Skip to content

F
ffmpeg.wasm-core
Commit a8d44f9d authored by Daniel Kang's avatar Daniel Kang Committed by Diego Biurrun
Browse files
Options

Add x86 assembly for some 10-bit H.264 intra predict functions. 

Parts are inspired from the 8-bit H.264 predict code in Libav.
Other parts ported from x264 with relicensing permission from author.
Signed-off-by: 's avatarDiego Biurrun <diego@biurrun.de>
parent 2c6fb9f0
Hide whitespace changes
Inline Side-by-side
Showing with 396 additions and 5 deletions
libavcodec/x86/Makefile
View file @ a8d44f9d
...@@ -16,7 +16,8 @@ YASM-OBJS-$(CONFIG_H264DSP) += x86/h264_deblock.o \ ...@@ -16,7 +16,8 @@ YASM-OBJS-$(CONFIG_H264DSP) += x86/h264_deblock.o \
x86/h264_idct_10bit.o \ x86/h264_idct_10bit.o \
x86/h264_weight.o \ x86/h264_weight.o \
YASM-OBJS-$(CONFIG_H264PRED) += x86/h264_intrapred.o YASM-OBJS-$(CONFIG_H264PRED) += x86/h264_intrapred.o \
x86/h264_intrapred_10bit.o
MMX-OBJS-$(CONFIG_H264PRED) += x86/h264_intrapred_init.o MMX-OBJS-$(CONFIG_H264PRED) += x86/h264_intrapred_init.o
YASM-OBJS-$(CONFIG_VC1_DECODER) += x86/vc1dsp_yasm.o YASM-OBJS-$(CONFIG_VC1_DECODER) += x86/vc1dsp_yasm.o
......
libavcodec/x86/h264_intrapred_10bit.asm 0 → 100644
View file @ a8d44f9d
;*****************************************************************************
;* MMX/SSE2/AVX-optimized 10-bit H.264 intra prediction code
;*****************************************************************************
;* Copyright (C) 2005-2011 x264 project
;*
;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
;*
;* 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 "x86inc.asm"
%include "x86util.asm"
SECTION_RODATA
SECTION .text
cextern pw_4
cextern pw_1
%macro PRED4x4_LOWPASS 4
paddw %2, %3
psrlw %2, 1
pavgw %1, %4, %2
%endmacro
;-----------------------------------------------------------------------------
; void pred4x4_down_right(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
%macro PRED4x4_DR 1
cglobal pred4x4_down_right_10_%1, 3,3
sub r0, r2
lea r1, [r0+r2*2]
movhps m1, [r1-8]
movhps m2, [r0+r2*1-8]
movhps m4, [r0-8]
punpckhwd m2, m4
movq m3, [r0]
punpckhdq m1, m2
PALIGNR m3, m1, 10, m1
mova m1, m3
movhps m4, [r1+r2*1-8]
PALIGNR m3, m4, 14, m4
mova m2, m3
movhps m4, [r1+r2*2-8]
PALIGNR m3, m4, 14, m4
PRED4x4_LOWPASS m0, m3, m1, m2
movq [r1+r2*2], m0
psrldq m0, 2
movq [r1+r2*1], m0
psrldq m0, 2
movq [r0+r2*2], m0
psrldq m0, 2
movq [r0+r2*1], m0
RET
%endmacro
INIT_XMM
%define PALIGNR PALIGNR_MMX
PRED4x4_DR sse2
%define PALIGNR PALIGNR_SSSE3
PRED4x4_DR ssse3
%ifdef HAVE_AVX
INIT_AVX
PRED4x4_DR avx
%endif
;-----------------------------------------------------------------------------
; void pred4x4_vertical_right(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
%macro PRED4x4_VR 1
cglobal pred4x4_vertical_right_10_%1, 3,3,6
sub r0, r2
lea r1, [r0+r2*2]
movq m5, [r0] ; ........t3t2t1t0
movhps m1, [r0-8]
PALIGNR m0, m5, m1, 14, m1 ; ......t3t2t1t0lt
pavgw m5, m0
movhps m1, [r0+r2*1-8]
PALIGNR m0, m1, 14, m1 ; ....t3t2t1t0ltl0
mova m1, m0
movhps m2, [r0+r2*2-8]
PALIGNR m0, m2, 14, m2 ; ..t3t2t1t0ltl0l1
mova m2, m0
movhps m3, [r1+r2*1-8]
PALIGNR m0, m3, 14, m3 ; t3t2t1t0ltl0l1l2
PRED4x4_LOWPASS m3, m1, m0, m2
pslldq m1, m3, 12
psrldq m3, 4
movq [r0+r2*1], m5
movq [r0+r2*2], m3
PALIGNR m5, m1, 14, m2
pslldq m1, 2
movq [r1+r2*1], m5
PALIGNR m3, m1, 14, m1
movq [r1+r2*2], m3
RET
%endmacro
INIT_XMM
%define PALIGNR PALIGNR_MMX
PRED4x4_VR sse2
%define PALIGNR PALIGNR_SSSE3
PRED4x4_VR ssse3
%ifdef HAVE_AVX
INIT_AVX
PRED4x4_VR avx
%endif
;-----------------------------------------------------------------------------
; void pred4x4_horizontal_down(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
%macro PRED4x4_HD 1
cglobal pred4x4_horizontal_down_10_%1, 3,3
sub r0, r2
lea r1, [r0+r2*2]
movq m0, [r0-8] ; lt ..
movhps m0, [r0]
pslldq m0, 2 ; t2 t1 t0 lt .. .. .. ..
movq m1, [r1+r2*2-8] ; l3
movq m3, [r1+r2*1-8]
punpcklwd m1, m3 ; l2 l3
movq m2, [r0+r2*2-8] ; l1
movq m3, [r0+r2*1-8]
punpcklwd m2, m3 ; l0 l1
punpckhdq m1, m2 ; l0 l1 l2 l3
punpckhqdq m1, m0 ; t2 t1 t0 lt l0 l1 l2 l3
psrldq m0, m1, 4 ; .. .. t2 t1 t0 lt l0 l1
psrldq m2, m1, 2 ; .. t2 t1 t0 lt l0 l1 l2
pavgw m5, m1, m2
PRED4x4_LOWPASS m3, m1, m0, m2
punpcklwd m5, m3
psrldq m3, 8
PALIGNR m3, m5, 12, m4
movq [r1+r2*2], m5
movhps [r0+r2*2], m5
psrldq m5, 4
movq [r1+r2*1], m5
movq [r0+r2*1], m3
RET
%endmacro
INIT_XMM
%define PALIGNR PALIGNR_MMX
PRED4x4_HD sse2
%define PALIGNR PALIGNR_SSSE3
PRED4x4_HD ssse3
%ifdef HAVE_AVX
INIT_AVX
PRED4x4_HD avx
%endif
;-----------------------------------------------------------------------------
; void pred4x4_dc(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
%macro HADDD 2 ; sum junk
%if mmsize == 16
movhlps %2, %1
paddd %1, %2
pshuflw %2, %1, 0xE
paddd %1, %2
%else
pshufw %2, %1, 0xE
paddd %1, %2
%endif
%endmacro
%macro HADDW 2
pmaddwd %1, [pw_1]
HADDD %1, %2
%endmacro
INIT_MMX
cglobal pred4x4_dc_10_mmxext, 3,3
sub r0, r2
lea r1, [r0+r2*2]
movq m2, [r0+r2*1-8]
paddw m2, [r0+r2*2-8]
paddw m2, [r1+r2*1-8]
paddw m2, [r1+r2*2-8]
psrlq m2, 48
movq m0, [r0]
HADDW m0, m1
paddw m0, [pw_4]
paddw m0, m2
psrlw m0, 3
SPLATW m0, m0, 0
movq [r0+r2*1], m0
movq [r0+r2*2], m0
movq [r1+r2*1], m0
movq [r1+r2*2], m0
RET
;-----------------------------------------------------------------------------
; void pred4x4_down_left(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
;TODO: more AVX here
%macro PRED4x4_DL 1
cglobal pred4x4_down_left_10_%1, 3,3
sub r0, r2
movq m1, [r0]
movhps m1, [r1]
pslldq m5, m1, 2
pxor m2, m5, m1
psrldq m2, 2
pxor m3, m1, m2
PRED4x4_LOWPASS m0, m5, m3, m1
lea r1, [r0+r2*2]
movhps [r1+r2*2], m0
psrldq m0, 2
movq [r0+r2*1], m0
psrldq m0, 2
movq [r0+r2*2], m0
psrldq m0, 2
movq [r1+r2*1], m0
RET
%endmacro
INIT_XMM
PRED4x4_DL sse2
%ifdef HAVE_AVX
INIT_AVX
PRED4x4_DL avx
%endif
;-----------------------------------------------------------------------------
; void pred4x4_vertical_left(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
%macro PRED4x4_VL 1
cglobal pred4x4_vertical_left_10_%1, 3,3
sub r0, r2
movu m1, [r0]
movhps m1, [r1]
psrldq m3, m1, 2
psrldq m2, m1, 4
pavgw m4, m3, m1
PRED4x4_LOWPASS m0, m1, m2, m3
lea r1, [r0+r2*2]
movq [r0+r2*1], m4
movq [r0+r2*2], m0
psrldq m4, 2
psrldq m0, 2
movq [r1+r2*1], m4
movq [r1+r2*2], m0
RET
%endmacro
INIT_XMM
PRED4x4_VL sse2
%ifdef HAVE_AVX
INIT_AVX
PRED4x4_VL avx
%endif
;-----------------------------------------------------------------------------
; void pred4x4_horizontal_up(pixel *src, const pixel *topright, int stride)
;-----------------------------------------------------------------------------
INIT_MMX
cglobal pred4x4_horizontal_up_10_mmxext, 3,3
sub r0, r2
lea r1, [r0+r2*2]
movq m0, [r0+r2*1-8]
punpckhwd m0, [r0+r2*2-8]
movq m1, [r1+r2*1-8]
punpckhwd m1, [r1+r2*2-8]
punpckhdq m0, m1
pshufw m1, m1, 0xFF
movq [r1+r2*2], m1
movd [r1+r2*1+4], m1
pshufw m2, m0, 11111001b
movq m1, m2
pavgw m2, m0
pshufw m5, m0, 11111110b
PRED4x4_LOWPASS m3, m0, m5, m1
movq m6, m2
punpcklwd m6, m3
movq [r0+r2*1], m6
psrlq m2, 16
psrlq m3, 16
punpcklwd m2, m3
movq [r0+r2*2], m2
psrlq m2, 32
movd [r1+r2*1], m2
RET
;-----------------------------------------------------------------------------
; void pred8x8_vertical(pixel *src, int stride)
;-----------------------------------------------------------------------------
INIT_XMM
cglobal pred8x8_vertical_10_sse2, 2,2
sub r0, r1
mova m0, [r0]
%rep 3
mova [r0+r1*1], m0
mova [r0+r1*2], m0
lea r0, [r0+r1*2]
%endrep
mova [r0+r1*1], m0
mova [r0+r1*2], m0
RET
;-----------------------------------------------------------------------------
; void pred8x8_horizontal(pixel *src, int stride)
;-----------------------------------------------------------------------------
INIT_XMM
cglobal pred8x8_horizontal_10_sse2, 2,3
mov r2, 4
.loop:
movq m0, [r0+r1*0-8]
movq m1, [r0+r1*1-8]
pshuflw m0, m0, 0xff
pshuflw m1, m1, 0xff
punpcklqdq m0, m0
punpcklqdq m1, m1
mova [r0+r1*0], m0
mova [r0+r1*1], m1
lea r0, [r0+r1*2]
dec r2
jg .loop
REP_RET
libavcodec/x86/h264_intrapred_init.c
View file @ a8d44f9d
...@@ -21,6 +21,31 @@ ...@@ -21,6 +21,31 @@
#include "libavutil/cpu.h" #include "libavutil/cpu.h"
#include "libavcodec/h264pred.h" #include "libavcodec/h264pred.h"
#define PRED4x4(TYPE, DEPTH, OPT) \
void ff_pred4x4_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, const uint8_t *topright, int stride);
PRED4x4(dc, 10, mmxext)
PRED4x4(down_left, 10, sse2)
PRED4x4(down_left, 10, avx)
PRED4x4(down_right, 10, sse2)
PRED4x4(down_right, 10, ssse3)
PRED4x4(down_right, 10, avx)
PRED4x4(vertical_left, 10, sse2)
PRED4x4(vertical_left, 10, avx)
PRED4x4(vertical_right, 10, sse2)
PRED4x4(vertical_right, 10, ssse3)
PRED4x4(vertical_right, 10, avx)
PRED4x4(horizontal_up, 10, mmxext)
PRED4x4(horizontal_down, 10, sse2)
PRED4x4(horizontal_down, 10, ssse3)
PRED4x4(horizontal_down, 10, avx)
#define PRED8x8(TYPE, DEPTH, OPT) \
void ff_pred8x8_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
PRED8x8(vertical, 10, sse2)
PRED8x8(horizontal, 10, sse2)
void ff_pred16x16_vertical_mmx (uint8_t *src, int stride); void ff_pred16x16_vertical_mmx (uint8_t *src, int stride);
void ff_pred16x16_vertical_sse (uint8_t *src, int stride); void ff_pred16x16_vertical_sse (uint8_t *src, int stride);
void ff_pred16x16_horizontal_mmx (uint8_t *src, int stride); void ff_pred16x16_horizontal_mmx (uint8_t *src, int stride);
...@@ -98,11 +123,8 @@ void ff_pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int s ...@@ -98,11 +123,8 @@ void ff_pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int s
void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth) void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth)
{ {
int mm_flags = av_get_cpu_flags(); int mm_flags = av_get_cpu_flags();
const int high_depth = bit_depth > 8;
if (high_depth)
return;
if (bit_depth == 8) {
#if HAVE_YASM #if HAVE_YASM
if (mm_flags & AV_CPU_FLAG_MMX) { if (mm_flags & AV_CPU_FLAG_MMX) {
h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx; h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx;
...@@ -226,4 +248,35 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth ...@@ -226,4 +248,35 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
} }
} }
#endif #endif
} else if (bit_depth == 10) {
#if HAVE_YASM
if (mm_flags & AV_CPU_FLAG_MMX2) {
h->pred4x4[DC_PRED ] = ff_pred4x4_dc_10_mmxext;
h->pred4x4[HOR_UP_PRED ] = ff_pred4x4_horizontal_up_10_mmxext;
}
if (mm_flags & AV_CPU_FLAG_SSE2) {
h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_sse2;
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_sse2;
h->pred4x4[VERT_LEFT_PRED ] = ff_pred4x4_vertical_left_10_sse2;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_sse2;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_sse2;
h->pred8x8[VERT_PRED8x8 ] = ff_pred8x8_vertical_10_sse2;
h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_horizontal_10_sse2;
}
if (mm_flags & AV_CPU_FLAG_SSSE3) {
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_ssse3;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_ssse3;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_ssse3;
}
#if HAVE_AVX
if (mm_flags&AV_CPU_FLAG_AVX) {
h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_avx;
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_avx;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_avx;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_avx;
}
#endif /* HAVE_AVX */
#endif /* HAVE_YASM */
}
} }
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment