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

15/16bit in C speedup

Browse Source 
Originally committed as revision 2585 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
pull/126/head
Michael Niedermayer 24 years ago
parent 397c035e91
commit b18ea156c4
2 changed files with 100 additions and 60 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. 80
      postproc/swscale.c
  2. 80
      postproc/swscale_template.c

80
postproc/swscale.c
Unescape View File

@ -98,6 +98,13 @@ static uint16_t pix_buf_uv[2][2048*2];
// clipping helper table for C implementations:
static unsigned char clip_table[768];
static unsigned short clip_table16b[768];
static unsigned short clip_table16g[768];
static unsigned short clip_table16r[768];
static unsigned short clip_table15b[768];
static unsigned short clip_table15g[768];
static unsigned short clip_table15r[768];
// yuv->rgb conversion tables:
static int yuvtab_2568[256];
static int yuvtab_3343[256];
@ -702,9 +709,9 @@ FULL_YSCALEYUV2RGB
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
((uint16_t*)dest)[i] =
(clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<3)&0x07E0) |
((clip_table[(Y + yuvtab_3343[V]) >>13]<<8)&0xF800);
clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
clip_table16r[(Y + yuvtab_3343[V]) >>13];
}
}
else if(dstbpp==15)
@ -716,9 +723,9 @@ FULL_YSCALEYUV2RGB
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
((uint16_t*)dest)[i] =
(clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<2)&0x03E0) |
((clip_table[(Y + yuvtab_3343[V]) >>13]<<7)&0x7C00);
clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
clip_table15r[(Y + yuvtab_3343[V]) >>13];
}
}
#endif
@ -846,14 +853,14 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y1 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y1 + Cb) >>13] |
clip_table16g[(Y1 + Cg) >>13] |
clip_table16r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y2 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y2 + Cb) >>13] |
clip_table16g[(Y2 + Cg) >>13] |
clip_table16r[(Y2 + Cr) >>13];
}
}
else if(dstbpp==15)
@ -870,13 +877,14 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y1 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y1 + Cb) >>13] |
clip_table15g[(Y1 + Cg) >>13] |
clip_table15r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y2 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y2 + Cb) >>13] |
clip_table15g[(Y2 + Cg) >>13] |
clip_table15r[(Y2 + Cr) >>13];
}
}
#endif
@ -1075,14 +1083,14 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y1 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y1 + Cb) >>13] |
clip_table16g[(Y1 + Cg) >>13] |
clip_table16r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y2 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y2 + Cb) >>13] |
clip_table16g[(Y2 + Cg) >>13] |
clip_table16r[(Y2 + Cr) >>13];
}
}
else if(dstbpp==15)
@ -1099,13 +1107,14 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y1 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y1 + Cb) >>13] |
clip_table15g[(Y1 + Cg) >>13] |
clip_table15r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y2 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y2 + Cb) >>13] |
clip_table15g[(Y2 + Cg) >>13] |
clip_table15r[(Y2 + Cr) >>13];
}
}
#endif
@ -1666,4 +1675,15 @@ void SwScale_Init(){
yuvtab_40cf[i]=0x40cf*(i-128);
}
for(i=0; i<768; i++)
{
int v= clip_table[i];
clip_table16b[i]= v>>3;
clip_table16g[i]= (v<<3)&0x07E0;
clip_table16r[i]= (v<<8)&0xF800;
clip_table15b[i]= v>>3;
clip_table15g[i]= (v<<2)&0x03E0;
clip_table15r[i]= (v<<7)&0x7C00;
}
}

80
postproc/swscale_template.c
Unescape View File

@ -98,6 +98,13 @@ static uint16_t pix_buf_uv[2][2048*2];
// clipping helper table for C implementations:
static unsigned char clip_table[768];
static unsigned short clip_table16b[768];
static unsigned short clip_table16g[768];
static unsigned short clip_table16r[768];
static unsigned short clip_table15b[768];
static unsigned short clip_table15g[768];
static unsigned short clip_table15r[768];
// yuv->rgb conversion tables:
static int yuvtab_2568[256];
static int yuvtab_3343[256];
@ -702,9 +709,9 @@ FULL_YSCALEYUV2RGB
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
((uint16_t*)dest)[i] =
(clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<3)&0x07E0) |
((clip_table[(Y + yuvtab_3343[V]) >>13]<<8)&0xF800);
clip_table16b[(Y + yuvtab_40cf[U]) >>13] |
clip_table16g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
clip_table16r[(Y + yuvtab_3343[V]) >>13];
}
}
else if(dstbpp==15)
@ -716,9 +723,9 @@ FULL_YSCALEYUV2RGB
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
((uint16_t*)dest)[i] =
(clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<2)&0x03E0) |
((clip_table[(Y + yuvtab_3343[V]) >>13]<<7)&0x7C00);
clip_table15b[(Y + yuvtab_40cf[U]) >>13] |
clip_table15g[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13] |
clip_table15r[(Y + yuvtab_3343[V]) >>13];
}
}
#endif
@ -846,14 +853,14 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y1 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y1 + Cb) >>13] |
clip_table16g[(Y1 + Cg) >>13] |
clip_table16r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y2 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y2 + Cb) >>13] |
clip_table16g[(Y2 + Cg) >>13] |
clip_table16r[(Y2 + Cr) >>13];
}
}
else if(dstbpp==15)
@ -870,13 +877,14 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y1 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y1 + Cb) >>13] |
clip_table15g[(Y1 + Cg) >>13] |
clip_table15r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y2 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y2 + Cb) >>13] |
clip_table15g[(Y2 + Cg) >>13] |
clip_table15r[(Y2 + Cr) >>13];
}
}
#endif
@ -1075,14 +1083,14 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y1 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y1 + Cb) >>13] |
clip_table16g[(Y1 + Cg) >>13] |
clip_table16r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<3)&0x07E0) |
((clip_table[(Y2 + Cr) >>13]<<8)&0xF800);
clip_table16b[(Y2 + Cb) >>13] |
clip_table16g[(Y2 + Cg) >>13] |
clip_table16r[(Y2 + Cr) >>13];
}
}
else if(dstbpp==15)
@ -1099,13 +1107,14 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] =
(clip_table[(Y1 + Cb) >>13]>>3) |
((clip_table[(Y1 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y1 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y1 + Cb) >>13] |
clip_table15g[(Y1 + Cg) >>13] |
clip_table15r[(Y1 + Cr) >>13];
((uint16_t*)dest)[i+1] =
(clip_table[(Y2 + Cb) >>13]>>3) |
((clip_table[(Y2 + Cg) >>13]<<2)&0x03E0) |
((clip_table[(Y2 + Cr) >>13]<<7)&0x7C00);
clip_table15b[(Y2 + Cb) >>13] |
clip_table15g[(Y2 + Cg) >>13] |
clip_table15r[(Y2 + Cr) >>13];
}
}
#endif
@ -1666,4 +1675,15 @@ void SwScale_Init(){
yuvtab_40cf[i]=0x40cf*(i-128);
}
for(i=0; i<768; i++)
{
int v= clip_table[i];
clip_table16b[i]= v>>3;
clip_table16g[i]= (v<<3)&0x07E0;
clip_table16r[i]= (v<<8)&0xF800;
clip_table15b[i]= v>>3;
clip_table15g[i]= (v<<2)&0x03E0;
clip_table15r[i]= (v<<7)&0x7C00;
}
}

Write Preview
Loading…
Cancel
Save