virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

15/16 bit dithering in C (5% slower, can be disabled by comenting #define DITHER1XBPP out)

Browse Source 
x/yInc bugfix (bug was introduced during the global var removial)
more emms & no messing with ARCH_X86 (=workaround against not cleared mmx state somewhere)
sharpening filter (-ssf cs=<int> & -ssf ls=<int>)

Originally committed as revision 4298 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
This commit is contained in:
Michael Niedermayer 2002-01-21 22:11:09 +00:00
parent fe8054c0ff
commit 5521b193d6
3 changed files with 224 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

132
postproc/swscale.c
View File

@ -161,6 +161,8 @@ float sws_lum_gblur= 0.0;
float sws_chr_gblur= 0.0; float sws_chr_gblur= 0.0;
int sws_chr_vshift= 0; int sws_chr_vshift= 0;
int sws_chr_hshift= 0; int sws_chr_hshift= 0;
float sws_chr_sharpen= 0.0;
float sws_lum_sharpen= 0.0;
/* cpuCaps combined from cpudetect and whats actually compiled in /* cpuCaps combined from cpudetect and whats actually compiled in
(if there is no support for something compiled in it wont appear here) */ (if there is no support for something compiled in it wont appear here) */
@ -298,6 +300,28 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
else if(dstFormat==IMGFMT_BGR16) else if(dstFormat==IMGFMT_BGR16)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<13;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<13;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<13;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<13;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<(dstW>>1); i++){ for(i=0; i<(dstW>>1); i++){
int j; int j;
int Y1=0; int Y1=0;
@ -325,19 +349,41 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
Cr= clip_yuvtab_3343[V+ 256]; Cr= clip_yuvtab_3343[V+ 256];
((uint16_t*)dest)[2*i] = ((uint16_t*)dest)[2*i] =
clip_table16b[(Y1 + Cb) >>13] | clip_table16b[(Y1 + Cb + ditherb1) >>13] |
clip_table16g[(Y1 + Cg) >>13] | clip_table16g[(Y1 + Cg + ditherg1) >>13] |
clip_table16r[(Y1 + Cr) >>13]; clip_table16r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[2*i+1] = ((uint16_t*)dest)[2*i+1] =
clip_table16b[(Y2 + Cb) >>13] | clip_table16b[(Y2 + Cb + ditherb2) >>13] |
clip_table16g[(Y2 + Cg) >>13] | clip_table16g[(Y2 + Cg + ditherg2) >>13] |
clip_table16r[(Y2 + Cr) >>13]; clip_table16r[(Y2 + Cr + ditherr2) >>13];
} }
} }
else if(dstFormat==IMGFMT_BGR15) else if(dstFormat==IMGFMT_BGR15)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<14;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<14;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<14;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<14;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<(dstW>>1); i++){ for(i=0; i<(dstW>>1); i++){
int j; int j;
int Y1=0; int Y1=0;
@ -365,14 +411,14 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
Cr= clip_yuvtab_3343[V+ 256]; Cr= clip_yuvtab_3343[V+ 256];
((uint16_t*)dest)[2*i] = ((uint16_t*)dest)[2*i] =
clip_table15b[(Y1 + Cb) >>13] | clip_table15b[(Y1 + Cb + ditherb1) >>13] |
clip_table15g[(Y1 + Cg) >>13] | clip_table15g[(Y1 + Cg + ditherg1) >>13] |
clip_table15r[(Y1 + Cr) >>13]; clip_table15r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[2*i+1] = ((uint16_t*)dest)[2*i+1] =
clip_table15b[(Y2 + Cb) >>13] | clip_table15b[(Y2 + Cb + ditherb2) >>13] |
clip_table15g[(Y2 + Cg) >>13] | clip_table15g[(Y2 + Cg + ditherg2) >>13] |
clip_table15r[(Y2 + Cr) >>13]; clip_table15r[(Y2 + Cr + ditherr2) >>13];
} }
} }
} }
@ -402,13 +448,11 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
#undef HAVE_MMX #undef HAVE_MMX
#undef HAVE_MMX2 #undef HAVE_MMX2
#undef HAVE_3DNOW #undef HAVE_3DNOW
#undef ARCH_X86
#ifdef COMPILE_C #ifdef COMPILE_C
#undef HAVE_MMX #undef HAVE_MMX
#undef HAVE_MMX2 #undef HAVE_MMX2
#undef HAVE_3DNOW #undef HAVE_3DNOW
#undef ARCH_X86
#define RENAME(a) a ## _C #define RENAME(a) a ## _C
#include "swscale_template.c" #include "swscale_template.c"
#endif #endif
@ -431,7 +475,6 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
#define HAVE_MMX #define HAVE_MMX
#undef HAVE_MMX2 #undef HAVE_MMX2
#undef HAVE_3DNOW #undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _MMX #define RENAME(a) a ## _MMX
#include "swscale_template.c" #include "swscale_template.c"
#endif #endif
@ -442,7 +485,6 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
#define HAVE_MMX #define HAVE_MMX
#define HAVE_MMX2 #define HAVE_MMX2
#undef HAVE_3DNOW #undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _MMX2 #define RENAME(a) a ## _MMX2
#include "swscale_template.c" #include "swscale_template.c"
#endif #endif
@ -453,7 +495,6 @@ static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilt
#define HAVE_MMX #define HAVE_MMX
#undef HAVE_MMX2 #undef HAVE_MMX2
#define HAVE_3DNOW #define HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _3DNow #define RENAME(a) a ## _3DNow
#include "swscale_template.c" #include "swscale_template.c"
#endif #endif
@ -477,6 +518,10 @@ void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
if(firstTime) if(firstTime)
{ {
#ifdef ARCH_X86
if(gCpuCaps.hasMMX)
asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
#endif
flags= SWS_PRINT_INFO; flags= SWS_PRINT_INFO;
firstTime=0; firstTime=0;
@ -501,12 +546,41 @@ void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
src_filter.chrV= getIdentityVec(); src_filter.chrV= getIdentityVec();
} }
if(sws_chr_sharpen!=0.0){
SwsVector *g= getConstVec(-1.0, 3);
SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
g->coeff[1]=2.0;
addVec(id, g);
convVec(src_filter.chrH, id);
convVec(src_filter.chrV, id);
freeVec(g);
freeVec(id);
}
if(sws_lum_sharpen!=0.0){
SwsVector *g= getConstVec(-1.0, 3);
SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
g->coeff[1]=2.0;
addVec(id, g);
convVec(src_filter.lumH, id);
convVec(src_filter.lumV, id);
freeVec(g);
freeVec(id);
}
if(sws_chr_hshift) if(sws_chr_hshift)
shiftVec(src_filter.chrH, sws_chr_hshift); shiftVec(src_filter.chrH, sws_chr_hshift);
if(sws_chr_vshift) if(sws_chr_vshift)
shiftVec(src_filter.chrV, sws_chr_vshift); shiftVec(src_filter.chrV, sws_chr_vshift);
normalizeVec(src_filter.chrH, 1.0);
normalizeVec(src_filter.chrV, 1.0);
normalizeVec(src_filter.lumH, 1.0);
normalizeVec(src_filter.lumV, 1.0);
if(verbose > 1) printVec(src_filter.chrH);
if(verbose > 1) printVec(src_filter.lumH);
} }
switch(dstbpp) switch(dstbpp)
@ -551,7 +625,6 @@ static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *out
#endif #endif
*filterPos = (int16_t*)memalign(8, dstW*sizeof(int16_t)); *filterPos = (int16_t*)memalign(8, dstW*sizeof(int16_t));
if(ABS(xInc - 0x10000) <10) // unscaled if(ABS(xInc - 0x10000) <10) // unscaled
{ {
int i; int i;
@ -1002,8 +1075,8 @@ SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH,
c->srcH= srcH; c->srcH= srcH;
c->dstW= dstW; c->dstW= dstW;
c->dstH= dstH; c->dstH= dstH;
c->lumXInc= ((srcW<<16) + (1<<15))/dstW; c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
c->lumYInc= ((srcH<<16) + (1<<15))/dstH; c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
c->flags= flags; c->flags= flags;
c->dstFormat= dstFormat; c->dstFormat= dstFormat;
c->srcFormat= srcFormat; c->srcFormat= srcFormat;
@ -1120,7 +1193,9 @@ SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH,
if(flags&SWS_PRINT_INFO) if(flags&SWS_PRINT_INFO)
{ {
#ifdef DITHER1XBPP #ifdef DITHER1XBPP
char *dither= cpuCaps.hasMMX ? " dithered" : ""; char *dither= " dithered";
#else
char *dither= "";
#endif #endif
if(flags&SWS_FAST_BILINEAR) if(flags&SWS_FAST_BILINEAR)
fprintf(stderr, "\nSwScaler: FAST_BILINEAR scaler "); fprintf(stderr, "\nSwScaler: FAST_BILINEAR scaler ");
@ -1244,6 +1319,21 @@ SwsVector *getGaussianVec(double variance, double quality){
return vec; return vec;
} }
SwsVector *getConstVec(double c, int length){
int i;
double *coeff= memalign(sizeof(double), length*sizeof(double));
SwsVector *vec= malloc(sizeof(SwsVector));
vec->coeff= coeff;
vec->length= length;
for(i=0; i<length; i++)
coeff[i]= c;
return vec;
}
SwsVector *getIdentityVec(void){ SwsVector *getIdentityVec(void){
double *coeff= memalign(sizeof(double), sizeof(double)); double *coeff= memalign(sizeof(double), sizeof(double));
SwsVector *vec= malloc(sizeof(SwsVector)); SwsVector *vec= malloc(sizeof(SwsVector));

1
postproc/swscale.h
View File

@ -108,6 +108,7 @@ extern void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int
int srcSliceH, uint8_t* dst[], int dstStride[]); int srcSliceH, uint8_t* dst[], int dstStride[]);
SwsVector *getGaussianVec(double variance, double quality); SwsVector *getGaussianVec(double variance, double quality);
SwsVector *getConstVec(double c, int length);
SwsVector *getIdentityVec(void); SwsVector *getIdentityVec(void);
void scaleVec(SwsVector *a, double scalar); void scaleVec(SwsVector *a, double scalar);
void normalizeVec(SwsVector *a, double height); void normalizeVec(SwsVector *a, double height);

136
postproc/swscale_template.c
View File

@ -1163,6 +1163,28 @@ FULL_YSCALEYUV2RGB
else if(dstFormat==IMGFMT_BGR16) else if(dstFormat==IMGFMT_BGR16)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<13;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<13;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<13;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<13;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<dstW-1; i+=2){ for(i=0; i<dstW-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step: // vertical linear interpolation && yuv2rgb in a single step:
int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)]; int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
@ -1175,19 +1197,41 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V]; int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] = ((uint16_t*)dest)[i] =
clip_table16b[(Y1 + Cb) >>13] | clip_table16b[(Y1 + Cb + ditherb1) >>13] |
clip_table16g[(Y1 + Cg) >>13] | clip_table16g[(Y1 + Cg + ditherg1) >>13] |
clip_table16r[(Y1 + Cr) >>13]; clip_table16r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[i+1] = ((uint16_t*)dest)[i+1] =
clip_table16b[(Y2 + Cb) >>13] | clip_table16b[(Y2 + Cb + ditherb2) >>13] |
clip_table16g[(Y2 + Cg) >>13] | clip_table16g[(Y2 + Cg + ditherg2) >>13] |
clip_table16r[(Y2 + Cr) >>13]; clip_table16r[(Y2 + Cr + ditherr2) >>13];
} }
} }
else if(dstFormat==IMGFMT_BGR15) else if(dstFormat==IMGFMT_BGR15)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<14;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<14;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<14;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<14;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<dstW-1; i+=2){ for(i=0; i<dstW-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step: // vertical linear interpolation && yuv2rgb in a single step:
int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)]; int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
@ -1200,14 +1244,14 @@ FULL_YSCALEYUV2RGB
int Cr= yuvtab_3343[V]; int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] = ((uint16_t*)dest)[i] =
clip_table15b[(Y1 + Cb) >>13] | clip_table15b[(Y1 + Cb + ditherb1) >>13] |
clip_table15g[(Y1 + Cg) >>13] | clip_table15g[(Y1 + Cg + ditherg1) >>13] |
clip_table15r[(Y1 + Cr) >>13]; clip_table15r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[i+1] = ((uint16_t*)dest)[i+1] =
clip_table15b[(Y2 + Cb) >>13] | clip_table15b[(Y2 + Cb + ditherb2) >>13] |
clip_table15g[(Y2 + Cg) >>13] | clip_table15g[(Y2 + Cg + ditherg2) >>13] |
clip_table15r[(Y2 + Cr) >>13]; clip_table15r[(Y2 + Cr + ditherr2) >>13];
} }
} }
#endif #endif
@ -1397,6 +1441,28 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *uvbuf0, uint16_t *
else if(dstFormat==IMGFMT_BGR16) else if(dstFormat==IMGFMT_BGR16)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<13;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<13;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<13;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<13;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<dstW-1; i+=2){ for(i=0; i<dstW-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step: // vertical linear interpolation && yuv2rgb in a single step:
int Y1=yuvtab_2568[buf0[i]>>7]; int Y1=yuvtab_2568[buf0[i]>>7];
@ -1409,19 +1475,41 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *uvbuf0, uint16_t *
int Cr= yuvtab_3343[V]; int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] = ((uint16_t*)dest)[i] =
clip_table16b[(Y1 + Cb) >>13] | clip_table16b[(Y1 + Cb + ditherb1) >>13] |
clip_table16g[(Y1 + Cg) >>13] | clip_table16g[(Y1 + Cg + ditherg1) >>13] |
clip_table16r[(Y1 + Cr) >>13]; clip_table16r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[i+1] = ((uint16_t*)dest)[i+1] =
clip_table16b[(Y2 + Cb) >>13] | clip_table16b[(Y2 + Cb + ditherb2) >>13] |
clip_table16g[(Y2 + Cg) >>13] | clip_table16g[(Y2 + Cg + ditherg2) >>13] |
clip_table16r[(Y2 + Cr) >>13]; clip_table16r[(Y2 + Cr + ditherr2) >>13];
} }
} }
else if(dstFormat==IMGFMT_BGR15) else if(dstFormat==IMGFMT_BGR15)
{ {
int i; int i;
#ifdef DITHER1XBPP
static int ditherb1=1<<14;
static int ditherg1=1<<14;
static int ditherr1=2<<14;
static int ditherb2=3<<14;
static int ditherg2=3<<14;
static int ditherr2=0<<14;
ditherb1 ^= (1^2)<<14;
ditherg1 ^= (1^2)<<14;
ditherr1 ^= (1^2)<<14;
ditherb2 ^= (3^0)<<14;
ditherg2 ^= (3^0)<<14;
ditherr2 ^= (3^0)<<14;
#else
const int ditherb1=0;
const int ditherg1=0;
const int ditherr1=0;
const int ditherb2=0;
const int ditherg2=0;
const int ditherr2=0;
#endif
for(i=0; i<dstW-1; i+=2){ for(i=0; i<dstW-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step: // vertical linear interpolation && yuv2rgb in a single step:
int Y1=yuvtab_2568[buf0[i]>>7]; int Y1=yuvtab_2568[buf0[i]>>7];
@ -1434,14 +1522,14 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *uvbuf0, uint16_t *
int Cr= yuvtab_3343[V]; int Cr= yuvtab_3343[V];
((uint16_t*)dest)[i] = ((uint16_t*)dest)[i] =
clip_table15b[(Y1 + Cb) >>13] | clip_table15b[(Y1 + Cb + ditherb1) >>13] |
clip_table15g[(Y1 + Cg) >>13] | clip_table15g[(Y1 + Cg + ditherg1) >>13] |
clip_table15r[(Y1 + Cr) >>13]; clip_table15r[(Y1 + Cr + ditherr1) >>13];
((uint16_t*)dest)[i+1] = ((uint16_t*)dest)[i+1] =
clip_table15b[(Y2 + Cb) >>13] | clip_table15b[(Y2 + Cb + ditherb2) >>13] |
clip_table15g[(Y2 + Cg) >>13] | clip_table15g[(Y2 + Cg + ditherg2) >>13] |
clip_table15r[(Y2 + Cr) >>13]; clip_table15r[(Y2 + Cr + ditherr2) >>13];
} }
} }
#endif #endif