opensource/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-06-29 12:26:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

interpolate chrominance for every second line in the 1:1 vertical scale function

Browse Source 
Originally committed as revision 2570 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
This commit is contained in:
Michael Niedermayer
2001-10-30 21:04:42 +00:00
parent 1faf086709
commit 497d4f9932
2 changed files with 308 additions and 84 deletions
Download Patch File Download Diff File Expand all files Collapse all files

116
postproc/swscale.c
View File

@ -249,8 +249,60 @@ static int canMMX2BeUsed=0;
"pmulhw ugCoeff, %%mm3 \n\t"\ "pmulhw ugCoeff, %%mm3 \n\t"\
"pmulhw vgCoeff, %%mm4 \n\t"\ "pmulhw vgCoeff, %%mm4 \n\t"\
/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\ /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
"movq (%1, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\ "movq (%0, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
"movq 8(%1, %%eax, 2), %%mm7 \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"\
"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*/\
"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, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\ "psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"pmulhw ubCoeff, %%mm2 \n\t"\ "pmulhw ubCoeff, %%mm2 \n\t"\
@ -803,6 +855,9 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
return; return;
} }
#ifdef HAVE_MMX #ifdef HAVE_MMX
if( yalpha > 2048 ) buf0 = buf1;
if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
{
if(dstbpp == 32) if(dstbpp == 32)
{ {
asm volatile( asm volatile(
@ -856,6 +911,63 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
: "%eax" : "%eax"
); );
} }
}
else
{
if(dstbpp == 32)
{
asm volatile(
YSCALEYUV2RGB1b
WRITEBGR32
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
else if(dstbpp==24)
{
asm volatile(
YSCALEYUV2RGB1b
WRITEBGR24
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax", "%ebx"
);
}
else if(dstbpp==15)
{
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
WRITEBGR16
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
}
#else #else
//FIXME unroll C loop and dont recalculate UV //FIXME unroll C loop and dont recalculate UV
asm volatile ("\n\t"::: "memory"); asm volatile ("\n\t"::: "memory");

116
postproc/swscale_template.c
View File

@ -249,8 +249,60 @@ static int canMMX2BeUsed=0;
"pmulhw ugCoeff, %%mm3 \n\t"\ "pmulhw ugCoeff, %%mm3 \n\t"\
"pmulhw vgCoeff, %%mm4 \n\t"\ "pmulhw vgCoeff, %%mm4 \n\t"\
/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\ /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
"movq (%1, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\ "movq (%0, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
"movq 8(%1, %%eax, 2), %%mm7 \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"\
"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*/\
"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, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\ "psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
"pmulhw ubCoeff, %%mm2 \n\t"\ "pmulhw ubCoeff, %%mm2 \n\t"\
@ -803,6 +855,9 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
return; return;
} }
#ifdef HAVE_MMX #ifdef HAVE_MMX
if( yalpha > 2048 ) buf0 = buf1;
if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
{
if(dstbpp == 32) if(dstbpp == 32)
{ {
asm volatile( asm volatile(
@ -856,6 +911,63 @@ static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, ui
: "%eax" : "%eax"
); );
} }
}
else
{
if(dstbpp == 32)
{
asm volatile(
YSCALEYUV2RGB1b
WRITEBGR32
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
else if(dstbpp==24)
{
asm volatile(
YSCALEYUV2RGB1b
WRITEBGR24
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax", "%ebx"
);
}
else if(dstbpp==15)
{
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
WRITEBGR16
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
}
#else #else
//FIXME unroll C loop and dont recalculate UV //FIXME unroll C loop and dont recalculate UV
asm volatile ("\n\t"::: "memory"); asm volatile ("\n\t"::: "memory");