falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
Browse Source

png: move DSP functions to their own DSP context.

tags/n0.11
Ronald S. Bultje 13 years ago
parent
f9708e9a0e
commit
e92003514d
9 changed files with 230 additions and 127 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +1
    -1
      libavcodec/Makefile
  2. +0
    -16
      libavcodec/dsputil.c
  3. +0
    -3
      libavcodec/dsputil.h
  4. +5
    -5
      libavcodec/pngdec.c
  5. +48
    -0
      libavcodec/pngdsp.c
  6. +40
    -0
      libavcodec/pngdsp.h
  7. +1
    -0
      libavcodec/x86/Makefile
  8. +0
    -102
      libavcodec/x86/dsputil_mmx.c
  9. +135
    -0
      libavcodec/x86/pngdsp-init.c

+ 1
- 1
libavcodec/Makefile View File

@@ -296,7 +296,7 @@ OBJS-$(CONFIG_PGMYUV_DECODER) += pnmdec.o pnm.o
OBJS-$(CONFIG_PGMYUV_ENCODER) += pnmenc.o pnm.o OBJS-$(CONFIG_PGMYUV_ENCODER) += pnmenc.o pnm.o
OBJS-$(CONFIG_PGSSUB_DECODER) += pgssubdec.o OBJS-$(CONFIG_PGSSUB_DECODER) += pgssubdec.o
OBJS-$(CONFIG_PICTOR_DECODER) += pictordec.o cga_data.o OBJS-$(CONFIG_PICTOR_DECODER) += pictordec.o cga_data.o
OBJS-$(CONFIG_PNG_DECODER) += png.o pngdec.o
OBJS-$(CONFIG_PNG_DECODER) += png.o pngdec.o pngdsp.o
OBJS-$(CONFIG_PNG_ENCODER) += png.o pngenc.o OBJS-$(CONFIG_PNG_ENCODER) += png.o pngenc.o
OBJS-$(CONFIG_PPM_DECODER) += pnmdec.o pnm.o OBJS-$(CONFIG_PPM_DECODER) += pnmdec.o pnm.o
OBJS-$(CONFIG_PPM_ENCODER) += pnmenc.o pnm.o OBJS-$(CONFIG_PPM_ENCODER) += pnmenc.o pnm.o


+ 0
- 16
libavcodec/dsputil.c View File

@@ -38,7 +38,6 @@
#include "config.h" #include "config.h"
#include "ac3dec.h" #include "ac3dec.h"
#include "vorbis.h" #include "vorbis.h"
#include "png.h"


uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, };
uint32_t ff_squareTbl[512] = {0, }; uint32_t ff_squareTbl[512] = {0, };
@@ -1882,17 +1881,6 @@ static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){
dst[i+0] += src[i+0]; dst[i+0] += src[i+0];
} }


static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
long i;
for(i=0; i<=w-sizeof(long); i+=sizeof(long)){
long a = *(long*)(src1+i);
long b = *(long*)(src2+i);
*(long*)(dst+i) = ((a&pb_7f) + (b&pb_7f)) ^ ((a^b)&pb_80);
}
for(; i<w; i++)
dst[i] = src1[i]+src2[i];
}

static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
long i; long i;
#if !HAVE_FAST_UNALIGNED #if !HAVE_FAST_UNALIGNED
@@ -3003,7 +2991,6 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c; c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c;


c->add_bytes= add_bytes_c; c->add_bytes= add_bytes_c;
c->add_bytes_l2= add_bytes_l2_c;
c->diff_bytes= diff_bytes_c; c->diff_bytes= diff_bytes_c;
c->add_hfyu_median_prediction= add_hfyu_median_prediction_c; c->add_hfyu_median_prediction= add_hfyu_median_prediction_c;
c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c; c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_c;
@@ -3011,9 +2998,6 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
c->add_hfyu_left_prediction_bgr32 = add_hfyu_left_prediction_bgr32_c; c->add_hfyu_left_prediction_bgr32 = add_hfyu_left_prediction_bgr32_c;
c->bswap_buf= bswap_buf; c->bswap_buf= bswap_buf;
c->bswap16_buf = bswap16_buf; c->bswap16_buf = bswap16_buf;
#if CONFIG_PNG_DECODER
c->add_png_paeth_prediction= ff_add_png_paeth_prediction;
#endif


if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
c->h263_h_loop_filter= h263_h_loop_filter_c; c->h263_h_loop_filter= h263_h_loop_filter_c;


+ 0
- 3
libavcodec/dsputil.h View File

@@ -378,7 +378,6 @@ typedef struct DSPContext {


/* huffyuv specific */ /* huffyuv specific */
void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
/** /**
* subtract huffyuv's variant of median prediction * subtract huffyuv's variant of median prediction
@@ -388,8 +387,6 @@ typedef struct DSPContext {
void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left); int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
/* this might write to dst[w] */
void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len); void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);




+ 5
- 5
libavcodec/pngdec.c View File

@@ -22,7 +22,7 @@
#include "avcodec.h" #include "avcodec.h"
#include "bytestream.h" #include "bytestream.h"
#include "png.h" #include "png.h"
#include "dsputil.h"
#include "pngdsp.h"


/* TODO: /* TODO:
* - add 2, 4 and 16 bit depth support * - add 2, 4 and 16 bit depth support
@@ -33,7 +33,7 @@
//#define DEBUG //#define DEBUG


typedef struct PNGDecContext { typedef struct PNGDecContext {
DSPContext dsp;
PNGDSPContext dsp;


const uint8_t *bytestream; const uint8_t *bytestream;
const uint8_t *bytestream_start; const uint8_t *bytestream_start;
@@ -191,7 +191,7 @@ void ff_add_png_paeth_prediction(uint8_t *dst, uint8_t *src, uint8_t *top, int w
} }


/* NOTE: 'dst' can be equal to 'last' */ /* NOTE: 'dst' can be equal to 'last' */
static void png_filter_row(DSPContext *dsp, uint8_t *dst, int filter_type,
static void png_filter_row(PNGDSPContext *dsp, uint8_t *dst, int filter_type,
uint8_t *src, uint8_t *last, int size, int bpp) uint8_t *src, uint8_t *last, int size, int bpp)
{ {
int i, p, r, g, b, a; int i, p, r, g, b, a;
@@ -235,7 +235,7 @@ static void png_filter_row(DSPContext *dsp, uint8_t *dst, int filter_type,
if(bpp > 1 && size > 4) { if(bpp > 1 && size > 4) {
// would write off the end of the array if we let it process the last pixel with bpp=3 // would write off the end of the array if we let it process the last pixel with bpp=3
int w = bpp==4 ? size : size-3; int w = bpp==4 ? size : size-3;
dsp->add_png_paeth_prediction(dst+i, src+i, last+i, w-i, bpp);
dsp->add_paeth_prediction(dst+i, src+i, last+i, w-i, bpp);
i = w; i = w;
} }
ff_add_png_paeth_prediction(dst+i, src+i, last+i, size-i, bpp); ff_add_png_paeth_prediction(dst+i, src+i, last+i, size-i, bpp);
@@ -639,7 +639,7 @@ static av_cold int png_dec_init(AVCodecContext *avctx){
s->last_picture = &s->picture2; s->last_picture = &s->picture2;
avcodec_get_frame_defaults(&s->picture1); avcodec_get_frame_defaults(&s->picture1);
avcodec_get_frame_defaults(&s->picture2); avcodec_get_frame_defaults(&s->picture2);
dsputil_init(&s->dsp, avctx);
ff_pngdsp_init(&s->dsp);


return 0; return 0;
} }


+ 48
- 0
libavcodec/pngdsp.c View File

@@ -0,0 +1,48 @@
/*
* PNG image format
* Copyright (c) 2008 Loren Merrit <lorenm@u.washington.edu>
*
* 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/common.h"
#include "png.h"
#include "pngdsp.h"

// 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size
#define pb_7f (~0UL/255 * 0x7f)
#define pb_80 (~0UL/255 * 0x80)

static void add_bytes_l2_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w)
{
long i;
for (i = 0; i <= w - sizeof(long); i += sizeof(long)) {
long a = *(long *)(src1 + i);
long b = *(long *)(src2 + i);
*(long *)(dst + i) = ((a & pb_7f) + (b & pb_7f)) ^ ((a ^ b) & pb_80);
}
for (; i < w; i++)
dst[i] = src1[i] + src2[i];
}

void ff_pngdsp_init(PNGDSPContext *dsp)
{
dsp->add_bytes_l2 = add_bytes_l2_c;
dsp->add_paeth_prediction = ff_add_png_paeth_prediction;

if (HAVE_MMX) ff_pngdsp_init_x86(dsp);
}

+ 40
- 0
libavcodec/pngdsp.h View File

@@ -0,0 +1,40 @@
/*
* PNG image format
* Copyright (c) 2008 Loren Merrit <lorenm@u.washington.edu>
*
* 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
*/

#ifndef AVCODEC_PNGDSP_H
#define AVCODEC_PNGDSP_H

#include <stdint.h>

typedef struct PNGDSPContext {
void (*add_bytes_l2)(uint8_t *dst /* align 16 */,
uint8_t *src1 /* align 16 */,
uint8_t *src2 /* align 16 */, int w);

/* this might write to dst[w] */
void (*add_paeth_prediction)(uint8_t *dst, uint8_t *src,
uint8_t *top, int w, int bpp);
} PNGDSPContext;

void ff_pngdsp_init(PNGDSPContext *dsp);
void ff_pngdsp_init_x86(PNGDSPContext *dsp);

#endif /* AVCDODEC_PNGDSP_H */

+ 1
- 0
libavcodec/x86/Makefile View File

@@ -42,6 +42,7 @@ MMX-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_mmx.o
YASM-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_yasm.o YASM-OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_yasm.o
MMX-OBJS-$(CONFIG_GPL) += x86/idct_mmx.o MMX-OBJS-$(CONFIG_GPL) += x86/idct_mmx.o
MMX-OBJS-$(CONFIG_LPC) += x86/lpc_mmx.o MMX-OBJS-$(CONFIG_LPC) += x86/lpc_mmx.o
MMX-OBJS-$(CONFIG_PNG_DECODER) += x86/pngdsp-init.o
YASM-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp.o YASM-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp.o
MMX-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp-init.o MMX-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp-init.o
MMX-OBJS-$(CONFIG_DWT) += x86/snowdsp_mmx.o MMX-OBJS-$(CONFIG_DWT) += x86/snowdsp_mmx.o


+ 0
- 102
libavcodec/x86/dsputil_mmx.c View File

@@ -582,28 +582,6 @@ static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
dst[i+0] += src[i+0]; dst[i+0] += src[i+0];
} }


static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
x86_reg i=0;
__asm__ volatile(
"jmp 2f \n\t"
"1: \n\t"
"movq (%2, %0), %%mm0 \n\t"
"movq 8(%2, %0), %%mm1 \n\t"
"paddb (%3, %0), %%mm0 \n\t"
"paddb 8(%3, %0), %%mm1 \n\t"
"movq %%mm0, (%1, %0) \n\t"
"movq %%mm1, 8(%1, %0) \n\t"
"add $16, %0 \n\t"
"2: \n\t"
"cmp %4, %0 \n\t"
" js 1b \n\t"
: "+r" (i)
: "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
);
for(; i<w; i++)
dst[i] = src1[i] + src2[i];
}

#if HAVE_7REGS #if HAVE_7REGS
static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) { static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) {
x86_reg w2 = -w; x86_reg w2 = -w;
@@ -879,80 +857,6 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w,
} }
} }


#define PAETH(cpu, abs3)\
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
{\
x86_reg i = -bpp;\
x86_reg end = w-3;\
__asm__ volatile(\
"pxor %%mm7, %%mm7 \n"\
"movd (%1,%0), %%mm0 \n"\
"movd (%2,%0), %%mm1 \n"\
"punpcklbw %%mm7, %%mm0 \n"\
"punpcklbw %%mm7, %%mm1 \n"\
"add %4, %0 \n"\
"1: \n"\
"movq %%mm1, %%mm2 \n"\
"movd (%2,%0), %%mm1 \n"\
"movq %%mm2, %%mm3 \n"\
"punpcklbw %%mm7, %%mm1 \n"\
"movq %%mm2, %%mm4 \n"\
"psubw %%mm1, %%mm3 \n"\
"psubw %%mm0, %%mm4 \n"\
"movq %%mm3, %%mm5 \n"\
"paddw %%mm4, %%mm5 \n"\
abs3\
"movq %%mm4, %%mm6 \n"\
"pminsw %%mm5, %%mm6 \n"\
"pcmpgtw %%mm6, %%mm3 \n"\
"pcmpgtw %%mm5, %%mm4 \n"\
"movq %%mm4, %%mm6 \n"\
"pand %%mm3, %%mm4 \n"\
"pandn %%mm3, %%mm6 \n"\
"pandn %%mm0, %%mm3 \n"\
"movd (%3,%0), %%mm0 \n"\
"pand %%mm1, %%mm6 \n"\
"pand %%mm4, %%mm2 \n"\
"punpcklbw %%mm7, %%mm0 \n"\
"movq %6, %%mm5 \n"\
"paddw %%mm6, %%mm0 \n"\
"paddw %%mm2, %%mm3 \n"\
"paddw %%mm3, %%mm0 \n"\
"pand %%mm5, %%mm0 \n"\
"movq %%mm0, %%mm3 \n"\
"packuswb %%mm3, %%mm3 \n"\
"movd %%mm3, (%1,%0) \n"\
"add %4, %0 \n"\
"cmp %5, %0 \n"\
"jle 1b \n"\
:"+r"(i)\
:"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
"m"(ff_pw_255)\
:"memory"\
);\
}

#define ABS3_MMX2\
"psubw %%mm5, %%mm7 \n"\
"pmaxsw %%mm7, %%mm5 \n"\
"pxor %%mm6, %%mm6 \n"\
"pxor %%mm7, %%mm7 \n"\
"psubw %%mm3, %%mm6 \n"\
"psubw %%mm4, %%mm7 \n"\
"pmaxsw %%mm6, %%mm3 \n"\
"pmaxsw %%mm7, %%mm4 \n"\
"pxor %%mm7, %%mm7 \n"

#define ABS3_SSSE3\
"pabsw %%mm3, %%mm3 \n"\
"pabsw %%mm4, %%mm4 \n"\
"pabsw %%mm5, %%mm5 \n"

PAETH(mmx2, ABS3_MMX2)
#if HAVE_SSSE3
PAETH(ssse3, ABS3_SSSE3)
#endif

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\ #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
"paddw " #m4 ", " #m3 " \n\t" /* x1 */\ "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
"movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\ "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\
@@ -2552,7 +2456,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
#endif #endif


c->add_bytes= add_bytes_mmx; c->add_bytes= add_bytes_mmx;
c->add_bytes_l2= add_bytes_l2_mmx;


if (!high_bit_depth) if (!high_bit_depth)
c->draw_edges = draw_edges_mmx; c->draw_edges = draw_edges_mmx;
@@ -2686,8 +2589,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
if (HAVE_AMD3DNOW && (mm_flags & AV_CPU_FLAG_3DNOW)) if (HAVE_AMD3DNOW && (mm_flags & AV_CPU_FLAG_3DNOW))
c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov; c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
#endif #endif

c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
} else if (HAVE_AMD3DNOW && (mm_flags & AV_CPU_FLAG_3DNOW)) { } else if (HAVE_AMD3DNOW && (mm_flags & AV_CPU_FLAG_3DNOW)) {
c->prefetch = prefetch_3dnow; c->prefetch = prefetch_3dnow;


@@ -2826,9 +2727,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
H264_QPEL_FUNCS_10(2, 0, ssse3_cache64) H264_QPEL_FUNCS_10(2, 0, ssse3_cache64)
H264_QPEL_FUNCS_10(3, 0, ssse3_cache64) H264_QPEL_FUNCS_10(3, 0, ssse3_cache64)
} }
#endif
c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
#if HAVE_YASM
if (!high_bit_depth && CONFIG_H264CHROMA) { if (!high_bit_depth && CONFIG_H264CHROMA) {
c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd; c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd;
c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd; c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd;


+ 135
- 0
libavcodec/x86/pngdsp-init.c View File

@@ -0,0 +1,135 @@
/*
* x86 PNG optimizations.
* Copyright (c) 2008 Loren Merrit <lorenm@u.washington.edu>
*
* 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/cpu.h"
#include "libavutil/x86_cpu.h"
#include "libavcodec/pngdsp.h"
#include "dsputil_mmx.h"

#define PAETH(cpu, abs3)\
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
{\
x86_reg i = -bpp;\
x86_reg end = w-3;\
__asm__ volatile(\
"pxor %%mm7, %%mm7 \n"\
"movd (%1,%0), %%mm0 \n"\
"movd (%2,%0), %%mm1 \n"\
"punpcklbw %%mm7, %%mm0 \n"\
"punpcklbw %%mm7, %%mm1 \n"\
"add %4, %0 \n"\
"1: \n"\
"movq %%mm1, %%mm2 \n"\
"movd (%2,%0), %%mm1 \n"\
"movq %%mm2, %%mm3 \n"\
"punpcklbw %%mm7, %%mm1 \n"\
"movq %%mm2, %%mm4 \n"\
"psubw %%mm1, %%mm3 \n"\
"psubw %%mm0, %%mm4 \n"\
"movq %%mm3, %%mm5 \n"\
"paddw %%mm4, %%mm5 \n"\
abs3\
"movq %%mm4, %%mm6 \n"\
"pminsw %%mm5, %%mm6 \n"\
"pcmpgtw %%mm6, %%mm3 \n"\
"pcmpgtw %%mm5, %%mm4 \n"\
"movq %%mm4, %%mm6 \n"\
"pand %%mm3, %%mm4 \n"\
"pandn %%mm3, %%mm6 \n"\
"pandn %%mm0, %%mm3 \n"\
"movd (%3,%0), %%mm0 \n"\
"pand %%mm1, %%mm6 \n"\
"pand %%mm4, %%mm2 \n"\
"punpcklbw %%mm7, %%mm0 \n"\
"movq %6, %%mm5 \n"\
"paddw %%mm6, %%mm0 \n"\
"paddw %%mm2, %%mm3 \n"\
"paddw %%mm3, %%mm0 \n"\
"pand %%mm5, %%mm0 \n"\
"movq %%mm0, %%mm3 \n"\
"packuswb %%mm3, %%mm3 \n"\
"movd %%mm3, (%1,%0) \n"\
"add %4, %0 \n"\
"cmp %5, %0 \n"\
"jle 1b \n"\
:"+r"(i)\
:"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
"m"(ff_pw_255)\
:"memory"\
);\
}

#define ABS3_MMX2\
"psubw %%mm5, %%mm7 \n"\
"pmaxsw %%mm7, %%mm5 \n"\
"pxor %%mm6, %%mm6 \n"\
"pxor %%mm7, %%mm7 \n"\
"psubw %%mm3, %%mm6 \n"\
"psubw %%mm4, %%mm7 \n"\
"pmaxsw %%mm6, %%mm3 \n"\
"pmaxsw %%mm7, %%mm4 \n"\
"pxor %%mm7, %%mm7 \n"

#define ABS3_SSSE3\
"pabsw %%mm3, %%mm3 \n"\
"pabsw %%mm4, %%mm4 \n"\
"pabsw %%mm5, %%mm5 \n"

PAETH(mmx2, ABS3_MMX2)
#if HAVE_SSSE3
PAETH(ssse3, ABS3_SSSE3)
#endif

static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w)
{
x86_reg i=0;

__asm__ volatile(
"jmp 2f \n\t"
"1: \n\t"
"movq (%2, %0), %%mm0 \n\t"
"movq 8(%2, %0), %%mm1 \n\t"
"paddb (%3, %0), %%mm0 \n\t"
"paddb 8(%3, %0), %%mm1 \n\t"
"movq %%mm0, (%1, %0) \n\t"
"movq %%mm1, 8(%1, %0) \n\t"
"add $16, %0 \n\t"
"2: \n\t"
"cmp %4, %0 \n\t"
" js 1b \n\t"
: "+r" (i)
: "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg) w - 15)
);
for (; i < w; i++)
dst[i] = src1[i] + src2[i];
}

void ff_pngdsp_init_x86(PNGDSPContext *dsp)
{
int flags = av_get_cpu_flags();

if (flags & AV_CPU_FLAG_MMX)
dsp->add_bytes_l2 = add_bytes_l2_mmx;
if (flags & AV_CPU_FLAG_MMX2)
dsp->add_paeth_prediction = add_png_paeth_prediction_mmx2;
if (flags & AV_CPU_FLAG_SSSE3)
dsp->add_paeth_prediction = add_png_paeth_prediction_ssse3;
}

Write Preview
Loading…
Cancel
Save