[FFmpeg-cvslog] hevc: x86: Add add_residual() SIMD optimizations
Pierre Edouard Lepere
git at videolan.org
Fri Mar 24 13:34:18 EET 2017
ffmpeg | branch: master | Pierre Edouard Lepere <Pierre-Edouard.Lepere at insa-rennes.fr> | Mon Aug 18 10:01:09 2014 +0200| [6d5636ad9ab6bd9bedf902051d88b7044385f88b] | committer: Diego Biurrun
hevc: x86: Add add_residual() SIMD optimizations
Initially written by Pierre Edouard Lepere <Pierre-Edouard.Lepere at insa-rennes.fr>,
extended by James Almer <jamrial at gmail.com>.
Signed-off-by: Alexandra Hájková <alexandra at khirnov.net>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=6d5636ad9ab6bd9bedf902051d88b7044385f88b
---
libavcodec/hevcdsp.h | 2 +-
libavcodec/x86/Makefile | 7 +-
libavcodec/x86/hevc_add_res.asm | 369 ++++++++++++++++++++++++++++++++++++++++
libavcodec/x86/hevcdsp_init.c | 42 +++++
4 files changed, 416 insertions(+), 4 deletions(-)
diff --git a/libavcodec/hevcdsp.h b/libavcodec/hevcdsp.h
index 199e5a9..49cb711 100644
--- a/libavcodec/hevcdsp.h
+++ b/libavcodec/hevcdsp.h
@@ -42,7 +42,7 @@ typedef struct HEVCDSPContext {
void (*put_pcm)(uint8_t *dst, ptrdiff_t stride, int size,
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);
void (*transform_4x4_luma)(int16_t *coeffs);
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index a38535b..094c1fa 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -115,9 +115,10 @@ YASM-OBJS-$(CONFIG_AAC_DECODER) += x86/sbrdsp.o
YASM-OBJS-$(CONFIG_APE_DECODER) += x86/apedsp.o
YASM-OBJS-$(CONFIG_DCA_DECODER) += x86/dcadsp.o
YASM-OBJS-$(CONFIG_DNXHD_ENCODER) += x86/dnxhdenc.o
-YASM-OBJS-$(CONFIG_HEVC_DECODER) += x86/hevc_deblock.o \
- x86/hevc_mc.o \
- x86/hevc_idct.o
+YASM-OBJS-$(CONFIG_HEVC_DECODER) += x86/hevc_add_res.o \
+ x86/hevc_deblock.o \
+ x86/hevc_idct.o \
+ x86/hevc_mc.o
YASM-OBJS-$(CONFIG_PNG_DECODER) += x86/pngdsp.o
YASM-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp.o
YASM-OBJS-$(CONFIG_RV40_DECODER) += x86/rv40dsp.o
diff --git a/libavcodec/x86/hevc_add_res.asm b/libavcodec/x86/hevc_add_res.asm
new file mode 100644
index 0000000..66b929c
--- /dev/null
+++ b/libavcodec/x86/hevc_add_res.asm
@@ -0,0 +1,369 @@
+; *****************************************************************************
+; * Provide SIMD optimizations for add_residual functions for HEVC decoding
+; * Copyright (c) 2014 Pierre-Edouard LEPERE
+; *
+; * This file is part of Libav.
+; *
+; * Libav is free software; you can redistribute it and/or
+; * modify it under the terms of the GNU Lesser General Public
+; * License as published by the Free Software Foundation; either
+; * version 2.1 of the License, or (at your option) any later version.
+; *
+; * Libav is distributed in the hope that it will be useful,
+; * but WITHOUT ANY WARRANTY; without even the implied warranty of
+; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+; * Lesser General Public License for more details.
+; *
+; * You should have received a copy of the GNU Lesser General Public
+; * License along with Libav; 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_RODATA 32
+max_pixels_10: times 16 dw ((1 << 10)-1)
+
+SECTION .text
+
+; 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 m0, [r1]
+ mova m2, [r1+8]
+ pxor m1, m1
+ pxor m3, m3
+ psubw m1, m0
+ psubw m3, m2
+ packuswb m0, m2
+ packuswb m1, m3
+
+ movd m2, [r0]
+ movd m3, [r0+r2]
+ punpckldq m2, m3
+ paddusb m0, m2
+ psubusb m0, m1
+ movd [r0], m0
+ psrlq m0, 32
+ movd [r0+r2], m0
+%endmacro
+
+
+INIT_MMX mmxext
+; void ff_hevc_add_residual_4_8_mmxext(uint8_t *dst, int16_t *res, ptrdiff_t stride)
+cglobal hevc_add_residual_4_8, 3, 3, 6
+ ADD_RES_MMX_4_8
+ add r1, 16
+ lea r0, [r0+r2*2]
+ ADD_RES_MMX_4_8
+ RET
+
+%macro ADD_RES_SSE_8_8 0
+ pxor m3, m3
+ mova m4, [r1]
+ mova m6, [r1+16]
+ mova m0, [r1+32]
+ mova m2, [r1+48]
+ psubw m5, m3, m4
+ psubw m7, m3, m6
+ psubw m1, m3, m0
+ packuswb m4, m0
+ packuswb m5, m1
+ psubw m3, m2
+ packuswb m6, m2
+ packuswb m7, m3
+
+ movq m0, [r0]
+ movq m1, [r0+r2]
+ movhps m0, [r0+r2*2]
+ movhps m1, [r0+r3]
+ paddusb m0, m4
+ paddusb m1, m6
+ psubusb m0, m5
+ psubusb m1, m7
+ movq [r0], m0
+ movq [r0+r2], m1
+ movhps [r0+2*r2], m0
+ movhps [r0+r3], m1
+%endmacro
+
+%macro ADD_RES_SSE_16_32_8 3
+ mova xm2, [r1+%1]
+ mova xm6, [r1+%1+16]
+%if cpuflag(avx2)
+ vinserti128 m2, m2, [r1+%1+32], 1
+ vinserti128 m6, m6, [r1+%1+48], 1
+%endif
+ psubw m1, m0, m2
+ psubw m5, m0, m6
+ packuswb m2, m6
+ packuswb m1, m5
+
+ mova xm4, [r1+%1+mmsize*2]
+ mova xm6, [r1+%1+mmsize*2+16]
+%if cpuflag(avx2)
+ vinserti128 m4, m4, [r1+%1+96 ], 1
+ vinserti128 m6, m6, [r1+%1+112], 1
+%endif
+ psubw m3, m0, m4
+ psubw m5, m0, m6
+ packuswb m4, m6
+ packuswb m3, m5
+
+ paddusb m2, [%2]
+ paddusb m4, [%3]
+ psubusb m2, m1
+ psubusb m4, m3
+ mova [%2], m2
+ mova [%3], m4
+%endmacro
+
+
+%macro TRANSFORM_ADD_8 0
+; 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]
+ ADD_RES_SSE_8_8
+ add r1, 64
+ lea r0, [r0+r2*4]
+ ADD_RES_SSE_8_8
+ RET
+
+; void ff_hevc_add_residual_16_8_<opt>(uint8_t *dst, int16_t *res, ptrdiff_t stride)
+cglobal hevc_add_residual_16_8, 3, 5, 7
+ pxor m0, m0
+ lea r3, [r2*3]
+ mov r4d, 4
+.loop:
+ ADD_RES_SSE_16_32_8 0, r0, r0+r2
+ ADD_RES_SSE_16_32_8 64, r0+r2*2, r0+r3
+ add r1, 128
+ lea r0, [r0+r2*4]
+ dec r4d
+ jg .loop
+ RET
+
+; void ff_hevc_add_residual_32_8_<opt>(uint8_t *dst, int16_t *res, ptrdiff_t stride)
+cglobal hevc_add_residual_32_8, 3, 5, 7
+ pxor m0, m0
+ mov r4d, 16
+.loop:
+ ADD_RES_SSE_16_32_8 0, r0, r0+16
+ ADD_RES_SSE_16_32_8 64, r0+r2, r0+r2+16
+ add r1, 128
+ lea r0, [r0+r2*2]
+ dec r4d
+ jg .loop
+ RET
+%endmacro
+
+INIT_XMM sse2
+TRANSFORM_ADD_8
+INIT_XMM avx
+TRANSFORM_ADD_8
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+; void ff_hevc_add_residual_32_8_avx2(uint8_t *dst, int16_t *res, ptrdiff_t stride)
+cglobal hevc_add_residual_32_8, 3, 5, 7
+ pxor m0, m0
+ lea r3, [r2*3]
+ mov r4d, 8
+.loop:
+ ADD_RES_SSE_16_32_8 0, r0, r0+r2
+ ADD_RES_SSE_16_32_8 128, r0+r2*2, r0+r3
+ add r1, 256
+ lea r0, [r0+r2*4]
+ dec r4d
+ jg .loop
+ RET
+%endif ;HAVE_AVX2_EXTERNAL
+
+%macro ADD_RES_SSE_8_10 4
+ mova m0, [%4]
+ mova m1, [%4+16]
+ mova m2, [%4+32]
+ mova m3, [%4+48]
+ paddw m0, [%1+0]
+ paddw m1, [%1+%2]
+ paddw m2, [%1+%2*2]
+ paddw m3, [%1+%3]
+ CLIPW m0, m4, m5
+ CLIPW m1, m4, m5
+ CLIPW m2, m4, m5
+ CLIPW m3, m4, m5
+ mova [%1+0], m0
+ mova [%1+%2], m1
+ mova [%1+%2*2], m2
+ mova [%1+%3], m3
+%endmacro
+
+%macro ADD_RES_MMX_4_10 3
+ mova m0, [%1+0]
+ mova m1, [%1+%2]
+ paddw m0, [%3]
+ paddw m1, [%3+8]
+ CLIPW m0, m2, m3
+ CLIPW m1, m2, m3
+ mova [%1+0], m0
+ mova [%1+%2], m1
+%endmacro
+
+%macro ADD_RES_SSE_16_10 3
+ mova m0, [%3]
+ mova m1, [%3+16]
+ mova m2, [%3+32]
+ mova m3, [%3+48]
+ paddw m0, [%1]
+ paddw m1, [%1+16]
+ paddw m2, [%1+%2]
+ paddw m3, [%1+%2+16]
+ CLIPW m0, m4, m5
+ CLIPW m1, m4, m5
+ CLIPW m2, m4, m5
+ CLIPW m3, m4, m5
+ mova [%1], m0
+ mova [%1+16], m1
+ mova [%1+%2], m2
+ mova [%1+%2+16], m3
+%endmacro
+
+%macro ADD_RES_SSE_32_10 2
+ mova m0, [%2]
+ mova m1, [%2+16]
+ mova m2, [%2+32]
+ mova m3, [%2+48]
+
+ paddw m0, [%1]
+ paddw m1, [%1+16]
+ paddw m2, [%1+32]
+ paddw m3, [%1+48]
+ CLIPW m0, m4, m5
+ CLIPW m1, m4, m5
+ CLIPW m2, m4, m5
+ CLIPW m3, m4, m5
+ mova [%1], m0
+ mova [%1+16], m1
+ mova [%1+32], m2
+ mova [%1+48], m3
+%endmacro
+
+%macro ADD_RES_AVX2_16_10 4
+ mova m0, [%4]
+ mova m1, [%4+32]
+ mova m2, [%4+64]
+ mova m3, [%4+96]
+
+ paddw m0, [%1+0]
+ paddw m1, [%1+%2]
+ paddw m2, [%1+%2*2]
+ paddw m3, [%1+%3]
+
+ CLIPW m0, m4, m5
+ CLIPW m1, m4, m5
+ CLIPW m2, m4, m5
+ CLIPW m3, m4, m5
+ mova [%1+0], m0
+ mova [%1+%2], m1
+ mova [%1+%2*2], m2
+ mova [%1+%3], m3
+%endmacro
+
+%macro ADD_RES_AVX2_32_10 3
+ mova m0, [%3]
+ mova m1, [%3+32]
+ mova m2, [%3+64]
+ mova m3, [%3+96]
+
+ paddw m0, [%1]
+ paddw m1, [%1+32]
+ paddw m2, [%1+%2]
+ paddw m3, [%1+%2+32]
+
+ CLIPW m0, m4, m5
+ CLIPW m1, m4, m5
+ CLIPW m2, m4, m5
+ CLIPW m3, m4, m5
+ mova [%1], m0
+ mova [%1+32], m1
+ mova [%1+%2], m2
+ 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_residual_4_10, 3, 3, 6
+ pxor m2, m2
+ mova m3, [max_pixels_10]
+ ADD_RES_MMX_4_10 r0, r2, r1
+ add r1, 16
+ lea r0, [r0+2*r2]
+ ADD_RES_MMX_4_10 r0, r2, r1
+ RET
+
+INIT_XMM sse2
+cglobal hevc_add_residual_8_10, 3, 4, 6
+ pxor m4, m4
+ mova m5, [max_pixels_10]
+ lea r3, [r2*3]
+
+ ADD_RES_SSE_8_10 r0, r2, r3, r1
+ lea r0, [r0+r2*4]
+ add r1, 64
+ ADD_RES_SSE_8_10 r0, r2, r3, r1
+ RET
+
+cglobal hevc_add_residual_16_10, 3, 5, 6
+ pxor m4, m4
+ mova m5, [max_pixels_10]
+
+ mov r4d, 8
+.loop:
+ ADD_RES_SSE_16_10 r0, r2, r1
+ lea r0, [r0+r2*2]
+ add r1, 64
+ dec r4d
+ jg .loop
+ RET
+
+cglobal hevc_add_residual_32_10, 3, 5, 6
+ pxor m4, m4
+ mova m5, [max_pixels_10]
+
+ mov r4d, 32
+.loop
+ ADD_RES_SSE_32_10 r0, r1
+ lea r0, [r0+r2]
+ add r1, 64
+ dec r4d
+ jg .loop
+ RET
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+cglobal hevc_add_residual_16_10, 3, 5, 6
+ pxor m4, m4
+ mova m5, [max_pixels_10]
+ lea r3, [r2*3]
+
+ mov r4d, 4
+.loop
+ ADD_RES_AVX2_16_10 r0, r2, r3, r1
+ lea r0, [r0+r2*4]
+ add r1, 128
+ dec r4d
+ jg .loop
+ RET
+
+cglobal hevc_add_residual_32_10, 3, 5, 6
+ pxor m4, m4
+ mova m5, [max_pixels_10]
+
+ mov r4d, 16
+.loop
+ ADD_RES_AVX2_32_10 r0, r2, r1
+ lea r0, [r0+r2*2]
+ add r1, 128
+ dec r4d
+ jg .loop
+ RET
+%endif ;HAVE_AVX2_EXTERNAL
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index 0a06347..a95fa30 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -91,6 +91,25 @@ void ff_hevc_idct_32x32_10_ ## opt(int16_t *coeffs, int col_limit);
IDCT_FUNCS(sse2)
IDCT_FUNCS(avx)
+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_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_residual_32_8_avx2(uint8_t *dst, int16_t *res, 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);
+
#define GET_PIXELS(width, depth, cf) \
void ff_hevc_get_pixels_ ## width ## _ ## depth ## _ ## cf(int16_t *dst, ptrdiff_t dststride, \
uint8_t *src, ptrdiff_t srcstride, \
@@ -278,17 +297,24 @@ 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_residual_4_8_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_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;
+
c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_sse2;
c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_sse2;
c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_sse2;
c->idct[0] = ff_hevc_idct_4x4_8_sse2;
c->idct[1] = ff_hevc_idct_8x8_8_sse2;
+
SET_QPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
SET_EPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
@@ -307,11 +333,19 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
if (EXTERNAL_AVX(cpu_flags)) {
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_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->add_residual[3] = ff_hevc_add_residual_32_8_avx2;
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
+
+ c->add_residual[0] = ff_hevc_add_residual_4_10_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
@@ -330,11 +364,19 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 10, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 10, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 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_AVX(cpu_flags)) {
c->idct[0] = ff_hevc_idct_4x4_10_avx;
c->idct[1] = ff_hevc_idct_8x8_10_avx;
}
+ if (EXTERNAL_AVX2(cpu_flags)) {
+ c->add_residual[2] = ff_hevc_add_residual_16_10_avx2;
+ c->add_residual[3] = ff_hevc_add_residual_32_10_avx2;
+ }
}
#if ARCH_X86_64
More information about the ffmpeg-cvslog
mailing list