for 64bit unset executable bit on ssse3 version (cosmetic) Change-Id: I1a5860839eb294ce4261f819caea2dcfa78e57ca
2011-01-11 15:41:57 +01:00
d->eob = vp8_regular_quantize_b_impl_sse2(b->coeff,
b->zbin,
d->qcoeff,
d->dequant,
vp8_default_zig_zag1d,
b->round,
b->quant,
d->dqcoeff,
b->zbin_extra,
b->zrun_zbin_boost,
b->quant_shift);
sse2 version of vp8_regular_quantize_b Added sse2 version of vp8_regular_quantize_b which improved encode performance(for the clip used) by ~10% for 32 bit builds and ~3% for 64 bit builds. Also updated SHADOW_ARGS_TO_STACK to allow for more than 9 arguments. Change-Id: I62f78eabc8040b39f3ffdf21be175811e96b39af
2010-06-14 20:07:56 +02:00
}
Initial WebM release
2010-05-18 17:58:33 +02:00
int vp8_mbblock_error_xmm_impl(short *coeff_ptr, short *dcoef_ptr, int dc);
int vp8_mbblock_error_xmm(MACROBLOCK *mb, int dc)
{
short *coeff_ptr = mb->block[0].coeff;
short *dcoef_ptr = mb->e_mbd.block[0].dqcoeff;
return vp8_mbblock_error_xmm_impl(coeff_ptr, dcoef_ptr, dc);
}
int vp8_mbuverror_xmm_impl(short *s_ptr, short *d_ptr);
int vp8_mbuverror_xmm(MACROBLOCK *mb)
{
short *s_ptr = &mb->coeff[256];
short *d_ptr = &mb->e_mbd.dqcoeff[256];
return vp8_mbuverror_xmm_impl(s_ptr, d_ptr);
}
Add SSE2 subtract functions Instead of doing 8-bit data unpack and 16-bit subtraction, use psubb to do 16 8-bit subtractions and pcmpgtb to preserve the sign information. This does not bring noticable gain since these functions are not called frequently. Change-Id: I90a0dfaa3db9d422e4ada324076596ffb178548e
2010-10-18 20:15:15 +02:00
void vp8_subtract_b_sse2_impl(unsigned char *z, int src_stride,
short *diff, unsigned char *predictor,
int pitch);
void vp8_subtract_b_sse2(BLOCK *be, BLOCKD *bd, int pitch)
{
unsigned char *z = *(be->base_src) + be->src;
unsigned int src_stride = be->src_stride;
short *diff = &be->src_diff[0];
unsigned char *predictor = &bd->predictor[0];
vp8_subtract_b_sse2_impl(z, src_stride, diff, predictor, pitch);
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
SSSE3 version of fast quantizer (test clip: tulip) For good quality mode with speed=1, this gave the encoder a small (2 - 3%) performance boost. Change-Id: I8a1d4269465944ac0819986c2f0be4b0a2ee0b35
2010-11-01 21:24:15 +01:00
#if HAVE_SSSE3
int vp8_fast_quantize_b_impl_ssse3(short *coeff_ptr,
short *qcoeff_ptr, short *dequant_ptr,
short *round_ptr,
short *quant_ptr, short *dqcoeff_ptr);
void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d)
{
d->eob = vp8_fast_quantize_b_impl_ssse3(
b->coeff,
d->qcoeff,
d->dequant,
b->round,
Use the fast quantizer for inter mode selection Use the fast quantizer for inter mode selection and the regular quantizer for the rest of the encode for good quality, speed 1. Both performance and quality were improved. The quality gains will make up for the quality loss mentioned in I9dc089007ca08129fb6c11fe7692777ebb8647b0. Change-Id: Ia90bc9cf326a7c65d60d31fa32f6465ab6984d21
2010-12-28 20:51:46 +01:00
b->quant_fast,
SSSE3 version of fast quantizer (test clip: tulip) For good quality mode with speed=1, this gave the encoder a small (2 - 3%) performance boost. Change-Id: I8a1d4269465944ac0819986c2f0be4b0a2ee0b35
2010-11-01 21:24:15 +01:00
d->dqcoeff
);
}
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
{
#if CONFIG_RUNTIME_CPU_DETECT
int flags = x86_simd_caps();
int mmx_enabled = flags & HAS_MMX;
int xmm_enabled = flags & HAS_SSE;
int wmt_enabled = flags & HAS_SSE2;
int SSE3Enabled = flags & HAS_SSE3;
int SSSE3Enabled = flags & HAS_SSSE3;
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
int SSE4_1Enabled = flags & HAS_SSE4_1;
Initial WebM release
2010-05-18 17:58:33 +02:00
/* Note:
*
* This platform can be built without runtime CPU detection as well. If
* you modify any of the function mappings present in this file, be sure
* to also update them in static mapings (<arch>/filename_<arch>.h)
*/
/* Override default functions with fastest ones for this CPU. */
#if HAVE_MMX
if (mmx_enabled)
{
cpi->rtcd.variance.sad16x16 = vp8_sad16x16_mmx;
cpi->rtcd.variance.sad16x8 = vp8_sad16x8_mmx;
cpi->rtcd.variance.sad8x16 = vp8_sad8x16_mmx;
cpi->rtcd.variance.sad8x8 = vp8_sad8x8_mmx;
cpi->rtcd.variance.sad4x4 = vp8_sad4x4_mmx;
cpi->rtcd.variance.var4x4 = vp8_variance4x4_mmx;
cpi->rtcd.variance.var8x8 = vp8_variance8x8_mmx;
cpi->rtcd.variance.var8x16 = vp8_variance8x16_mmx;
cpi->rtcd.variance.var16x8 = vp8_variance16x8_mmx;
cpi->rtcd.variance.var16x16 = vp8_variance16x16_mmx;
cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_mmx;
cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_mmx;
cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_mmx;
cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_mmx;
cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_mmx;
Fix half-pixel variance RTCD functions This patch fixes the system dependent entries for the half-pixel variance functions in both the RTCD and non-RTCD cases: - The generic C versions of these functions are now correct. Before all three cases called the hv code. - Wire up the ARM functions in RTCD mode - Created stubs for x86 to call the optimized subpixel functions with the correct parameters, rather than falling back to C code. Change-Id: I1d937d074d929e0eb93aacb1232cc5e0ad1c6184
2010-10-27 17:28:43 +02:00
cpi->rtcd.variance.halfpixvar16x16_h = vp8_variance_halfpixvar16x16_h_mmx;
cpi->rtcd.variance.halfpixvar16x16_v = vp8_variance_halfpixvar16x16_v_mmx;
cpi->rtcd.variance.halfpixvar16x16_hv = vp8_variance_halfpixvar16x16_hv_mmx;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->rtcd.variance.subpixmse16x16 = vp8_sub_pixel_mse16x16_mmx;
cpi->rtcd.variance.mse16x16 = vp8_mse16x16_mmx;
cpi->rtcd.variance.getmbss = vp8_get_mb_ss_mmx;
cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_mmx;
cpi->rtcd.variance.get8x8var = vp8_get8x8var_mmx;
cpi->rtcd.variance.get16x16var = vp8_get16x16var_mmx;
cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_mmx;
FDCT optimizations. Fixed up the fdct for mmx and 8x4 sse2 to match them most recent changes. Change-Id: Ibee2d6c536fe14dcf75cd6eb1c73f4848a56d719
2010-10-21 19:53:15 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_mmx;
cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_mmx;
Redo the forward 4x4 dct The new fdct lowers the round trip sum squared error for a 4x4 block ~0.12. or ~0.008/pixel. For reference, the old matrix multiply version has average round trip error 1.46 for a 4x4 block. Thanks to "derf" for his suggestions and references. Change-Id: I5559d1e81d333b319404ab16b336b739f87afc79
2010-06-16 21:52:18 +02:00
cpi->rtcd.fdct.fast4x4 = vp8_short_fdct4x4_mmx;
cpi->rtcd.fdct.fast8x4 = vp8_short_fdct8x4_mmx;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_c;
cpi->rtcd.encodemb.berr = vp8_block_error_mmx;
cpi->rtcd.encodemb.mberr = vp8_mbblock_error_mmx;
cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_mmx;
cpi->rtcd.encodemb.subb = vp8_subtract_b_mmx;
cpi->rtcd.encodemb.submby = vp8_subtract_mby_mmx;
cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_mmx;
Make the quantizer exact. This replaces the approximate division-by-multiplication in the quantizer with an exact one that costs just one add and one shift extra. The asm versions have not been updated in this patch, and thus have been disabled, since the new method requires different multipliers which are not compatible with the old method. Change-Id: I53ac887af0f969d906e464c88b1f4be69c6b1206
2010-06-29 02:15:09 +02:00
/*cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_mmx;*/
Initial WebM release
2010-05-18 17:58:33 +02:00
}
#endif
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
#if HAVE_SSE2
Initial WebM release
2010-05-18 17:58:33 +02:00
if (wmt_enabled)
{
cpi->rtcd.variance.sad16x16 = vp8_sad16x16_wmt;
cpi->rtcd.variance.sad16x8 = vp8_sad16x8_wmt;
cpi->rtcd.variance.sad8x16 = vp8_sad8x16_wmt;
cpi->rtcd.variance.sad8x8 = vp8_sad8x8_wmt;
cpi->rtcd.variance.sad4x4 = vp8_sad4x4_wmt;
cpi->rtcd.variance.var4x4 = vp8_variance4x4_wmt;
cpi->rtcd.variance.var8x8 = vp8_variance8x8_wmt;
cpi->rtcd.variance.var8x16 = vp8_variance8x16_wmt;
cpi->rtcd.variance.var16x8 = vp8_variance16x8_wmt;
cpi->rtcd.variance.var16x16 = vp8_variance16x16_wmt;
cpi->rtcd.variance.subpixvar4x4 = vp8_sub_pixel_variance4x4_wmt;
cpi->rtcd.variance.subpixvar8x8 = vp8_sub_pixel_variance8x8_wmt;
cpi->rtcd.variance.subpixvar8x16 = vp8_sub_pixel_variance8x16_wmt;
cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_wmt;
cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_wmt;
Fix half-pixel variance RTCD functions This patch fixes the system dependent entries for the half-pixel variance functions in both the RTCD and non-RTCD cases: - The generic C versions of these functions are now correct. Before all three cases called the hv code. - Wire up the ARM functions in RTCD mode - Created stubs for x86 to call the optimized subpixel functions with the correct parameters, rather than falling back to C code. Change-Id: I1d937d074d929e0eb93aacb1232cc5e0ad1c6184
2010-10-27 17:28:43 +02:00
cpi->rtcd.variance.halfpixvar16x16_h = vp8_variance_halfpixvar16x16_h_wmt;
cpi->rtcd.variance.halfpixvar16x16_v = vp8_variance_halfpixvar16x16_v_wmt;
cpi->rtcd.variance.halfpixvar16x16_hv = vp8_variance_halfpixvar16x16_hv_wmt;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->rtcd.variance.subpixmse16x16 = vp8_sub_pixel_mse16x16_wmt;
cpi->rtcd.variance.mse16x16 = vp8_mse16x16_wmt;
cpi->rtcd.variance.getmbss = vp8_get_mb_ss_sse2;
cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_sse2;
cpi->rtcd.variance.get8x8var = vp8_get8x8var_sse2;
cpi->rtcd.variance.get16x16var = vp8_get16x16var_sse2;
/* cpi->rtcd.variance.get4x4sse_cs not implemented for wmt */;
Added first-pass sse2 version of Yaowu's new fdct. Change-Id: Ib479210067510162879c368428b92690591120b2
2010-06-24 19:11:30 +02:00
cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_sse2;
cpi->rtcd.fdct.short8x4 = vp8_short_fdct8x4_sse2;
cpi->rtcd.fdct.fast4x4 = vp8_short_fdct4x4_sse2;
cpi->rtcd.fdct.fast8x4 = vp8_short_fdct8x4_sse2;
Rewrite vp8_short_walsh4x4_sse2() This rewriting reflects changes made in commit "Improve the accuracy of forward walsh-hadamard transform". Since this function is not called much, only a small encoder performance gain (~0.5% ) is seen. Change-Id: Ie9df58a43028a11fd5b115c4bbe3141f7596578b
2010-10-21 16:26:50 +02:00
cpi->rtcd.fdct.walsh_short4x4 = vp8_short_walsh4x4_sse2 ;
Initial WebM release
2010-05-18 17:58:33 +02:00
cpi->rtcd.encodemb.berr = vp8_block_error_xmm;
cpi->rtcd.encodemb.mberr = vp8_mbblock_error_xmm;
cpi->rtcd.encodemb.mbuverr = vp8_mbuverror_xmm;
Add SSE2 subtract functions Instead of doing 8-bit data unpack and 16-bit subtraction, use psubb to do 16 8-bit subtractions and pcmpgtb to preserve the sign information. This does not bring noticable gain since these functions are not called frequently. Change-Id: I90a0dfaa3db9d422e4ada324076596ffb178548e
2010-10-18 20:15:15 +02:00
cpi->rtcd.encodemb.subb = vp8_subtract_b_sse2;
cpi->rtcd.encodemb.submby = vp8_subtract_mby_sse2;
cpi->rtcd.encodemb.submbuv = vp8_subtract_mbuv_sse2;
Initial WebM release
2010-05-18 17:58:33 +02:00
update sse2 regular quantizer about ~5% gain on 32bit. disabled for 64bit unset executable bit on ssse3 version (cosmetic) Change-Id: I1a5860839eb294ce4261f819caea2dcfa78e57ca
2011-01-11 15:41:57 +01:00
#if ARCH_X86
cpi->rtcd.quantize.quantb = vp8_regular_quantize_b_sse2;
#endif
Added vp8_fast_quantize_b_sse2 Moved vp8_fast_quantize_b_sse from quantize_mmx.asm into quantize_sse2.asm and renamed. Updated the assembly code to match the C version. Change-Id: I1766d9e1ca60e173f65badc0ca0c160c2b51b200
2010-10-07 17:43:19 +02:00
cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_sse2;
x86 sse2 temporal_filter_apply count can be reduced to short because the max number of filtered frames is set to 15. the max value for any frame is 32 (modifier = 16, filter_weight = 2). 15*32 = 480 which requires 9 bits this function goes from about 7000 us / 1000 iterations for the C code to < 275 us / 1000 iterations for sse2 for block_size = 16 and from about 1800 us / 1000 iters to < 100 us / 1000 iters for block_size = 8 Change-Id: I64a32607f58a2d33c39286f468b04ccd457d9e6e
2010-12-22 17:23:51 +01:00
cpi->rtcd.temporal.apply = vp8_temporal_filter_apply_sse2;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
#endif
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
#if HAVE_SSE3
Initial WebM release
2010-05-18 17:58:33 +02:00
if (SSE3Enabled)
{
cpi->rtcd.variance.sad16x16 = vp8_sad16x16_sse3;
cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_sse3;
cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_sse3;
cpi->rtcd.variance.sad8x16x3 = vp8_sad8x16x3_sse3;
cpi->rtcd.variance.sad8x8x3 = vp8_sad8x8x3_sse3;
cpi->rtcd.variance.sad4x4x3 = vp8_sad4x4x3_sse3;
cpi->rtcd.search.full_search = vp8_full_search_sadx3;
cpi->rtcd.variance.sad16x16x4d = vp8_sad16x16x4d_sse3;
cpi->rtcd.variance.sad16x8x4d = vp8_sad16x8x4d_sse3;
cpi->rtcd.variance.sad8x16x4d = vp8_sad8x16x4d_sse3;
cpi->rtcd.variance.sad8x8x4d = vp8_sad8x8x4d_sse3;
cpi->rtcd.variance.sad4x4x4d = vp8_sad4x4x4d_sse3;
cpi->rtcd.search.diamond_search = vp8_diamond_search_sadx4;
}
#endif
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
#if HAVE_SSSE3
Initial WebM release
2010-05-18 17:58:33 +02:00
if (SSSE3Enabled)
{
cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_ssse3;
cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_ssse3;
SSSE3 version of fast quantizer (test clip: tulip) For good quality mode with speed=1, this gave the encoder a small (2 - 3%) performance boost. Change-Id: I8a1d4269465944ac0819986c2f0be4b0a2ee0b35
2010-11-01 21:24:15 +01:00
cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3;
Initial WebM release
2010-05-18 17:58:33 +02:00
}
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
#endif
Initial WebM release
2010-05-18 17:58:33 +02:00
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
#if HAVE_SSE4_1
if (SSE4_1Enabled)
{
cpi->rtcd.variance.sad16x16x8 = vp8_sad16x16x8_sse4;
cpi->rtcd.variance.sad16x8x8 = vp8_sad16x8x8_sse4;
cpi->rtcd.variance.sad8x16x8 = vp8_sad8x16x8_sse4;
cpi->rtcd.variance.sad8x8x8 = vp8_sad8x8x8_sse4;
cpi->rtcd.variance.sad4x4x8 = vp8_sad4x4x8_sse4;
cpi->rtcd.search.full_search = vp8_full_search_sadx8;
}
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
Initial WebM release
2010-05-18 17:58:33 +02:00
#endif
}
Reference in New Issue Copy Permalink