vpx/vp8/common/x86/vp8_asm_stubs.c

593 lines
17 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include "vpx_ports/config.h"
#include "vpx_ports/mem.h"
#include "vp8/common/subpixel.h"
2010-05-18 17:58:33 +02:00
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#if CONFIG_SIXTEENTH_SUBPEL_UV
extern const short vp8_six_tap_mmx[16][6*8];
extern const short vp8_bilinear_filters_mmx[16][2*8];
#else
2010-05-18 17:58:33 +02:00
extern const short vp8_six_tap_mmx[8][6*8];
extern const short vp8_bilinear_filters_mmx[8][2*8];
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#endif
//#define ANNOUNCE_FUNCTION
2010-05-18 17:58:33 +02:00
extern void vp8_filter_block1d_h6_mmx
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
extern void vp8_filter_block1dc_v6_mmx
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int output_pitch,
unsigned int pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
extern void vp8_filter_block1d8_h6_sse2
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
extern void vp8_filter_block1d16_h6_sse2
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
extern void vp8_filter_block1d8_v6_sse2
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int dst_ptich,
unsigned int pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
extern void vp8_filter_block1d16_v6_sse2
(
unsigned short *src_ptr,
unsigned char *output_ptr,
int dst_ptich,
unsigned int pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const short *vp8_filter
);
2010-05-18 17:58:33 +02:00
extern void vp8_unpack_block1d16_h6_sse2
(
unsigned char *src_ptr,
unsigned short *output_ptr,
unsigned int src_pixels_per_line,
unsigned int output_height,
unsigned int output_width
);
extern void vp8_filter_block1d8_h6_only_sse2
2010-05-18 17:58:33 +02:00
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
unsigned char *output_ptr,
int dst_ptich,
2010-05-18 17:58:33 +02:00
unsigned int output_height,
const short *vp8_filter
2010-05-18 17:58:33 +02:00
);
extern void vp8_filter_block1d16_h6_only_sse2
2010-05-18 17:58:33 +02:00
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
unsigned char *output_ptr,
2010-05-18 17:58:33 +02:00
int dst_ptich,
unsigned int output_height,
const short *vp8_filter
2010-05-18 17:58:33 +02:00
);
extern void vp8_filter_block1d8_v6_only_sse2
2010-05-18 17:58:33 +02:00
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
2010-05-18 17:58:33 +02:00
unsigned char *output_ptr,
int dst_ptich,
unsigned int output_height,
const short *vp8_filter
2010-05-18 17:58:33 +02:00
);
extern prototype_subpixel_predict(vp8_bilinear_predict8x8_mmx);
#if HAVE_MMX
void vp8_sixtap_predict4x4_mmx
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict4x4_mmx\n");
#endif
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 16*16); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 8, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1dc_v6_mmx(FData2 + 8, dst_ptr, dst_pitch, 8, 4 , 4, 4, VFilter);
}
void vp8_sixtap_predict16x16_mmx
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict16x16_mmx\n");
#endif
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 24*24); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 21, 32, HFilter);
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 21, 32, HFilter);
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 8, FData2 + 8, src_pixels_per_line, 1, 21, 32, HFilter);
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 12, FData2 + 12, src_pixels_per_line, 1, 21, 32, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1dc_v6_mmx(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, 16, VFilter);
vp8_filter_block1dc_v6_mmx(FData2 + 36, dst_ptr + 4, dst_pitch, 32, 16 , 16, 16, VFilter);
vp8_filter_block1dc_v6_mmx(FData2 + 40, dst_ptr + 8, dst_pitch, 32, 16 , 16, 16, VFilter);
vp8_filter_block1dc_v6_mmx(FData2 + 44, dst_ptr + 12, dst_pitch, 32, 16 , 16, 16, VFilter);
}
void vp8_sixtap_predict8x8_mmx
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x8_mmx\n");
#endif
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 256); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 13, 16, HFilter);
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 13, 16, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1dc_v6_mmx(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 8, 8, VFilter);
vp8_filter_block1dc_v6_mmx(FData2 + 20, dst_ptr + 4, dst_pitch, 16, 8 , 8, 8, VFilter);
}
void vp8_sixtap_predict8x4_mmx
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x4_mmx\n");
#endif
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 256); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 16, HFilter);
vp8_filter_block1d_h6_mmx(src_ptr - (2 * src_pixels_per_line) + 4, FData2 + 4, src_pixels_per_line, 1, 9, 16, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1dc_v6_mmx(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 4, 8, VFilter);
vp8_filter_block1dc_v6_mmx(FData2 + 20, dst_ptr + 4, dst_pitch, 16, 8 , 4, 8, VFilter);
}
void vp8_bilinear_predict16x16_mmx
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
vp8_bilinear_predict8x8_mmx(src_ptr, src_pixels_per_line, xoffset, yoffset, dst_ptr, dst_pitch);
vp8_bilinear_predict8x8_mmx(src_ptr + 8, src_pixels_per_line, xoffset, yoffset, dst_ptr + 8, dst_pitch);
vp8_bilinear_predict8x8_mmx(src_ptr + 8 * src_pixels_per_line, src_pixels_per_line, xoffset, yoffset, dst_ptr + dst_pitch * 8, dst_pitch);
vp8_bilinear_predict8x8_mmx(src_ptr + 8 * src_pixels_per_line + 8, src_pixels_per_line, xoffset, yoffset, dst_ptr + dst_pitch * 8 + 8, dst_pitch);
}
#endif
#if HAVE_SSE2
void vp8_sixtap_predict16x16_sse2
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 24*24); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict16x16_sse2\n");
#endif
2010-05-18 17:58:33 +02:00
if (xoffset)
{
if (yoffset)
{
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d16_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 21, 32, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1d16_v6_sse2(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, dst_pitch, VFilter);
}
else
{
/* First-pass only */
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d16_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 16, HFilter);
}
2010-05-18 17:58:33 +02:00
}
else
{
/* Second-pass only */
2010-05-18 17:58:33 +02:00
VFilter = vp8_six_tap_mmx[yoffset];
vp8_unpack_block1d16_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 21, 32);
vp8_filter_block1d16_v6_sse2(FData2 + 32, dst_ptr, dst_pitch, 32, 16 , 16, dst_pitch, VFilter);
2010-05-18 17:58:33 +02:00
}
}
void vp8_sixtap_predict8x8_sse2
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 256); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x8_sse2\n");
#endif
2010-05-18 17:58:33 +02:00
if (xoffset)
{
if (yoffset)
{
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d8_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 13, 16, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1d8_v6_sse2(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 8, dst_pitch, VFilter);
}
else
{
/* First-pass only */
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d8_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 8, HFilter);
}
2010-05-18 17:58:33 +02:00
}
else
{
/* Second-pass only */
2010-05-18 17:58:33 +02:00
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1d8_v6_only_sse2(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 8, VFilter);
2010-05-18 17:58:33 +02:00
}
}
void vp8_sixtap_predict8x4_sse2
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned short, FData2, 256); /* Temp data bufffer used in filtering */
2010-05-18 17:58:33 +02:00
const short *HFilter, *VFilter;
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x4_sse2\n");
#endif
2010-05-18 17:58:33 +02:00
if (xoffset)
{
if (yoffset)
{
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d8_h6_sse2(src_ptr - (2 * src_pixels_per_line), FData2, src_pixels_per_line, 1, 9, 16, HFilter);
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1d8_v6_sse2(FData2 + 16, dst_ptr, dst_pitch, 16, 8 , 4, dst_pitch, VFilter);
}
else
{
/* First-pass only */
HFilter = vp8_six_tap_mmx[xoffset];
vp8_filter_block1d8_h6_only_sse2(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 4, HFilter);
}
2010-05-18 17:58:33 +02:00
}
else
{
/* Second-pass only */
2010-05-18 17:58:33 +02:00
VFilter = vp8_six_tap_mmx[yoffset];
vp8_filter_block1d8_v6_only_sse2(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 4, VFilter);
2010-05-18 17:58:33 +02:00
}
}
2010-05-18 17:58:33 +02:00
#endif
#if HAVE_SSSE3
extern void vp8_filter_block1d8_h6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
unsigned char *output_ptr,
unsigned int output_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
extern void vp8_filter_block1d16_h6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
unsigned char *output_ptr,
unsigned int output_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
extern void vp8_filter_block1d16_v6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
extern void vp8_filter_block1d8_v6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
extern void vp8_filter_block1d4_h6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pixels_per_line,
unsigned char *output_ptr,
unsigned int output_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
extern void vp8_filter_block1d4_v6_ssse3
(
unsigned char *src_ptr,
unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
unsigned int vp8_filter_index
);
void vp8_sixtap_predict16x16_ssse3
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned char, FData2, 24*24);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict16x16_ssse3\n");
#endif
if (xoffset)
{
if (yoffset)
{
vp8_filter_block1d16_h6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, FData2, 16, 21, xoffset);
vp8_filter_block1d16_v6_ssse3(FData2 , 16, dst_ptr, dst_pitch, 16, yoffset);
}
else
{
/* First-pass only */
vp8_filter_block1d16_h6_ssse3(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 16, xoffset);
}
}
else
{
/* Second-pass only */
vp8_filter_block1d16_v6_ssse3(src_ptr - (2 * src_pixels_per_line) , src_pixels_per_line, dst_ptr, dst_pitch, 16, yoffset);
}
}
void vp8_sixtap_predict8x8_ssse3
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned char, FData2, 256);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x8_ssse3\n");
#endif
if (xoffset)
{
if (yoffset)
{
vp8_filter_block1d8_h6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, FData2, 8, 13, xoffset);
vp8_filter_block1d8_v6_ssse3(FData2, 8, dst_ptr, dst_pitch, 8, yoffset);
}
else
{
vp8_filter_block1d8_h6_ssse3(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 8, xoffset);
}
}
else
{
/* Second-pass only */
vp8_filter_block1d8_v6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 8, yoffset);
}
}
void vp8_sixtap_predict8x4_ssse3
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned char, FData2, 256);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict8x4_ssse3\n");
#endif
if (xoffset)
{
if (yoffset)
{
vp8_filter_block1d8_h6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, FData2, 8, 9, xoffset);
vp8_filter_block1d8_v6_ssse3(FData2, 8, dst_ptr, dst_pitch, 4, yoffset);
}
else
{
/* First-pass only */
vp8_filter_block1d8_h6_ssse3(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 4, xoffset);
}
}
else
{
/* Second-pass only */
vp8_filter_block1d8_v6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 4, yoffset);
}
}
void vp8_sixtap_predict4x4_ssse3
(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch
)
{
DECLARE_ALIGNED_ARRAY(16, unsigned char, FData2, 4*9);
Supporting high precision 1/8-pel motion vectors This is the initial patch for supporting 1/8th pel motion. Currently if we configure with enable-high-precision-mv, all motion vectors would default to 1/8 pel. Encode and decode syncs fine with the current code. In the next phase the code will be refactored so that we can choose the 1/8 pel mode adaptively at a frame/segment/mb level. Derf results: http://www.corp.google.com/~debargha/vp8_results/enhinterp_hpmv.html (about 0.83% better than 8-tap interpoaltion) Patch 3: Rebased. Also adding 1/16th pel interpolation for U and V Patch 4: HD results. http://www.corp.google.com/~debargha/vp8_results/enhinterp_hd_hpmv.html Seems impressive (unless I am doing something wrong). Patch 5: Added mmx/sse for bilateral filtering, as well as enforced use of c-versions of subpel filters with 8-taps and 1/16th pel; Also redesigned the 8-tap filters to reduce the cut-off in order to introduce a denoising effect. There is a new configure option sixteenth-subpel-uv which will use 1/16 th pel interpolation for uv, if the motion vectors have 1/8 pel accuracy. With the fixes the results are promising on the derf set. The enhanced interpolation option with 8-taps alone gives 3% improvement over thei derf set: http://www.corp.google.com/~debargha/vp8_results/enhinterpn.html Results on high precision mv and on the hd set are to follow. Patch 6: Adding a missing condition for CONFIG_SIXTEENTH_SUBPEL_UV in vp8/common/x86/x86_systemdependent.c Patch 7: Cleaning up various debug messages. Patch 8: Merge conflict Change-Id: I5b1d844457aefd7414a9e4e0e06c6ed38fd8cc04
2012-02-16 18:29:54 +01:00
#ifdef ANNOUNCE_FUNCTION
printf("vp8_sixtap_predict4x4_ssse3\n");
#endif
if (xoffset)
{
if (yoffset)
{
vp8_filter_block1d4_h6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, FData2, 4, 9, xoffset);
vp8_filter_block1d4_v6_ssse3(FData2, 4, dst_ptr, dst_pitch, 4, yoffset);
}
else
{
vp8_filter_block1d4_h6_ssse3(src_ptr, src_pixels_per_line, dst_ptr, dst_pitch, 4, xoffset);
}
}
else
{
vp8_filter_block1d4_v6_ssse3(src_ptr - (2 * src_pixels_per_line), src_pixels_per_line, dst_ptr, dst_pitch, 4, yoffset);
}
}
#endif