#include "swscale.h"
#include "../mmx_defs.h"
#undef MOVNTQ
+#undef PAVGB
//#undef HAVE_MMX2
//#undef HAVE_MMX
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)))
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;
"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"\
- "paddw %%mm5, %%mm4 \n\t"\
- "psraw $5, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
- "psraw $5, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
+ "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*/\
" 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 */\
"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)
for(i=0;i<dstw;i++)
{
- ((uint8_t*)dest)[0] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
- dest++;
+ ((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
}
if(uvalpha != -1)
{
- for(i=0; i<dstw/2; i++)
+ for(i=0; i<(dstw>>1); i++)
{
- ((uint8_t*)uDest)[0] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
- ((uint8_t*)vDest)[0] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
- uDest++;
- vDest++;
+ ((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
+ ((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
}
}
}
{
int yalpha1=yalpha^4095;
int uvalpha1=uvalpha^4095;
- int i;
if(fullUVIpol)
{
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)];
}
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_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);
- 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_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);
- 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
asm volatile ("\n\t"::: "memory");
- if(dstbpp==32 || dstbpp==24)
+ 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_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;
+ 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_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);
- 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_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);
- 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;
}
-#ifdef HAVE_MMX
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)
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"
);
else if(dstbpp==24)
{
asm volatile(
+ "movl %4, %%ebx \n\t"
YSCALEYUV2RGB1b
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
+//FIXME write 2 versions (for even & odd lines)
asm volatile ("\n\t"::: "memory");
- if(dstbpp==32 || dstbpp==24)
+ 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]>>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_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;
+ 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)
{
- 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]>>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_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);
- dest+=2;
+ 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)
{
- 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]>>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_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);
- dest+=2;
+ 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 inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
{
- int i;
- unsigned int xpos=0;
// *** horizontal scale Y line to temp buffer
#ifdef ARCH_X86
#ifdef HAVE_MMX2
+ int i;
if(canMMX2BeUsed)
{
asm volatile(
} //if MMX2 cant be used
#endif
#else
- 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;
- }
+ 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)
{
- int xpos=0;
- int i;
#ifdef ARCH_X86
#ifdef HAVE_MMX2
+ int i;
if(canMMX2BeUsed)
{
asm volatile(
} //if MMX2 cant be used
#endif
#else
- 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);
+ 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;
- }
+ xpos+=xInc;
+ }
#endif
}
static int s_last_ypos;
static int s_last_y1pos;
-static int static_dstw;
-
#ifdef HAVE_MMX2
// used to detect a horizontal size change
static int old_dstw= -1;
static int old_s_xinc= -1;
#endif
-int srcWidth= (dstw*s_xinc + 0x8000)>>16;
-int dstUVw= fullUVIpol ? dstw : dstw/2;
+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;
// force calculation of the horizontal interpolation of the first line
if(y==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
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(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
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;
+ }
+
}