// the parts written by michael are under GNU GPL
#include <inttypes.h>
+#include <string.h>
#include "../config.h"
#include "swscale.h"
+#include "../mmx_defs.h"
+#undef MOVNTQ
+#undef PAVGB
//#undef HAVE_MMX2
//#undef HAVE_MMX
known BUGS with known cause (no bugreports please!, but patches are welcome :) )
horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
-Supported output formats BGR15 BGR16 BGR24 BGR32 (15,24 are untested)
+Supported output formats BGR15 BGR16 BGR24 BGR32
BGR15 & BGR16 MMX verions support dithering
Special versions: fast Y 1:1 scaling (no interpolation in y direction)
TODO
more intelligent missalignment avoidance for the horizontal scaler
+bicubic scaler
+dither in C
+change the distance of the u & v buffer
*/
#define ABS(a) ((a) > 0 ? (a) : (-(a)))
+#define MIN(a,b) ((a) > (b) ? (b) : (a))
+#define MAX(a,b) ((a) < (b) ? (b) : (a))
#ifdef HAVE_MMX2
#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
#ifdef HAVE_MMX
static uint64_t __attribute__((aligned(8))) yCoeff= 0x2568256825682568LL;
-static uint64_t __attribute__((aligned(8))) ubCoeff= 0x3343334333433343LL;
-static uint64_t __attribute__((aligned(8))) vrCoeff= 0x40cf40cf40cf40cfLL;
-static uint64_t __attribute__((aligned(8))) ugCoeff= 0xE5E2E5E2E5E2E5E2LL;
-static uint64_t __attribute__((aligned(8))) vgCoeff= 0xF36EF36EF36EF36ELL;
+static uint64_t __attribute__((aligned(8))) vrCoeff= 0x3343334333433343LL;
+static uint64_t __attribute__((aligned(8))) ubCoeff= 0x40cf40cf40cf40cfLL;
+static uint64_t __attribute__((aligned(8))) vgCoeff= 0xE5E2E5E2E5E2E5E2LL;
+static uint64_t __attribute__((aligned(8))) ugCoeff= 0xF36EF36EF36EF36ELL;
+static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
+static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
static uint64_t __attribute__((aligned(8))) w400= 0x0400040004000400LL;
static uint64_t __attribute__((aligned(8))) w80= 0x0080008000800080LL;
static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
// 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];
static int yuvtab_1a1e[256];
static int yuvtab_40cf[256];
-
+#ifdef HAVE_MMX2
static uint8_t funnyYCode[10000];
static uint8_t funnyUVCode[10000];
+#endif
+
+static int canMMX2BeUsed=0;
#define FULL_YSCALEYUV2RGB \
"pxor %%mm7, %%mm7 \n\t"\
"pmulhw ugCoeff, %%mm3 \n\t"\
"pmulhw vgCoeff, %%mm4 \n\t"\
/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
- "movq (%1, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
- "movq 8(%1, %%eax, 2), %%mm7 \n\t" /*buf0[eax]*/\
+ "movq (%0, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
+ "movq 8(%0, %%eax, 2), %%mm7 \n\t" /*buf0[eax]*/\
+ "psraw $4, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
+ "psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
+ "pmulhw ubCoeff, %%mm2 \n\t"\
+ "pmulhw vrCoeff, %%mm5 \n\t"\
+ "psubw w80, %%mm1 \n\t" /* 8(Y-16)*/\
+ "psubw w80, %%mm7 \n\t" /* 8(Y-16)*/\
+ "pmulhw yCoeff, %%mm1 \n\t"\
+ "pmulhw yCoeff, %%mm7 \n\t"\
+ /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
+ "paddw %%mm3, %%mm4 \n\t"\
+ "movq %%mm2, %%mm0 \n\t"\
+ "movq %%mm5, %%mm6 \n\t"\
+ "movq %%mm4, %%mm3 \n\t"\
+ "punpcklwd %%mm2, %%mm2 \n\t"\
+ "punpcklwd %%mm5, %%mm5 \n\t"\
+ "punpcklwd %%mm4, %%mm4 \n\t"\
+ "paddw %%mm1, %%mm2 \n\t"\
+ "paddw %%mm1, %%mm5 \n\t"\
+ "paddw %%mm1, %%mm4 \n\t"\
+ "punpckhwd %%mm0, %%mm0 \n\t"\
+ "punpckhwd %%mm6, %%mm6 \n\t"\
+ "punpckhwd %%mm3, %%mm3 \n\t"\
+ "paddw %%mm7, %%mm0 \n\t"\
+ "paddw %%mm7, %%mm6 \n\t"\
+ "paddw %%mm7, %%mm3 \n\t"\
+ /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
+ "packuswb %%mm0, %%mm2 \n\t"\
+ "packuswb %%mm6, %%mm5 \n\t"\
+ "packuswb %%mm3, %%mm4 \n\t"\
+ "pxor %%mm7, %%mm7 \n\t"
+
+// do vertical chrominance interpolation
+#define YSCALEYUV2RGB1b \
+ "xorl %%eax, %%eax \n\t"\
+ "1: \n\t"\
+ "movq (%2, %%eax), %%mm2 \n\t" /* uvbuf0[eax]*/\
+ "movq (%3, %%eax), %%mm3 \n\t" /* uvbuf1[eax]*/\
+ "movq 4096(%2, %%eax), %%mm5 \n\t" /* uvbuf0[eax+2048]*/\
+ "movq 4096(%3, %%eax), %%mm4 \n\t" /* uvbuf1[eax+2048]*/\
+ "paddw %%mm2, %%mm3 \n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
+ "paddw %%mm5, %%mm4 \n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
+ "psrlw $5, %%mm3 \n\t"\
+ "psrlw $5, %%mm4 \n\t"\
+ "psubw w400, %%mm3 \n\t" /* (U-128)8*/\
+ "psubw w400, %%mm4 \n\t" /* (V-128)8*/\
+ "movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
+ "movq %%mm4, %%mm5 \n\t" /* (V-128)8*/\
+ "pmulhw ugCoeff, %%mm3 \n\t"\
+ "pmulhw vgCoeff, %%mm4 \n\t"\
+ /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
+ "movq (%0, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
+ "movq 8(%0, %%eax, 2), %%mm7 \n\t" /*buf0[eax]*/\
"psraw $4, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"pmulhw ubCoeff, %%mm2 \n\t"\
" jb 1b \n\t"
#define WRITEBGR16 \
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm4, %%mm3 \n\t" /* G */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
+ "pand bF8, %%mm2 \n\t" /* B */\
+ "pand bFC, %%mm4 \n\t" /* G */\
+ "pand bF8, %%mm5 \n\t" /* R */\
+ "psrlq $3, %%mm2 \n\t"\
\
- "punpcklbw %%mm7, %%mm3 \n\t" /* 0G0G0G0G */\
- "punpcklbw %%mm7, %%mm2 \n\t" /* 0B0B0B0B */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R */\
+ "movq %%mm2, %%mm1 \n\t"\
+ "movq %%mm4, %%mm3 \n\t"\
\
- "psrlw $3, %%mm2 \n\t"\
- "psllw $3, %%mm3 \n\t"\
- "psllw $8, %%mm5 \n\t"\
+ "punpcklbw %%mm7, %%mm3 \n\t"\
+ "punpcklbw %%mm5, %%mm2 \n\t"\
+ "punpckhbw %%mm7, %%mm4 \n\t"\
+ "punpckhbw %%mm5, %%mm1 \n\t"\
\
- "pand g16Mask, %%mm3 \n\t"\
- "pand r16Mask, %%mm5 \n\t"\
+ "psllq $3, %%mm3 \n\t"\
+ "psllq $3, %%mm4 \n\t"\
\
"por %%mm3, %%mm2 \n\t"\
- "por %%mm5, %%mm2 \n\t"\
-\
- "punpckhbw %%mm7, %%mm4 \n\t" /* 0G0G0G0G */\
- "punpckhbw %%mm7, %%mm1 \n\t" /* 0B0B0B0B */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm1 \n\t"\
- "psllw $3, %%mm4 \n\t"\
- "psllw $8, %%mm6 \n\t"\
-\
- "pand g16Mask, %%mm4 \n\t"\
- "pand r16Mask, %%mm6 \n\t"\
-\
"por %%mm4, %%mm1 \n\t"\
- "por %%mm6, %%mm1 \n\t"\
\
MOVNTQ(%%mm2, (%4, %%eax, 2))\
MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
" jb 1b \n\t"
#define WRITEBGR15 \
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm4, %%mm3 \n\t" /* G */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
+ "pand bF8, %%mm2 \n\t" /* B */\
+ "pand bF8, %%mm4 \n\t" /* G */\
+ "pand bF8, %%mm5 \n\t" /* R */\
+ "psrlq $3, %%mm2 \n\t"\
+ "psrlq $1, %%mm5 \n\t"\
\
- "punpcklbw %%mm7, %%mm3 \n\t" /* 0G0G0G0G */\
- "punpcklbw %%mm7, %%mm2 \n\t" /* 0B0B0B0B */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R */\
+ "movq %%mm2, %%mm1 \n\t"\
+ "movq %%mm4, %%mm3 \n\t"\
\
- "psrlw $3, %%mm2 \n\t"\
- "psllw $2, %%mm3 \n\t"\
- "psllw $7, %%mm5 \n\t"\
+ "punpcklbw %%mm7, %%mm3 \n\t"\
+ "punpcklbw %%mm5, %%mm2 \n\t"\
+ "punpckhbw %%mm7, %%mm4 \n\t"\
+ "punpckhbw %%mm5, %%mm1 \n\t"\
\
- "pand g15Mask, %%mm3 \n\t"\
- "pand r15Mask, %%mm5 \n\t"\
+ "psllq $2, %%mm3 \n\t"\
+ "psllq $2, %%mm4 \n\t"\
\
"por %%mm3, %%mm2 \n\t"\
- "por %%mm5, %%mm2 \n\t"\
-\
- "punpckhbw %%mm7, %%mm4 \n\t" /* 0G0G0G0G */\
- "punpckhbw %%mm7, %%mm1 \n\t" /* 0B0B0B0B */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm1 \n\t"\
- "psllw $2, %%mm4 \n\t"\
- "psllw $7, %%mm6 \n\t"\
-\
- "pand g15Mask, %%mm4 \n\t"\
- "pand r15Mask, %%mm6 \n\t"\
-\
"por %%mm4, %%mm1 \n\t"\
- "por %%mm6, %%mm1 \n\t"\
\
MOVNTQ(%%mm2, (%4, %%eax, 2))\
MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
"addl $8, %%eax \n\t"\
"cmpl %5, %%eax \n\t"\
" jb 1b \n\t"
+
// FIXME find a faster way to shuffle it to BGR24
#define WRITEBGR24 \
/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
"punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R 2 */\
"movq %%mm2, %%mm0 \n\t" /* GBGBGBGB 0 */\
"movq %%mm1, %%mm3 \n\t" /* GBGBGBGB 2 */\
- "punpcklbw %%mm5, %%mm0 \n\t" /* 0RGB0RGB 0 */\
- "punpckhbw %%mm5, %%mm2 \n\t" /* 0RGB0RGB 1 */\
- "punpcklbw %%mm6, %%mm1 \n\t" /* 0RGB0RGB 2 */\
- "punpckhbw %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */\
+ "punpcklwd %%mm5, %%mm0 \n\t" /* 0RGB0RGB 0 */\
+ "punpckhwd %%mm5, %%mm2 \n\t" /* 0RGB0RGB 1 */\
+ "punpcklwd %%mm6, %%mm1 \n\t" /* 0RGB0RGB 2 */\
+ "punpckhwd %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */\
\
"movq %%mm0, %%mm4 \n\t" /* 0RGB0RGB 0 */\
"psrlq $8, %%mm0 \n\t" /* 00RGB0RG 0 */\
"psllq $16, %%mm3 \n\t" /* RGBRGB00 3 */\
"por %%mm4, %%mm3 \n\t" /* RGBRGBRG 2.5 */\
\
- "leal (%%eax, %%eax, 2), %%ebx \n\t"\
- MOVNTQ(%%mm0, (%4, %%ebx))\
- MOVNTQ(%%mm2, 8(%4, %%ebx))\
- MOVNTQ(%%mm3, 16(%4, %%ebx))\
+ MOVNTQ(%%mm0, (%%ebx))\
+ MOVNTQ(%%mm2, 8(%%ebx))\
+ MOVNTQ(%%mm3, 16(%%ebx))\
+ "addl $24, %%ebx \n\t"\
\
"addl $8, %%eax \n\t"\
"cmpl %5, %%eax \n\t"\
" jb 1b \n\t"
+#ifdef HAVE_MMX
+void in_asm_used_var_warning_killer()
+{
+ int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
+ bm00001111+bm00000111+bm11111000+b16Dither+b16Dither1+b16Dither2+g16Dither+g16Dither1+
+ g16Dither2+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+temp0+asm_yalpha1+ asm_uvalpha1;
+ if(i) i=0;
+}
+#endif
+
+static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
+ uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
+{
+ int yalpha1=yalpha^4095;
+ int uvalpha1=uvalpha^4095;
+ int i;
+
+ asm volatile ("\n\t"::: "memory");
+
+ for(i=0;i<dstw;i++)
+ {
+ ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
+ }
+
+ if(uvalpha != -1)
+ {
+ for(i=0; i<(dstw>>1); i++)
+ {
+ ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
+ ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
+ }
+ }
+}
/**
* vertical scale YV12 to RGB
{
int yalpha1=yalpha^4095;
int uvalpha1=uvalpha^4095;
- int i;
if(fullUVIpol)
{
);
}
#else
+ asm volatile ("\n\t"::: "memory");
+
if(dstbpp==32 || dstbpp==24)
{
+ int i;
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
+ dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
+ dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
+ dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
dest+=dstbpp>>3;
}
}
else if(dstbpp==16)
{
+ int i;
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
- dest+=2;
+ ((uint16_t*)dest)[i] =
+ 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)
{
+ int i;
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
- dest+=2;
+ ((uint16_t*)dest)[i] =
+ 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
else if(dstbpp==24)
{
asm volatile(
+ "movl %4, %%ebx \n\t"
YSCALEYUV2RGB
WRITEBGR24
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax", "%ebx"
);
);
}
#else
-//FIXME unroll C loop and dont recalculate UV
- if(dstbpp==32 || dstbpp==24)
+ asm volatile ("\n\t"::: "memory");
+
+ if(dstbpp==32)
{
- for(i=0;i<dstw;i++){
+ int i;
+ for(i=0; i<dstw-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
- dest+=dstbpp>>3;
+ int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
+ int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
+ dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
+ dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
+ dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
+ dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
+ }
+ }
+ if(dstbpp==24)
+ {
+ int i;
+ for(i=0; i<dstw-1; i+=2){
+ // vertical linear interpolation && yuv2rgb in a single step:
+ int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
+ int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ dest[0]=clip_table[((Y1 + Cb) >>13)];
+ dest[1]=clip_table[((Y1 + Cg) >>13)];
+ dest[2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[3]=clip_table[((Y2 + Cb) >>13)];
+ dest[4]=clip_table[((Y2 + Cg) >>13)];
+ dest[5]=clip_table[((Y2 + Cr) >>13)];
+ dest+=6;
}
}
else if(dstbpp==16)
{
- for(i=0;i<dstw;i++){
+ int i;
+ for(i=0; i<dstw-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
- dest+=2;
+ int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
+ int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ ((uint16_t*)dest)[i] =
+ clip_table16b[(Y1 + Cb) >>13] |
+ clip_table16g[(Y1 + Cg) >>13] |
+ clip_table16r[(Y1 + Cr) >>13];
+
+ ((uint16_t*)dest)[i+1] =
+ clip_table16b[(Y2 + Cb) >>13] |
+ clip_table16g[(Y2 + Cg) >>13] |
+ clip_table16r[(Y2 + Cr) >>13];
}
}
else if(dstbpp==15)
{
- for(i=0;i<dstw;i++){
+ int i;
+ for(i=0; i<dstw-1; i+=2){
// vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
- dest+=2;
+ int Y1=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
+ int Y2=yuvtab_2568[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19)];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ ((uint16_t*)dest)[i] =
+ clip_table15b[(Y1 + Cb) >>13] |
+ clip_table15g[(Y1 + Cg) >>13] |
+ clip_table15r[(Y1 + Cr) >>13];
+
+ ((uint16_t*)dest)[i+1] =
+ clip_table15b[(Y2 + Cb) >>13] |
+ clip_table15g[(Y2 + Cg) >>13] |
+ clip_table15r[(Y2 + Cr) >>13];
}
}
#endif
static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
{
- int yalpha1=yalpha^4095;
int uvalpha1=uvalpha^4095;
- int i;
+#ifdef HAVE_MMX
+ int yalpha1=yalpha^4095;
+#endif
+
if(fullUVIpol || allwaysIpol)
{
yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
return;
}
+ if( yalpha > 2048 ) buf0 = buf1;
+
#ifdef HAVE_MMX
+ if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
+ {
if(dstbpp == 32)
{
asm volatile(
else if(dstbpp==24)
{
asm volatile(
+ "movl %4, %%ebx \n\t"
YSCALEYUV2RGB1
WRITEBGR24
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax", "%ebx"
);
: "%eax"
);
}
-#else
-//FIXME unroll C loop and dont recalculate UV
- if(dstbpp==32 || dstbpp==24)
+ }
+ else
+ {
+ if(dstbpp == 32)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
- dest+=dstbpp>>3;
- }
+ asm volatile(
+ YSCALEYUV2RGB1b
+ WRITEBGR32
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
+ "m" (yalpha1), "m" (uvalpha1)
+ : "%eax"
+ );
}
- else if(dstbpp==16)
+ else if(dstbpp==24)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
- dest+=2;
- }
+ asm volatile(
+ "movl %4, %%ebx \n\t"
+ YSCALEYUV2RGB1b
+ WRITEBGR24
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
+ "m" (yalpha1), "m" (uvalpha1)
+ : "%eax", "%ebx"
+ );
}
else if(dstbpp==15)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
- (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
- (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
- dest+=2;
- }
+ asm volatile(
+ YSCALEYUV2RGB1b
+ /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
+#ifdef DITHER1XBPP
+ "paddusb b16Dither, %%mm2 \n\t"
+ "paddusb b16Dither, %%mm4 \n\t"
+ "paddusb b16Dither, %%mm5 \n\t"
+#endif
+ WRITEBGR15
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
+ "m" (yalpha1), "m" (uvalpha1)
+ : "%eax"
+ );
}
+ else if(dstbpp==16)
+ {
+ asm volatile(
+ YSCALEYUV2RGB1b
+ /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
+#ifdef DITHER1XBPP
+ "paddusb g16Dither, %%mm2 \n\t"
+ "paddusb b16Dither, %%mm4 \n\t"
+ "paddusb b16Dither, %%mm5 \n\t"
#endif
-}
-
-
-
-
-// *** bilinear scaling and yuv->rgb conversion of yv12 slices:
-// *** Note: it's called multiple times while decoding a frame, first time y==0
-// *** Designed to upscale, but may work for downscale too.
-// s_xinc = (src_width << 16) / dst_width
-// s_yinc = (src_height << 16) / dst_height
-void SwScale_YV12slice_brg24(unsigned char* srcptr[],int stride[], int y, int h,
- unsigned char* dstptr, int dststride, int dstw, int dstbpp,
- unsigned int s_xinc,unsigned int s_yinc){
-
-// scaling factors:
-//static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
-//static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
-unsigned int s_xinc2;
-
-static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
-static int s_ypos;
+ WRITEBGR16
+ :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
+ "m" (yalpha1), "m" (uvalpha1)
+ : "%eax"
+ );
+ }
+ }
+#else
+//FIXME write 2 versions (for even & odd lines)
+ asm volatile ("\n\t"::: "memory");
-// last horzontally interpolated lines, used to avoid unnecessary calculations
-static int s_last_ypos;
-static int s_last_y1pos;
+ if(dstbpp==32)
+ {
+ int i;
+ for(i=0; i<dstw-1; i+=2){
+ // vertical linear interpolation && yuv2rgb in a single step:
+ int Y1=yuvtab_2568[buf0[i]>>7];
+ int Y2=yuvtab_2568[buf0[i+1]>>7];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ dest[4*i+0]=clip_table[((Y1 + Cb) >>13)];
+ dest[4*i+1]=clip_table[((Y1 + Cg) >>13)];
+ dest[4*i+2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[4*i+4]=clip_table[((Y2 + Cb) >>13)];
+ dest[4*i+5]=clip_table[((Y2 + Cg) >>13)];
+ dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
+ }
+ }
+ if(dstbpp==24)
+ {
+ int i;
+ for(i=0; i<dstw-1; i+=2){
+ // vertical linear interpolation && yuv2rgb in a single step:
+ int Y1=yuvtab_2568[buf0[i]>>7];
+ int Y2=yuvtab_2568[buf0[i+1]>>7];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ dest[0]=clip_table[((Y1 + Cb) >>13)];
+ dest[1]=clip_table[((Y1 + Cg) >>13)];
+ dest[2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[3]=clip_table[((Y2 + Cb) >>13)];
+ dest[4]=clip_table[((Y2 + Cg) >>13)];
+ dest[5]=clip_table[((Y2 + Cr) >>13)];
+ dest+=6;
+ }
+ }
+ else if(dstbpp==16)
+ {
+ int i;
+ for(i=0; i<dstw-1; i+=2){
+ // vertical linear interpolation && yuv2rgb in a single step:
+ int Y1=yuvtab_2568[buf0[i]>>7];
+ int Y2=yuvtab_2568[buf0[i+1]>>7];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ ((uint16_t*)dest)[i] =
+ clip_table16b[(Y1 + Cb) >>13] |
+ clip_table16g[(Y1 + Cg) >>13] |
+ clip_table16r[(Y1 + Cr) >>13];
+
+ ((uint16_t*)dest)[i+1] =
+ clip_table16b[(Y2 + Cb) >>13] |
+ clip_table16g[(Y2 + Cg) >>13] |
+ clip_table16r[(Y2 + Cr) >>13];
+ }
+ }
+ else if(dstbpp==15)
+ {
+ int i;
+ for(i=0; i<dstw-1; i+=2){
+ // vertical linear interpolation && yuv2rgb in a single step:
+ int Y1=yuvtab_2568[buf0[i]>>7];
+ int Y2=yuvtab_2568[buf0[i+1]>>7];
+ int U=((uvbuf0[i>>1]*uvalpha1+uvbuf1[i>>1]*uvalpha)>>19);
+ int V=((uvbuf0[(i>>1)+2048]*uvalpha1+uvbuf1[(i>>1)+2048]*uvalpha)>>19);
+
+ int Cb= yuvtab_40cf[U];
+ int Cg= yuvtab_1a1e[V] + yuvtab_0c92[U];
+ int Cr= yuvtab_3343[V];
+
+ ((uint16_t*)dest)[i] =
+ clip_table15b[(Y1 + Cb) >>13] |
+ clip_table15g[(Y1 + Cg) >>13] |
+ clip_table15r[(Y1 + Cr) >>13];
+
+ ((uint16_t*)dest)[i+1] =
+ clip_table15b[(Y2 + Cb) >>13] |
+ clip_table15g[(Y2 + Cg) >>13] |
+ clip_table15r[(Y2 + Cr) >>13];
+ }
+ }
+#endif
+}
-static int static_dstw;
+static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
+{
+ // *** horizontal scale Y line to temp buffer
+#ifdef ARCH_X86
#ifdef HAVE_MMX2
-// used to detect a horizontal size change
-static int old_dstw= -1;
-static int old_s_xinc= -1;
-#endif
+ int i;
+ if(canMMX2BeUsed)
+ {
+ asm volatile(
+ "pxor %%mm7, %%mm7 \n\t"
+ "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
+ "movd %5, %%mm6 \n\t" // xInc&0xFFFF
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "movq %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t"
+ "paddw %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t"
+ "paddw %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=xInc&0xFF
+ "movq %%mm2, temp0 \n\t"
+ "movd %4, %%mm6 \n\t" //(xInc*4)&0xFFFF
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "xorl %%eax, %%eax \n\t" // i
+ "movl %0, %%esi \n\t" // src
+ "movl %1, %%edi \n\t" // buf1
+ "movl %3, %%edx \n\t" // (xInc*4)>>16
+ "xorl %%ecx, %%ecx \n\t"
+ "xorl %%ebx, %%ebx \n\t"
+ "movw %4, %%bx \n\t" // (xInc*4)&0xFFFF
-int canMMX2BeUsed=0;
-int srcWidth= (dstw*s_xinc + 0x8000)>>16;
-int dstUVw= fullUVIpol ? dstw : dstw/2;
+#define FUNNY_Y_CODE \
+ PREFETCH" 1024(%%esi) \n\t"\
+ PREFETCH" 1056(%%esi) \n\t"\
+ PREFETCH" 1088(%%esi) \n\t"\
+ "call funnyYCode \n\t"\
+ "movq temp0, %%mm2 \n\t"\
+ "xorl %%ecx, %%ecx \n\t"
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
+FUNNY_Y_CODE
-#ifdef HAVE_MMX2
-canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
+ :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
+ "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
+ : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
+ );
+ for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
+ }
+ else
+ {
#endif
-
-// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
-// n-2 is the last chrominance sample available
-// FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
+ //NO MMX just normal asm ...
+ asm volatile(
+ "xorl %%eax, %%eax \n\t" // i
+ "xorl %%ebx, %%ebx \n\t" // xx
+ "xorl %%ecx, %%ecx \n\t" // 2*xalpha
+ "1: \n\t"
+ "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
+ "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
+ "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
+ "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
+ "shll $16, %%edi \n\t"
+ "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
+ "movl %1, %%edi \n\t"
+ "shrl $9, %%esi \n\t"
+ "movw %%si, (%%edi, %%eax, 2) \n\t"
+ "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
+ "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
+
+ "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
+ "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
+ "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
+ "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
+ "shll $16, %%edi \n\t"
+ "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
+ "movl %1, %%edi \n\t"
+ "shrl $9, %%esi \n\t"
+ "movw %%si, 2(%%edi, %%eax, 2) \n\t"
+ "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
+ "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
+
+
+ "addl $2, %%eax \n\t"
+ "cmpl %2, %%eax \n\t"
+ " jb 1b \n\t"
+
+
+ :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
+ : "%eax", "%ebx", "%ecx", "%edi", "%esi"
+ );
+#ifdef HAVE_MMX2
+ } //if MMX2 cant be used
+#endif
+#else
+ int i;
+ unsigned int xpos=0;
+ for(i=0;i<dstWidth;i++)
+ {
+ register unsigned int xx=xpos>>16;
+ register unsigned int xalpha=(xpos&0xFFFF)>>9;
+ dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
+ xpos+=xInc;
+ }
+#endif
+}
+
+inline static void hcscale(uint16_t *dst, int dstWidth,
+ uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
+{
+#ifdef ARCH_X86
+#ifdef HAVE_MMX2
+ int i;
+ if(canMMX2BeUsed)
+ {
+ asm volatile(
+ "pxor %%mm7, %%mm7 \n\t"
+ "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
+ "movd %5, %%mm6 \n\t" // xInc&0xFFFF
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "movq %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t"
+ "paddw %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t"
+ "paddw %%mm6, %%mm2 \n\t"
+ "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=xInc&0xFFFF
+ "movq %%mm2, temp0 \n\t"
+ "movd %4, %%mm6 \n\t" //(xInc*4)&0xFFFF
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "punpcklwd %%mm6, %%mm6 \n\t"
+ "xorl %%eax, %%eax \n\t" // i
+ "movl %0, %%esi \n\t" // src
+ "movl %1, %%edi \n\t" // buf1
+ "movl %3, %%edx \n\t" // (xInc*4)>>16
+ "xorl %%ecx, %%ecx \n\t"
+ "xorl %%ebx, %%ebx \n\t"
+ "movw %4, %%bx \n\t" // (xInc*4)&0xFFFF
+
+#define FUNNYUVCODE \
+ PREFETCH" 1024(%%esi) \n\t"\
+ PREFETCH" 1056(%%esi) \n\t"\
+ PREFETCH" 1088(%%esi) \n\t"\
+ "call funnyUVCode \n\t"\
+ "movq temp0, %%mm2 \n\t"\
+ "xorl %%ecx, %%ecx \n\t"
+
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+ "xorl %%eax, %%eax \n\t" // i
+ "movl %6, %%esi \n\t" // src
+ "movl %1, %%edi \n\t" // buf1
+ "addl $4096, %%edi \n\t"
+
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+FUNNYUVCODE
+
+ :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
+ "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
+ : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
+ );
+ for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
+ {
+ dst[i] = src1[srcWidth/2-1]*128;
+ dst[i+2048] = src2[srcWidth/2-1]*128;
+ }
+ }
+ else
+ {
+#endif
+ asm volatile(
+ "xorl %%eax, %%eax \n\t" // i
+ "xorl %%ebx, %%ebx \n\t" // xx
+ "xorl %%ecx, %%ecx \n\t" // 2*xalpha
+ "1: \n\t"
+ "movl %0, %%esi \n\t"
+ "movzbl (%%esi, %%ebx), %%edi \n\t" //src[xx]
+ "movzbl 1(%%esi, %%ebx), %%esi \n\t" //src[xx+1]
+ "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
+ "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
+ "shll $16, %%edi \n\t"
+ "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
+ "movl %1, %%edi \n\t"
+ "shrl $9, %%esi \n\t"
+ "movw %%si, (%%edi, %%eax, 2) \n\t"
+
+ "movzbl (%5, %%ebx), %%edi \n\t" //src[xx]
+ "movzbl 1(%5, %%ebx), %%esi \n\t" //src[xx+1]
+ "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
+ "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
+ "shll $16, %%edi \n\t"
+ "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
+ "movl %1, %%edi \n\t"
+ "shrl $9, %%esi \n\t"
+ "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
+
+ "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
+ "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
+ "addl $1, %%eax \n\t"
+ "cmpl %2, %%eax \n\t"
+ " jb 1b \n\t"
+
+ :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
+ "r" (src2)
+ : "%eax", "%ebx", "%ecx", "%edi", "%esi"
+ );
+#ifdef HAVE_MMX2
+ } //if MMX2 cant be used
+#endif
+#else
+ int i;
+ unsigned int xpos=0;
+ for(i=0;i<dstWidth;i++)
+ {
+ register unsigned int xx=xpos>>16;
+ register unsigned int xalpha=(xpos&0xFFFF)>>9;
+ dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
+ dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
+/* slower
+ dst[i]= (src1[xx]<<7) + (src1[xx+1] - src1[xx])*xalpha;
+ dst[i+2048]=(src2[xx]<<7) + (src2[xx+1] - src2[xx])*xalpha;
+*/
+ xpos+=xInc;
+ }
+#endif
+}
+
+
+// *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
+// *** Note: it's called multiple times while decoding a frame, first time y==0
+// *** Designed to upscale, but may work for downscale too.
+// s_xinc = (src_width << 16) / dst_width
+// s_yinc = (src_height << 16) / dst_height
+void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
+ uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
+ unsigned int s_xinc,unsigned int s_yinc){
+
+// scaling factors:
+//static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
+//static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
+
+unsigned int s_xinc2;
+
+static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
+static int s_ypos;
+
+// last horzontally interpolated lines, used to avoid unnecessary calculations
+static int s_last_ypos;
+static int s_last_y1pos;
+
+#ifdef HAVE_MMX2
+// used to detect a horizontal size change
+static int old_dstw= -1;
+static int old_s_xinc= -1;
+#endif
+
+int srcWidth;
+int dstUVw;
+int i;
+
+if(((dstw + 7)&(~7)) >= dststride) dstw&= ~7;
+
+srcWidth= (dstw*s_xinc + 0x8000)>>16;
+dstUVw= fullUVIpol ? dstw : dstw/2;
+
+#ifdef HAVE_MMX2
+canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
+#endif
+
+// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
+// n-2 is the last chrominance sample available
+// FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
// would be like the vertical one, but that would require some special code for the
// first and last pixel
if(canMMX2BeUsed) s_xinc+= 20;
else s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
-if(fullUVIpol) s_xinc2= s_xinc>>1;
-else s_xinc2= s_xinc;
+if(fullUVIpol && !(dstbpp==12)) s_xinc2= s_xinc>>1;
+else s_xinc2= s_xinc;
// force calculation of the horizontal interpolation of the first line
- s_last_ypos=-99;
- s_last_y1pos=-99;
if(y==0){
- s_srcypos=-0x8000;
- s_ypos=0;
+// printf("dstw %d, srcw %d, mmx2 %d\n", dstw, srcWidth, canMMX2BeUsed);
+ s_last_ypos=-99;
+ s_last_y1pos=-99;
+ s_srcypos= s_yinc/2 - 0x8000;
+ s_ypos=0;
+
+ // clean the buffers so that no green stuff is drawen if the width is not sane (%8=0)
+ for(i=dstw-2; i<dstw+20; i++)
+ {
+ pix_buf_uv[0][i] = pix_buf_uv[1][i]
+ = pix_buf_uv[0][2048+i] = pix_buf_uv[1][2048+i] = 128*128;
+ pix_buf_uv[0][i/2] = pix_buf_uv[1][i/2]
+ = pix_buf_uv[0][2048+i/2] = pix_buf_uv[1][2048+i/2] = 128*128;
+ pix_buf_y[0][i]= pix_buf_y[1][i]= 0;
+ }
+
#ifdef HAVE_MMX2
// cant downscale !!!
if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
int imm8OfPShufW2;
int fragmentLength;
- int xpos, xx, xalpha, i;
+ int xpos, i;
old_s_xinc= s_xinc;
old_dstw= dstw;
- static_dstw= dstw;
-
// create an optimized horizontal scaling routine
//code fragment
#endif // HAVE_MMX2
} // reset counters
-
while(1){
- unsigned char *dest=dstptr+dststride*s_ypos;
+ unsigned char *dest =dstptr[0]+dststride*s_ypos;
+ unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
+ unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
+
int y0=(s_srcypos + 0xFFFF)>>16; // first luminance source line number below the dst line
// points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
- int srcuvpos= s_srcypos + s_yinc/2 - 0x8000;
+ int srcuvpos= dstbpp==12 ? s_srcypos + s_yinc/2 - 0x8000 :
+ s_srcypos - 0x8000;
int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
int yalpha=((s_srcypos-1)&0xFFFF)>>4;
int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
uint16_t *buf1=pix_buf_y[((y0+1)&1)]; // bottom line of the interpolated slice
uint16_t *uvbuf0=pix_buf_uv[y1&1]; // top line of the interpolated slice
uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1]; // bottom line of the interpolated slice
- int i;
-
- // if this is before the first line than use only the first src line
- if(y0==0) buf0= buf1;
- if(y1==0) uvbuf0= uvbuf1; // yes we do have to check this, its not the same as y0==0
if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
- // if this is after the last line than use only the last src line
- if(y0>=y+h)
- {
- buf1= buf0;
- s_last_ypos=y0;
- }
- if(y1>=(y+h)/2)
- {
- uvbuf1= uvbuf0;
- s_last_y1pos=y1;
- }
-
+ if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
s_ypos++; s_srcypos+=s_yinc;
//only interpolate the src line horizontally if we didnt do it allready
- if(s_last_ypos!=y0){
- unsigned char *src=srcptr[0]+(y0-y)*stride[0];
- unsigned int xpos=0;
- s_last_ypos=y0;
- // *** horizontal scale Y line to temp buffer
-#ifdef ARCH_X86
-
-#ifdef HAVE_MMX2
- if(canMMX2BeUsed)
- {
- asm volatile(
- "pxor %%mm7, %%mm7 \n\t"
- "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
- "movd %5, %%mm6 \n\t" // s_xinc&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "movq %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFF
- "movq %%mm2, temp0 \n\t"
- "movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "xorl %%eax, %%eax \n\t" // i
- "movl %0, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "movl %3, %%edx \n\t" // (s_xinc*4)>>16
- "xorl %%ecx, %%ecx \n\t"
- "xorl %%ebx, %%ebx \n\t"
- "movw %4, %%bx \n\t" // (s_xinc*4)&0xFFFF
-#ifdef HAVE_MMX2
-#define FUNNY_Y_CODE \
- "prefetchnta 1024(%%esi) \n\t"\
- "prefetchnta 1056(%%esi) \n\t"\
- "prefetchnta 1088(%%esi) \n\t"\
- "call funnyYCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-#else
-#define FUNNY_Y_CODE \
- "call funnyYCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-#endif
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-
- :: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>16),
- "m" ((s_xinc*4)&0xFFFF), "m" (s_xinc&0xFFFF)
- : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
- );
- for(i=dstw-1; (i*s_xinc)>>16 >=srcWidth-1; i--) buf1[i] = src[srcWidth-1]*128;
- }
- else
+ if(s_last_ypos!=y0)
{
-#endif
- //NO MMX just normal asm ...
- asm volatile(
- "xorl %%eax, %%eax \n\t" // i
- "xorl %%ebx, %%ebx \n\t" // xx
- "xorl %%ecx, %%ecx \n\t" // 2*xalpha
- "1: \n\t"
- "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, (%%edi, %%eax, 2) \n\t"
- "addw %4, %%cx \n\t" //2*xalpha += s_xinc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
-
- "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, 2(%%edi, %%eax, 2) \n\t"
- "addw %4, %%cx \n\t" //2*xalpha += s_xinc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
-
-
- "addl $2, %%eax \n\t"
- "cmpl %2, %%eax \n\t"
- " jb 1b \n\t"
+ unsigned char *src;
+ // skip if first line has been horiz scaled alleady
+ if(s_last_ypos != y0-1)
+ {
+ // check if first line is before any available src lines
+ if(y0-1 < y) src=srcptr[0]+(0 )*stride[0];
+ else src=srcptr[0]+(y0-y-1)*stride[0];
+ hyscale(buf0, dstw, src, srcWidth, s_xinc);
+ }
+ // check if second line is after any available src lines
+ if(y0-y >= h) src=srcptr[0]+(h-1)*stride[0];
+ else src=srcptr[0]+(y0-y)*stride[0];
- :: "r" (src), "m" (buf1), "m" (dstw), "m" (s_xinc>>16), "m" (s_xinc&0xFFFF)
- : "%eax", "%ebx", "%ecx", "%edi", "%esi"
- );
-#ifdef HAVE_MMX2
- } //if MMX2 cant be used
-#endif
-#else
- for(i=0;i<dstw;i++){
- register unsigned int xx=xpos>>16;
- register unsigned int xalpha=(xpos&0xFFFF)>>9;
- buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
- xpos+=s_xinc;
- }
-#endif
- }
+ // the min() is required to avoid reuseing lines which where not available
+ s_last_ypos= MIN(y0, y+h-1);
+ hyscale(buf1, dstw, src, srcWidth, s_xinc);
+ }
+// printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
// *** horizontal scale U and V lines to temp buffer
- if(s_last_y1pos!=y1){
- unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
- unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
- int xpos=0;
- s_last_y1pos= y1;
-#ifdef ARCH_X86
-#ifdef HAVE_MMX2
- if(canMMX2BeUsed)
+ if(s_last_y1pos!=y1)
{
- asm volatile(
- "pxor %%mm7, %%mm7 \n\t"
- "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
- "movd %5, %%mm6 \n\t" // s_xinc&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "movq %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFFFF
- "movq %%mm2, temp0 \n\t"
- "movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "xorl %%eax, %%eax \n\t" // i
- "movl %0, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "movl %3, %%edx \n\t" // (s_xinc*4)>>16
- "xorl %%ecx, %%ecx \n\t"
- "xorl %%ebx, %%ebx \n\t"
- "movw %4, %%bx \n\t" // (s_xinc*4)&0xFFFF
-
-#ifdef HAVE_MMX2
-#define FUNNYUVCODE \
- "prefetchnta 1024(%%esi) \n\t"\
- "prefetchnta 1056(%%esi) \n\t"\
- "prefetchnta 1088(%%esi) \n\t"\
- "call funnyUVCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-#else
-#define FUNNYUVCODE \
- "call funnyUVCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-#endif
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-
- "xorl %%eax, %%eax \n\t" // i
- "movl %6, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "addl $4096, %%edi \n\t"
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
- :: "m" (src1), "m" (uvbuf1), "m" (dstUVw), "m" ((s_xinc2*4)>>16),
- "m" ((s_xinc2*4)&0xFFFF), "m" (s_xinc2&0xFFFF), "m" (src2)
- : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
- );
- for(i=dstUVw-1; (i*s_xinc2)>>16 >=srcWidth/2-1; i--)
+ uint8_t *src1, *src2;
+ // skip if first line has been horiz scaled alleady
+ if(s_last_y1pos != y1-1)
{
- uvbuf1[i] = src1[srcWidth/2-1]*128;
- uvbuf1[i+2048] = src2[srcWidth/2-1]*128;
+ // check if first line is before any available src lines
+ if(y1-y/2-1 < 0)
+ {
+ src1= srcptr[1]+(0)*stride[1];
+ src2= srcptr[2]+(0)*stride[2];
+ }else{
+ src1= srcptr[1]+(y1-y/2-1)*stride[1];
+ src2= srcptr[2]+(y1-y/2-1)*stride[2];
+ }
+ hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
}
- }
- else
- {
-#endif
- asm volatile(
- "xorl %%eax, %%eax \n\t" // i
- "xorl %%ebx, %%ebx \n\t" // xx
- "xorl %%ecx, %%ecx \n\t" // 2*xalpha
- "1: \n\t"
- "movl %0, %%esi \n\t"
- "movzbl (%%esi, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%%esi, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, (%%edi, %%eax, 2) \n\t"
-
- "movzbl (%5, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%5, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
- "addw %4, %%cx \n\t" //2*xalpha += s_xinc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
- "addl $1, %%eax \n\t"
- "cmpl %2, %%eax \n\t"
- " jb 1b \n\t"
+ // check if second line is after any available src lines
+ if(y1 - y/2 >= h/2)
+ {
+ src1= srcptr[1]+(h/2-1)*stride[1];
+ src2= srcptr[2]+(h/2-1)*stride[2];
+ }else{
+ src1= srcptr[1]+(y1-y/2)*stride[1];
+ src2= srcptr[2]+(y1-y/2)*stride[2];
+ }
+ hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
- :: "m" (src1), "m" (uvbuf1), "m" (dstUVw), "m" (s_xinc2>>16), "m" (s_xinc2&0xFFFF),
- "r" (src2)
- : "%eax", "%ebx", "%ecx", "%edi", "%esi"
- );
-#ifdef HAVE_MMX2
- } //if MMX2 cant be used
-#endif
-#else
- for(i=0;i<dstUVw;i++){
- register unsigned int xx=xpos>>16;
- register unsigned int xalpha=(xpos&0xFFFF)>>9;
- uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
- uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
- xpos+=s_xinc2;
- }
-#endif
- // if this is the line before the first line
- if(s_srcypos == s_xinc - 0x8000)
- {
- s_srcypos= s_yinc/2 - 0x8000;
- continue;
+ // the min() is required to avoid reuseing lines which where not available
+ s_last_y1pos= MIN(y1, y/2+h/2-1);
}
- }
- if(ABS(s_yinc - 0x10000) < 10)
+ if(dstbpp==12) //YV12
+ yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
+ else if(ABS(s_yinc - 0x10000) < 10)
yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
else
yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
#endif
}
-#ifdef HAVE_3DNOW
- asm volatile("femms");
-#elif defined (HAVE_MMX)
- asm volatile("emms");
+#ifdef HAVE_MMX
+ __asm __volatile(SFENCE:::"memory");
+ __asm __volatile(EMMS:::"memory");
#endif
}
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;
+ }
+
}