Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge commit '6d5636ad9ab6bd9bedf902051d88b7044385f88b'

Browse Source 
* commit '6d5636ad9ab6bd9bedf902051d88b7044385f88b':
  hevc: x86: Add add_residual() SIMD optimizations

See a6af4bf64d

This merge is only cosmetics (renames, space shuffling, etc).

The functionnal changes in the ASM are *not* merged:
- unrolling with %rep is kept
- ADD_RES_MMX_4_8 is left untouched: this needs investigation

Merged-by: Clément Bœsch <u@pkh.me>
pull/62/merge
Clément Bœsch 8 years ago
parent 40ac226014 6d5636ad9a
commit 3d65359832
5 changed files with 163 additions and 168 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 1
      doc/libav-merge.txt
  2. 2
      libavcodec/hevcdsp.h
  3. 125
      libavcodec/x86/hevc_add_res.asm
  4. 29
      libavcodec/x86/hevcdsp.h
  5. 30
      libavcodec/x86/hevcdsp_init.c

1
doc/libav-merge.txt
Unescape View File

@ -97,6 +97,7 @@ Stuff that didn't reach the codebase:
- VAAPI VP8 decode hwaccel (currently under review: http://ffmpeg.org/pipermail/ffmpeg-devel/2017-February/thread.html#207348)
- Removal of the custom atomic API (5cc0057f49, see http://ffmpeg.org/pipermail/ffmpeg-devel/2017-March/209003.html)
- Use the new bitstream filter for extracting extradata (see 8e2ea69135 and 096a8effa3)
- ADD_RES_MMX_4_8 in libavcodec/x86/hevc_add_res.asm probably needs updating (see 589880710)
Collateral damage that needs work locally:
------------------------------------------

2
libavcodec/hevcdsp.h
Unescape View File

@ -46,7 +46,7 @@ typedef struct HEVCDSPContext {
void (*put_pcm)(uint8_t *_dst, ptrdiff_t _stride, int width, int height,
struct GetBitContext *gb, int pcm_bit_depth);
void (*add_residual[4])(uint8_t *_dst, int16_t *coeffs, ptrdiff_t _stride);
void (*add_residual[4])(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void (*dequant)(int16_t *coeffs, int16_t log2_size);

125
libavcodec/x86/hevc_add_res.asm
Unescape View File

@ -1,4 +1,4 @@
; /*
; *****************************************************************************
; * Provide SIMD optimizations for add_residual functions for HEVC decoding
; * Copyright (c) 2014 Pierre-Edouard LEPERE
; *
@ -17,7 +17,8 @@
; * You should have received a copy of the GNU Lesser General Public
; * License along with FFmpeg; if not, write to the Free Software
; * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
; */
; ******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION .text
@ -25,9 +26,8 @@ SECTION .text
cextern pw_1023
%define max_pixels_10 pw_1023
;the tr_add macros and functions were largely inspired by x264 project's code in the h264_idct.asm file
%macro TR_ADD_MMX_4_8 0
; the add_res macros and functions were largely inspired by h264_idct.asm from the x264 project
%macro ADD_RES_MMX_4_8 0
mova m2, [r1]
mova m4, [r1+8]
pxor m3, m3
@ -51,15 +51,15 @@ cextern pw_1023
INIT_MMX mmxext
; void ff_hevc_tranform_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_add_residual4_8, 3, 4, 6
TR_ADD_MMX_4_8
; void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride)
cglobal hevc_add_residual_4_8, 3, 4, 6
ADD_RES_MMX_4_8
add r1, 16
lea r0, [r0+r2*2]
TR_ADD_MMX_4_8
ADD_RES_MMX_4_8
RET
%macro TR_ADD_SSE_8_8 0
%macro ADD_RES_SSE_8_8 0
pxor m3, m3
mova m4, [r1]
mova m6, [r1+16]
@ -88,7 +88,7 @@ cglobal hevc_add_residual4_8, 3, 4, 6
movhps [r0+r3], m1
%endmacro
%macro TR_ADD_SSE_16_32_8 3
%macro ADD_RES_SSE_16_32_8 3
mova xm2, [r1+%1]
mova xm6, [r1+%1+16]
%if cpuflag(avx2)
@ -135,39 +135,39 @@ cglobal hevc_add_residual4_8, 3, 4, 6
%macro TRANSFORM_ADD_8 0
; void ff_hevc_add_residual8_8_<opt>(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_add_residual8_8, 3, 4, 8
; void ff_hevc_add_residual_8_8_<opt>(uint8_t *dst, int16_t *res, ptrdiff_t stride)
cglobal hevc_add_residual_8_8, 3, 4, 8
lea r3, [r2*3]
TR_ADD_SSE_8_8
ADD_RES_SSE_8_8
add r1, 64
lea r0, [r0+r2*4]
TR_ADD_SSE_8_8
ADD_RES_SSE_8_8
RET
; void ff_hevc_add_residual16_8_<opt>(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_add_residual16_8, 3, 4, 7
; void ff_hevc_add_residual_16_8_<opt>(uint8_t *dst, int16_t *res, ptrdiff_t stride)
cglobal hevc_add_residual_16_8, 3, 4, 7
pxor m0, m0
lea r3, [r2*3]
TR_ADD_SSE_16_32_8 0, r0, r0+r2
TR_ADD_SSE_16_32_8 64, r0+r2*2, r0+r3
ADD_RES_SSE_16_32_8 0, r0, r0+r2
ADD_RES_SSE_16_32_8 64, r0+r2*2, r0+r3
%rep 3
add r1, 128
lea r0, [r0+r2*4]
TR_ADD_SSE_16_32_8 0, r0, r0+r2
TR_ADD_SSE_16_32_8 64, r0+r2*2, r0+r3
ADD_RES_SSE_16_32_8 0, r0, r0+r2
ADD_RES_SSE_16_32_8 64, r0+r2*2, r0+r3
%endrep
RET
; void ff_hevc_add_residual32_8_<opt>(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_add_residual32_8, 3, 4, 7
; void ff_hevc_add_residual_32_8_<opt>(uint8_t *dst, int16_t *res, ptrdiff_t stride)
cglobal hevc_add_residual_32_8, 3, 4, 7
pxor m0, m0
TR_ADD_SSE_16_32_8 0, r0, r0+16
TR_ADD_SSE_16_32_8 64, r0+r2, r0+r2+16
ADD_RES_SSE_16_32_8 0, r0, r0+16
ADD_RES_SSE_16_32_8 64, r0+r2, r0+r2+16
%rep 15
add r1, 128
lea r0, [r0+r2*2]
TR_ADD_SSE_16_32_8 0, r0, r0+16
TR_ADD_SSE_16_32_8 64, r0+r2, r0+r2+16
ADD_RES_SSE_16_32_8 0, r0, r0+16
ADD_RES_SSE_16_32_8 64, r0+r2, r0+r2+16
%endrep
RET
%endmacro
@ -179,25 +179,22 @@ TRANSFORM_ADD_8
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
; void ff_hevc_add_residual32_8_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_add_residual32_8, 3, 4, 7
; void ff_hevc_add_residual_32_8_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride)
cglobal hevc_add_residual_32_8, 3, 4, 7
pxor m0, m0
lea r3, [r2*3]
TR_ADD_SSE_16_32_8 0, r0, r0+r2
TR_ADD_SSE_16_32_8 128, r0+r2*2, r0+r3
ADD_RES_SSE_16_32_8 0, r0, r0+r2
ADD_RES_SSE_16_32_8 128, r0+r2*2, r0+r3
%rep 7
add r1, 256
lea r0, [r0+r2*4]
TR_ADD_SSE_16_32_8 0, r0, r0+r2
TR_ADD_SSE_16_32_8 128, r0+r2*2, r0+r3
ADD_RES_SSE_16_32_8 0, r0, r0+r2
ADD_RES_SSE_16_32_8 128, r0+r2*2, r0+r3
%endrep
RET
%endif
;-----------------------------------------------------------------------------
; void ff_hevc_add_residual_10(pixel *dst, int16_t *block, int stride)
;-----------------------------------------------------------------------------
%macro TR_ADD_SSE_8_10 4
%macro ADD_RES_SSE_8_10 4
mova m0, [%4]
mova m1, [%4+16]
mova m2, [%4+32]
@ -216,7 +213,7 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+%3], m3
%endmacro
%macro TR_ADD_MMX4_10 3
%macro ADD_RES_MMX_4_10 3
mova m0, [%1+0]
mova m1, [%1+%2]
paddw m0, [%3]
@ -227,7 +224,7 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+%2], m1
%endmacro
%macro TRANS_ADD_SSE_16_10 3
%macro ADD_RES_SSE_16_10 3
mova m0, [%3]
mova m1, [%3+16]
mova m2, [%3+32]
@ -246,7 +243,7 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+%2+16], m3
%endmacro
%macro TRANS_ADD_SSE_32_10 2
%macro ADD_RES_SSE_32_10 2
mova m0, [%2]
mova m1, [%2+16]
mova m2, [%2+32]
@ -266,7 +263,7 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+48], m3
%endmacro
%macro TRANS_ADD16_AVX2 4
%macro ADD_RES_AVX2_16_10 4
mova m0, [%4]
mova m1, [%4+32]
mova m2, [%4+64]
@ -287,7 +284,7 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+%3], m3
%endmacro
%macro TRANS_ADD32_AVX2 3
%macro ADD_RES_AVX2_32_10 3
mova m0, [%3]
mova m1, [%3+32]
mova m2, [%3+64]
@ -308,81 +305,77 @@ cglobal hevc_add_residual32_8, 3, 4, 7
mova [%1+%2+32], m3
%endmacro
; void ff_hevc_add_residual_<4|8|16|32>_10(pixel *dst, int16_t *block, ptrdiff_t stride)
INIT_MMX mmxext
cglobal hevc_add_residual4_10,3,4, 6
cglobal hevc_add_residual_4_10, 3, 4, 6
pxor m2, m2
mova m3, [max_pixels_10]
TR_ADD_MMX4_10 r0, r2, r1
ADD_RES_MMX_4_10 r0, r2, r1
add r1, 16
lea r0, [r0+2*r2]
TR_ADD_MMX4_10 r0, r2, r1
ADD_RES_MMX_4_10 r0, r2, r1
RET
;-----------------------------------------------------------------------------
; void ff_hevc_add_residual_10(pixel *dst, int16_t *block, int stride)
;-----------------------------------------------------------------------------
INIT_XMM sse2
cglobal hevc_add_residual8_10,3,4,6
cglobal hevc_add_residual_8_10, 3, 4, 6
pxor m4, m4
mova m5, [max_pixels_10]
lea r3, [r2*3]
TR_ADD_SSE_8_10 r0, r2, r3, r1
ADD_RES_SSE_8_10 r0, r2, r3, r1
lea r0, [r0+r2*4]
add r1, 64
TR_ADD_SSE_8_10 r0, r2, r3, r1
ADD_RES_SSE_8_10 r0, r2, r3, r1
RET
cglobal hevc_add_residual16_10,3,4,6
cglobal hevc_add_residual_16_10, 3, 4, 6
pxor m4, m4
mova m5, [max_pixels_10]
TRANS_ADD_SSE_16_10 r0, r2, r1
ADD_RES_SSE_16_10 r0, r2, r1
%rep 7
lea r0, [r0+r2*2]
add r1, 64
TRANS_ADD_SSE_16_10 r0, r2, r1
ADD_RES_SSE_16_10 r0, r2, r1
%endrep
RET
cglobal hevc_add_residual32_10,3,4,6
cglobal hevc_add_residual_32_10, 3, 4, 6
pxor m4, m4
mova m5, [max_pixels_10]
TRANS_ADD_SSE_32_10 r0, r1
ADD_RES_SSE_32_10 r0, r1
%rep 31
lea r0, [r0+r2]
add r1, 64
TRANS_ADD_SSE_32_10 r0, r1
ADD_RES_SSE_32_10 r0, r1
%endrep
RET
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
cglobal hevc_add_residual16_10,3,4,6
cglobal hevc_add_residual_16_10, 3, 4, 6
pxor m4, m4
mova m5, [max_pixels_10]
lea r3, [r2*3]
TRANS_ADD16_AVX2 r0, r2, r3, r1
ADD_RES_AVX2_16_10 r0, r2, r3, r1
%rep 3
lea r0, [r0+r2*4]
add r1, 128
TRANS_ADD16_AVX2 r0, r2, r3, r1
ADD_RES_AVX2_16_10 r0, r2, r3, r1
%endrep
RET
cglobal hevc_add_residual32_10,3,4,6
cglobal hevc_add_residual_32_10, 3, 4, 6
pxor m4, m4
mova m5, [max_pixels_10]
TRANS_ADD32_AVX2 r0, r2, r1
ADD_RES_AVX2_32_10 r0, r2, r1
%rep 15
lea r0, [r0+r2*2]
add r1, 128
TRANS_ADD32_AVX2 r0, r2, r1
ADD_RES_AVX2_32_10 r0, r2, r1
%endrep
RET
%endif ;HAVE_AVX_EXTERNAL
%endif ;HAVE_AVX2_EXTERNAL

29
libavcodec/x86/hevcdsp.h
Unescape View File

@ -236,23 +236,24 @@ WEIGHTING_PROTOTYPES(12, sse4);
///////////////////////////////////////////////////////////////////////////////
// TRANSFORM_ADD
///////////////////////////////////////////////////////////////////////////////
void ff_hevc_add_residual4_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual8_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual16_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual32_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual8_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual16_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual32_8_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_8_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_16_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_32_8_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual32_8_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual_8_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_16_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_32_8_avx(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual4_10_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual8_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual16_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual32_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual_32_8_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual16_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual32_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void ff_hevc_add_residual_4_10_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_8_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_16_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_32_10_sse2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_16_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
void ff_hevc_add_residual_32_10_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride);
#endif // AVCODEC_X86_HEVCDSP_H

30
libavcodec/x86/hevcdsp_init.c
Unescape View File

@ -713,7 +713,8 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_mmxext;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_mmxext;
c->add_residual[0] = ff_hevc_add_residual4_8_mmxext;
c->add_residual[0] = ff_hevc_add_residual_4_8_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
@ -734,9 +735,9 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
c->idct[0] = ff_hevc_idct_4x4_8_sse2;
c->idct[1] = ff_hevc_idct_8x8_8_sse2;
c->add_residual[1] = ff_hevc_add_residual8_8_sse2;
c->add_residual[2] = ff_hevc_add_residual16_8_sse2;
c->add_residual[3] = ff_hevc_add_residual32_8_sse2;
c->add_residual[1] = ff_hevc_add_residual_8_8_sse2;
c->add_residual[2] = ff_hevc_add_residual_16_8_sse2;
c->add_residual[3] = ff_hevc_add_residual_32_8_sse2;
}
if (EXTERNAL_SSSE3(cpu_flags)) {
if(ARCH_X86_64) {
@ -772,9 +773,9 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
c->idct[0] = ff_hevc_idct_4x4_8_avx;
c->idct[1] = ff_hevc_idct_8x8_8_avx;
c->add_residual[1] = ff_hevc_add_residual8_8_avx;
c->add_residual[2] = ff_hevc_add_residual16_8_avx;
c->add_residual[3] = ff_hevc_add_residual32_8_avx;
c->add_residual[1] = ff_hevc_add_residual_8_8_avx;
c->add_residual[2] = ff_hevc_add_residual_16_8_avx;
c->add_residual[3] = ff_hevc_add_residual_32_8_avx;
}
if (EXTERNAL_AVX2(cpu_flags)) {
c->sao_band_filter[0] = ff_hevc_sao_band_filter_8_8_avx2;
@ -874,11 +875,11 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
c->sao_edge_filter[3] = ff_hevc_sao_edge_filter_48_8_avx2;
c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_64_8_avx2;
c->add_residual[3] = ff_hevc_add_residual32_8_avx2;
c->add_residual[3] = ff_hevc_add_residual_32_8_avx2;
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->add_residual[0] = ff_hevc_add_residual4_10_mmxext;
c->add_residual[0] = ff_hevc_add_residual_4_10_mmxext;
c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
}
@ -902,9 +903,9 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
c->idct[0] = ff_hevc_idct_4x4_10_sse2;
c->idct[1] = ff_hevc_idct_8x8_10_sse2;
c->add_residual[1] = ff_hevc_add_residual8_10_sse2;
c->add_residual[2] = ff_hevc_add_residual16_10_sse2;
c->add_residual[3] = ff_hevc_add_residual32_10_sse2;
c->add_residual[1] = ff_hevc_add_residual_8_10_sse2;
c->add_residual[2] = ff_hevc_add_residual_16_10_sse2;
c->add_residual[3] = ff_hevc_add_residual_32_10_sse2;
}
if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
@ -1090,9 +1091,8 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
SAO_BAND_INIT(10, avx2);
SAO_EDGE_INIT(10, avx2);
c->add_residual[2] = ff_hevc_add_residual16_10_avx2;
c->add_residual[3] = ff_hevc_add_residual32_10_avx2;
c->add_residual[2] = ff_hevc_add_residual_16_10_avx2;
c->add_residual[3] = ff_hevc_add_residual_32_10_avx2;
}
} else if (bit_depth == 12) {
if (EXTERNAL_MMXEXT(cpu_flags)) {

Write Preview
Loading…
Cancel
Save