runtime cpu detection

rgb24toyv12 in C


git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3133 b3059339-0415-0410-9bf9-f77b7e298cf2

2 files changed, 320 insertions, 857 deletions

diff --git a/postproc/rgb2rgb.c b/postproc/rgb2rgb.c
index d345854d95..52ddcf5749 100644
--- a/postproc/rgb2rgb.c
+++ b/postproc/rgb2rgb.c
@@ -5,14 +5,18 @@
  *               Software YUV to YUV convertor
  *               Software YUV to RGB convertor
  *  Written by Nick Kurshev.
- *  palette stuff & yuv stuff by Michael
+ *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
  */
 #include <inttypes.h>
 #include "../config.h"
 #include "rgb2rgb.h"
-#include "../mmx_defs.h"
+#include "../cpudetect.h"
 
-#ifdef HAVE_MMX
+#ifdef ARCH_X86
+#define CAN_COMPILE_X86_ASM
+#endif
+
+#ifdef CAN_COMPILE_X86_ASM
 static const uint64_t mask32b  __attribute__((aligned(8))) = 0x000000FF000000FFULL;
 static const uint64_t mask32g  __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
 static const uint64_t mask32r  __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
@@ -47,133 +51,89 @@ static uint64_t __attribute__((aligned(8))) dither8[2]={
 #endif
 #endif
 
+#define RGB2YUV_SHIFT 8
+#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
+#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
+#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
+#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
+#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
+#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
+#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
+
+//Note: we have C, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
+//Plain C versions
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#undef ARCH_X86
+#define RENAME(a) a ## _C
+#include "rgb2rgb_template.c"
+
+#ifdef CAN_COMPILE_X86_ASM
+
+//MMX versions
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _MMX
+#include "rgb2rgb_template.c"
+
+//MMX2 versions
+#undef RENAME
+#define HAVE_MMX
+#define HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _MMX2
+#include "rgb2rgb_template.c"
+
+//3DNOW versions
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#define HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _3DNow
+#include "rgb2rgb_template.c"
+
+#endif //CAN_COMPILE_X86_ASM
+
 void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
 {
-  uint8_t *dest = dst;
-  const uint8_t *s = src;
-  const uint8_t *end;
-#ifdef HAVE_MMX
-  uint8_t *mm_end;
-#endif
-  end = s + src_size;
-#ifdef HAVE_MMX
-  __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
-  mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4));
-  __asm __volatile("movq	%0, %%mm7"::"m"(mask32):"memory");
-  if(mm_end == end) mm_end -= MMREG_SIZE*4;
-  while(s < mm_end)
-  {
-    __asm __volatile(
-	PREFETCH"	32%1\n\t"
-	"movd	%1, %%mm0\n\t"
-	"punpckldq 3%1, %%mm0\n\t"
-	"movd	6%1, %%mm1\n\t"
-	"punpckldq 9%1, %%mm1\n\t"
-	"movd	12%1, %%mm2\n\t"
-	"punpckldq 15%1, %%mm2\n\t"
-	"movd	18%1, %%mm3\n\t"
-	"punpckldq 21%1, %%mm3\n\t"
-	"pand	%%mm7, %%mm0\n\t"
-	"pand	%%mm7, %%mm1\n\t"
-	"pand	%%mm7, %%mm2\n\t"
-	"pand	%%mm7, %%mm3\n\t"
-	MOVNTQ"	%%mm0, %0\n\t"
-	MOVNTQ"	%%mm1, 8%0\n\t"
-	MOVNTQ"	%%mm2, 16%0\n\t"
-	MOVNTQ"	%%mm3, 24%0"
-	:"=m"(*dest)
-	:"m"(*s)
-	:"memory");
-    dest += 32;
-    s += 24;
-  }
-  __asm __volatile(SFENCE:::"memory");
-  __asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb24to32_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb24to32_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb24to32_MMX(src, dst, src_size);
+	else
+		rgb24to32_C(src, dst, src_size);
+#else
+		rgb24to32_C(src, dst, src_size);
 #endif
-  while(s < end)
-  {
-    *dest++ = *s++;
-    *dest++ = *s++;
-    *dest++ = *s++;
-    *dest++ = 0;
-  }
 }
 
 void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
 {
-  uint8_t *dest = dst;
-  const uint8_t *s = src;
-  const uint8_t *end;
-#ifdef HAVE_MMX
-  uint8_t *mm_end;
-#endif
-  end = s + src_size;
-#ifdef HAVE_MMX
-  __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
-  mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4));
-  while(s < mm_end)
-  {
-    __asm __volatile(
-	PREFETCH"	32%1\n\t"
-	"movq	%1, %%mm0\n\t"
-	"movq	8%1, %%mm1\n\t"
-	"movq	16%1, %%mm4\n\t"
-	"movq	24%1, %%mm5\n\t"
-	"movq	%%mm0, %%mm2\n\t"
-	"movq	%%mm1, %%mm3\n\t"
-	"movq	%%mm4, %%mm6\n\t"
-	"movq	%%mm5, %%mm7\n\t"
-	"psrlq	$8, %%mm2\n\t"
-	"psrlq	$8, %%mm3\n\t"
-	"psrlq	$8, %%mm6\n\t"
-	"psrlq	$8, %%mm7\n\t"
-	"pand	%2, %%mm0\n\t"
-	"pand	%2, %%mm1\n\t"
-	"pand	%2, %%mm4\n\t"
-	"pand	%2, %%mm5\n\t"
-	"pand	%3, %%mm2\n\t"
-	"pand	%3, %%mm3\n\t"
-	"pand	%3, %%mm6\n\t"
-	"pand	%3, %%mm7\n\t"
-	"por	%%mm2, %%mm0\n\t"
-	"por	%%mm3, %%mm1\n\t"
-	"por	%%mm6, %%mm4\n\t"
-	"por	%%mm7, %%mm5\n\t"
-
-	"movq	%%mm1, %%mm2\n\t"
-	"movq	%%mm4, %%mm3\n\t"
-	"psllq	$48, %%mm2\n\t"
-	"psllq	$32, %%mm3\n\t"
-	"pand	%4, %%mm2\n\t"
-	"pand	%5, %%mm3\n\t"
-	"por	%%mm2, %%mm0\n\t"
-	"psrlq	$16, %%mm1\n\t"
-	"psrlq	$32, %%mm4\n\t"
-	"psllq	$16, %%mm5\n\t"
-	"por	%%mm3, %%mm1\n\t"
-	"pand	%6, %%mm5\n\t"
-	"por	%%mm5, %%mm4\n\t"
-	
-	MOVNTQ"	%%mm0, %0\n\t"
-	MOVNTQ"	%%mm1, 8%0\n\t"
-	MOVNTQ"	%%mm4, 16%0"
-	:"=m"(*dest)
-	:"m"(*s),"m"(mask24l),
-	 "m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
-	:"memory");
-    dest += 24;
-    s += 32;
-  }
-  __asm __volatile(SFENCE:::"memory");
-  __asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb32to24_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb32to24_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb32to24_MMX(src, dst, src_size);
+	else
+		rgb32to24_C(src, dst, src_size);
+#else
+		rgb32to24_C(src, dst, src_size);
 #endif
-  while(s < end)
-  {
-    *dest++ = *s++;
-    *dest++ = *s++;
-    *dest++ = *s++;
-    s++;
-  }
 }
 
 /*
@@ -184,61 +144,18 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
 */
 void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
 {
-#ifdef HAVE_MMX
-  register const char* s=src+src_size;
-  register char* d=dst+src_size;
-  register int offs=-src_size;
-  __asm __volatile(PREFETCH"	%0"::"m"(*(s+offs)));
-  __asm __volatile(
-	"movq	%0, %%mm4\n\t"
-	::"m"(mask15s));
-  while(offs<0)
-  {
-	__asm __volatile(
-		PREFETCH"	32%1\n\t"
-		"movq	%1, %%mm0\n\t"
-		"movq	8%1, %%mm2\n\t"
-		"movq	%%mm0, %%mm1\n\t"
-		"movq	%%mm2, %%mm3\n\t"
-		"pand	%%mm4, %%mm0\n\t"
-		"pand	%%mm4, %%mm2\n\t"
-		"paddw	%%mm1, %%mm0\n\t"
-		"paddw	%%mm3, %%mm2\n\t"
-		MOVNTQ"	%%mm0, %0\n\t"
-		MOVNTQ"	%%mm2, 8%0"
-		:"=m"(*(d+offs))
-		:"m"(*(s+offs))
-		);
-	offs+=16;
-  }
-  __asm __volatile(SFENCE:::"memory");
-  __asm __volatile(EMMS:::"memory");
-#else
-#if 0
-   const uint16_t *s1=( uint16_t * )src;
-   uint16_t *d1=( uint16_t * )dst;
-   uint16_t *e=((uint8_t *)s1)+src_size;
-   while( s1<e ){
-     register int x=*( s1++ );
-     /* rrrrrggggggbbbbb
-        0rrrrrgggggbbbbb
-        0111 1111 1110 0000=0x7FE0
-        00000000000001 1111=0x001F */
-     *( d1++ )=( x&0x001F )|( ( x&0x7FE0 )<<1 );
-   }
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb15to16_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb15to16_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb15to16_MMX(src, dst, src_size);
+	else
+		rgb15to16_C(src, dst, src_size);
 #else
-	const unsigned *s1=( unsigned * )src;
-	unsigned *d1=( unsigned * )dst;
-	int i;
-	int size= src_size>>2;
-	for(i=0; i<size; i++)
-	{
-		register int x= s1[i];
-//		d1[i] = x + (x&0x7FE07FE0); //faster but need msbit =0 which might not allways be true
-		d1[i] = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
-
-	}
-#endif
+		rgb15to16_C(src, dst, src_size);
 #endif
 }
 
@@ -275,287 +192,69 @@ void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons
 
 void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
 {
-#ifdef HAVE_MMX
-	const uint8_t *s = src;
-	const uint8_t *end,*mm_end;
-	uint16_t *d = (uint16_t *)dst;
-	end = s + src_size;
-	mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
-	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");
-	__asm __volatile(
-	    "movq	%0, %%mm7\n\t"
-	    "movq	%1, %%mm6\n\t"
-	    ::"m"(red_16mask),"m"(green_16mask));
-	while(s < mm_end)
-	{
-	    __asm __volatile(
-		PREFETCH" 32%1\n\t"
-		"movd	%1, %%mm0\n\t"
-		"movd	4%1, %%mm3\n\t"
-		"punpckldq 8%1, %%mm0\n\t"
-		"punpckldq 12%1, %%mm3\n\t"
-		"movq	%%mm0, %%mm1\n\t"
-		"movq	%%mm0, %%mm2\n\t"
-		"movq	%%mm3, %%mm4\n\t"
-		"movq	%%mm3, %%mm5\n\t"
-		"psrlq	$3, %%mm0\n\t"
-		"psrlq	$3, %%mm3\n\t"
-		"pand	%2, %%mm0\n\t"
-		"pand	%2, %%mm3\n\t"
-		"psrlq	$5, %%mm1\n\t"
-		"psrlq	$5, %%mm4\n\t"
-		"pand	%%mm6, %%mm1\n\t"
-		"pand	%%mm6, %%mm4\n\t"
-		"psrlq	$8, %%mm2\n\t"
-		"psrlq	$8, %%mm5\n\t"
-		"pand	%%mm7, %%mm2\n\t"
-		"pand	%%mm7, %%mm5\n\t"
-		"por	%%mm1, %%mm0\n\t"
-		"por	%%mm4, %%mm3\n\t"
-		"por	%%mm2, %%mm0\n\t"
-		"por	%%mm5, %%mm3\n\t"
-		"psllq	$16, %%mm3\n\t"
-		"por	%%mm3, %%mm0\n\t"
-		MOVNTQ"	%%mm0, %0\n\t"
-		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
-		d += 4;
-		s += 16;
-	}
-	while(s < end)
-	{
-		const int b= *s++;
-		const int g= *s++;
-		const int r= *s++;
-		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
-	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb32to16_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb32to16_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb32to16_MMX(src, dst, src_size);
+	else
+		rgb32to16_C(src, dst, src_size);
 #else
-	unsigned j,i,num_pixels=src_size/4;
-	uint16_t *d = (uint16_t *)dst;
-	for(i=0,j=0; j<num_pixels; i+=4,j++)
-	{
-		const int b= src[i+0];
-		const int g= src[i+1];
-		const int r= src[i+2];
-
-		d[j]= (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
-	}
+		rgb32to16_C(src, dst, src_size);
 #endif
 }
 
 void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
 {
-#ifdef HAVE_MMX
-	const uint8_t *s = src;
-	const uint8_t *end,*mm_end;
-	uint16_t *d = (uint16_t *)dst;
-	end = s + src_size;
-	mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
-	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");
-	__asm __volatile(
-	    "movq	%0, %%mm7\n\t"
-	    "movq	%1, %%mm6\n\t"
-	    ::"m"(red_15mask),"m"(green_15mask));
-	while(s < mm_end)
-	{
-	    __asm __volatile(
-		PREFETCH" 32%1\n\t"
-		"movd	%1, %%mm0\n\t"
-		"movd	4%1, %%mm3\n\t"
-		"punpckldq 8%1, %%mm0\n\t"
-		"punpckldq 12%1, %%mm3\n\t"
-		"movq	%%mm0, %%mm1\n\t"
-		"movq	%%mm0, %%mm2\n\t"
-		"movq	%%mm3, %%mm4\n\t"
-		"movq	%%mm3, %%mm5\n\t"
-		"psrlq	$3, %%mm0\n\t"
-		"psrlq	$3, %%mm3\n\t"
-		"pand	%2, %%mm0\n\t"
-		"pand	%2, %%mm3\n\t"
-		"psrlq	$6, %%mm1\n\t"
-		"psrlq	$6, %%mm4\n\t"
-		"pand	%%mm6, %%mm1\n\t"
-		"pand	%%mm6, %%mm4\n\t"
-		"psrlq	$9, %%mm2\n\t"
-		"psrlq	$9, %%mm5\n\t"
-		"pand	%%mm7, %%mm2\n\t"
-		"pand	%%mm7, %%mm5\n\t"
-		"por	%%mm1, %%mm0\n\t"
-		"por	%%mm4, %%mm3\n\t"
-		"por	%%mm2, %%mm0\n\t"
-		"por	%%mm5, %%mm3\n\t"
-		"psllq	$16, %%mm3\n\t"
-		"por	%%mm3, %%mm0\n\t"
-		MOVNTQ"	%%mm0, %0\n\t"
-		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
-		d += 4;
-		s += 16;
-	}
-	while(s < end)
-	{
-		const int b= *s++;
-		const int g= *s++;
-		const int r= *s++;
-		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
-	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb32to15_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb32to15_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb32to15_MMX(src, dst, src_size);
+	else
+		rgb32to15_C(src, dst, src_size);
 #else
-	unsigned j,i,num_pixels=src_size/4;
-	uint16_t *d = (uint16_t *)dst;
-	for(i=0,j=0; j<num_pixels; i+=4,j++)
-	{
-		const int b= src[i+0];
-		const int g= src[i+1];
-		const int r= src[i+2];
-
-		d[j]= (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
-	}
+		rgb32to15_C(src, dst, src_size);
 #endif
 }
 
 void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
 {
-#ifdef HAVE_MMX
-	const uint8_t *s = src;
-	const uint8_t *end,*mm_end;
-	uint16_t *d = (uint16_t *)dst;
-	end = s + src_size;
-	mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
-	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");
-	__asm __volatile(
-	    "movq	%0, %%mm7\n\t"
-	    "movq	%1, %%mm6\n\t"
-	    ::"m"(red_16mask),"m"(green_16mask));
-	if(mm_end == end) mm_end -= MMREG_SIZE*2;
-	while(s < mm_end)
-	{
-	    __asm __volatile(
-		PREFETCH" 32%1\n\t"
-		"movd	%1, %%mm0\n\t"
-		"movd	3%1, %%mm3\n\t"
-		"punpckldq 6%1, %%mm0\n\t"
-		"punpckldq 9%1, %%mm3\n\t"
-		"movq	%%mm0, %%mm1\n\t"
-		"movq	%%mm0, %%mm2\n\t"
-		"movq	%%mm3, %%mm4\n\t"
-		"movq	%%mm3, %%mm5\n\t"
-		"psrlq	$3, %%mm0\n\t"
-		"psrlq	$3, %%mm3\n\t"
-		"pand	%2, %%mm0\n\t"
-		"pand	%2, %%mm3\n\t"
-		"psrlq	$5, %%mm1\n\t"
-		"psrlq	$5, %%mm4\n\t"
-		"pand	%%mm6, %%mm1\n\t"
-		"pand	%%mm6, %%mm4\n\t"
-		"psrlq	$8, %%mm2\n\t"
-		"psrlq	$8, %%mm5\n\t"
-		"pand	%%mm7, %%mm2\n\t"
-		"pand	%%mm7, %%mm5\n\t"
-		"por	%%mm1, %%mm0\n\t"
-		"por	%%mm4, %%mm3\n\t"
-		"por	%%mm2, %%mm0\n\t"
-		"por	%%mm5, %%mm3\n\t"
-		"psllq	$16, %%mm3\n\t"
-		"por	%%mm3, %%mm0\n\t"
-		MOVNTQ"	%%mm0, %0\n\t"
-		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
-		d += 4;
-		s += 12;
-	}
-	while(s < end)
-	{
-		const int b= *s++;
-		const int g= *s++;
-		const int r= *s++;
-		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
-	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb24to16_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb24to16_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb24to16_MMX(src, dst, src_size);
+	else
+		rgb24to16_C(src, dst, src_size);
 #else
-	unsigned j,i,num_pixels=src_size/3;
-	uint16_t *d = (uint16_t *)dst;
-	for(i=0,j=0; j<num_pixels; i+=3,j++)
-	{
-		const int b= src[i+0];
-		const int g= src[i+1];
-		const int r= src[i+2];
-
-		d[j]= (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
-	}
+		rgb24to16_C(src, dst, src_size);
 #endif
 }
 
 void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
 {
-#ifdef HAVE_MMX
-	const uint8_t *s = src;
-	const uint8_t *end,*mm_end;
-	uint16_t *d = (uint16_t *)dst;
-	end = s + src_size;
-	mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
-	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");
-	__asm __volatile(
-	    "movq	%0, %%mm7\n\t"
-	    "movq	%1, %%mm6\n\t"
-	    ::"m"(red_15mask),"m"(green_15mask));
-	if(mm_end == end) mm_end -= MMREG_SIZE*2;
-	while(s < mm_end)
-	{
-	    __asm __volatile(
-		PREFETCH" 32%1\n\t"
-		"movd	%1, %%mm0\n\t"
-		"movd	3%1, %%mm3\n\t"
-		"punpckldq 6%1, %%mm0\n\t"
-		"punpckldq 9%1, %%mm3\n\t"
-		"movq	%%mm0, %%mm1\n\t"
-		"movq	%%mm0, %%mm2\n\t"
-		"movq	%%mm3, %%mm4\n\t"
-		"movq	%%mm3, %%mm5\n\t"
-		"psrlq	$3, %%mm0\n\t"
-		"psrlq	$3, %%mm3\n\t"
-		"pand	%2, %%mm0\n\t"
-		"pand	%2, %%mm3\n\t"
-		"psrlq	$6, %%mm1\n\t"
-		"psrlq	$6, %%mm4\n\t"
-		"pand	%%mm6, %%mm1\n\t"
-		"pand	%%mm6, %%mm4\n\t"
-		"psrlq	$9, %%mm2\n\t"
-		"psrlq	$9, %%mm5\n\t"
-		"pand	%%mm7, %%mm2\n\t"
-		"pand	%%mm7, %%mm5\n\t"
-		"por	%%mm1, %%mm0\n\t"
-		"por	%%mm4, %%mm3\n\t"
-		"por	%%mm2, %%mm0\n\t"
-		"por	%%mm5, %%mm3\n\t"
-		"psllq	$16, %%mm3\n\t"
-		"por	%%mm3, %%mm0\n\t"
-		MOVNTQ"	%%mm0, %0\n\t"
-		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
-		d += 4;
-		s += 12;
-	}
-	while(s < end)
-	{
-		const int b= *s++;
-		const int g= *s++;
-		const int r= *s++;
-		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
-	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb24to15_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb24to15_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb24to15_MMX(src, dst, src_size);
+	else
+		rgb24to15_C(src, dst, src_size);
 #else
-	unsigned j,i,num_pixels=src_size/3;
-	uint16_t *d = (uint16_t *)dst;
-	for(i=0,j=0; j<num_pixels; i+=3,j++)
-	{
-		const int b= src[i+0];
-		const int g= src[i+1];
-		const int r= src[i+2];
-
-		d[j]= (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
-	}
+		rgb24to15_C(src, dst, src_size);
 #endif
 }
 
@@ -581,41 +280,18 @@ void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons
 
 void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
 {
-	int num_pixels= src_size >> 2;
-#ifdef HAVE_MMX
-	asm volatile (
-		"xorl %%eax, %%eax		\n\t"
-		".balign 16			\n\t"
-		"1:				\n\t"
-		PREFETCH" 32(%0, %%eax)		\n\t"
-		"movq (%0, %%eax), %%mm0	\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"movq %%mm0, %%mm2		\n\t"
-		"pslld $16, %%mm0		\n\t"
-		"psrld $16, %%mm1		\n\t"
-		"pand mask32r, %%mm0		\n\t"
-		"pand mask32g, %%mm2		\n\t"
-		"pand mask32b, %%mm1		\n\t"
-		"por %%mm0, %%mm2		\n\t"
-		"por %%mm1, %%mm2		\n\t"
-		MOVNTQ" %%mm2, (%1, %%eax)	\n\t"
-		"addl $2, %%eax			\n\t"
-		"cmpl %2, %%eax			\n\t"
-		" jb 1b				\n\t"
-		:: "r" (src), "r"(dst), "r" (num_pixels)
-		: "%eax"
-	);
-
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb32tobgr32_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb32tobgr32_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb32tobgr32_MMX(src, dst, src_size);
+	else
+		rgb32tobgr32_C(src, dst, src_size);
 #else
-	int i;
-	for(i=0; i<num_pixels; i++)
-	{
-		dst[4*i + 0] = src[4*i + 2];
-		dst[4*i + 1] = src[4*i + 1];
-		dst[4*i + 2] = src[4*i + 0];
-	}
+		rgb32tobgr32_C(src, dst, src_size);
 #endif
 }
 
@@ -628,67 +304,18 @@ void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, u
 	unsigned int width, unsigned int height,
 	unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
 {
-	int y;
-	const int chromWidth= width>>1;
-	for(y=0; y<height; y++)
-	{
-#ifdef HAVE_MMX
-//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
-		asm volatile(
-			"xorl %%eax, %%eax		\n\t"
-			".balign 16			\n\t"
-			"1:				\n\t"
-			PREFETCH" 32(%1, %%eax, 2)	\n\t"
-			PREFETCH" 32(%2, %%eax)		\n\t"
-			PREFETCH" 32(%3, %%eax)		\n\t"
-			"movq (%2, %%eax), %%mm0	\n\t" // U(0)
-			"movq %%mm0, %%mm2		\n\t" // U(0)
-			"movq (%3, %%eax), %%mm1	\n\t" // V(0)
-			"punpcklbw %%mm1, %%mm0		\n\t" // UVUV UVUV(0)
-			"punpckhbw %%mm1, %%mm2		\n\t" // UVUV UVUV(8)
-
-			"movq (%1, %%eax,2), %%mm3	\n\t" // Y(0)
-			"movq 8(%1, %%eax,2), %%mm5	\n\t" // Y(8)
-			"movq %%mm3, %%mm4		\n\t" // Y(0)
-			"movq %%mm5, %%mm6		\n\t" // Y(8)
-			"punpcklbw %%mm0, %%mm3		\n\t" // YUYV YUYV(0)
-			"punpckhbw %%mm0, %%mm4		\n\t" // YUYV YUYV(4)
-			"punpcklbw %%mm2, %%mm5		\n\t" // YUYV YUYV(8)
-			"punpckhbw %%mm2, %%mm6		\n\t" // YUYV YUYV(12)
-
-			MOVNTQ" %%mm3, (%0, %%eax, 4)	\n\t"
-			MOVNTQ" %%mm4, 8(%0, %%eax, 4)	\n\t"
-			MOVNTQ" %%mm5, 16(%0, %%eax, 4)	\n\t"
-			MOVNTQ" %%mm6, 24(%0, %%eax, 4)	\n\t"
-
-			"addl $8, %%eax			\n\t"
-			"cmpl %4, %%eax			\n\t"
-			" jb 1b				\n\t"
-			::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (chromWidth)
-			: "%eax"
-		);
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		yv12toyuy2_MMX2(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
+	else if(gCpuCaps.has3DNow)
+		yv12toyuy2_3DNow(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
+	else if(gCpuCaps.hasMMX)
+		yv12toyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
+	else
+		yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 #else
-		int i;
-		for(i=0; i<chromWidth; i++)
-		{
-			dst[4*i+0] = ysrc[2*i+0];
-			dst[4*i+1] = usrc[i];
-			dst[4*i+2] = ysrc[2*i+1];
-			dst[4*i+3] = vsrc[i];
-		}
-#endif
-		if(y&1)
-		{
-			usrc += chromStride;
-			vsrc += chromStride;
-		}
-		ysrc += lumStride;
-		dst += dstStride;
-	}
-#ifdef HAVE_MMX
-asm(    EMMS" \n\t"
-        SFENCE" \n\t"
-        :::"memory");
+		yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 #endif
 }
 
@@ -701,119 +328,18 @@ void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
 	unsigned int width, unsigned int height,
 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
 {
-	int y;
-	const int chromWidth= width>>1;
-	for(y=0; y<height; y+=2)
-	{
-#ifdef HAVE_MMX
-		asm volatile(
-			"xorl %%eax, %%eax		\n\t"
-			"pcmpeqw %%mm7, %%mm7		\n\t"
-			"psrlw $8, %%mm7		\n\t" // FF,00,FF,00...
-			".balign 16			\n\t"
-			"1:				\n\t"
-			PREFETCH" 64(%0, %%eax, 4)	\n\t"
-			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)
-			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)
-			"movq %%mm0, %%mm2		\n\t" // YUYV YUYV(0)
-			"movq %%mm1, %%mm3		\n\t" // YUYV YUYV(4)
-			"psrlw $8, %%mm0		\n\t" // U0V0 U0V0(0)
-			"psrlw $8, %%mm1		\n\t" // U0V0 U0V0(4)
-			"pand %%mm7, %%mm2		\n\t" // Y0Y0 Y0Y0(0)
-			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(4)
-			"packuswb %%mm1, %%mm0		\n\t" // UVUV UVUV(0)
-			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(0)
-
-			MOVNTQ" %%mm2, (%1, %%eax, 2)	\n\t"
-
-			"movq 16(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(8)
-			"movq 24(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(12)
-			"movq %%mm1, %%mm3		\n\t" // YUYV YUYV(8)
-			"movq %%mm2, %%mm4		\n\t" // YUYV YUYV(12)
-			"psrlw $8, %%mm1		\n\t" // U0V0 U0V0(8)
-			"psrlw $8, %%mm2		\n\t" // U0V0 U0V0(12)
-			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(8)
-			"pand %%mm7, %%mm4		\n\t" // Y0Y0 Y0Y0(12)
-			"packuswb %%mm2, %%mm1		\n\t" // UVUV UVUV(8)
-			"packuswb %%mm4, %%mm3		\n\t" // YYYY YYYY(8)
-
-			MOVNTQ" %%mm3, 8(%1, %%eax, 2)	\n\t"
-
-			"movq %%mm0, %%mm2		\n\t" // UVUV UVUV(0)
-			"movq %%mm1, %%mm3		\n\t" // UVUV UVUV(8)
-			"psrlw $8, %%mm0		\n\t" // V0V0 V0V0(0)
-			"psrlw $8, %%mm1		\n\t" // V0V0 V0V0(8)
-			"pand %%mm7, %%mm2		\n\t" // U0U0 U0U0(0)
-			"pand %%mm7, %%mm3		\n\t" // U0U0 U0U0(8)
-			"packuswb %%mm1, %%mm0		\n\t" // VVVV VVVV(0)
-			"packuswb %%mm3, %%mm2		\n\t" // UUUU UUUU(0)
-
-			MOVNTQ" %%mm0, (%3, %%eax)	\n\t"
-			MOVNTQ" %%mm2, (%2, %%eax)	\n\t"
-
-			"addl $8, %%eax			\n\t"
-			"cmpl %4, %%eax			\n\t"
-			" jb 1b				\n\t"
-			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)
-			: "memory", "%eax"
-		);
-
-		ydst += lumStride;
-		src  += srcStride;
-
-		asm volatile(
-			"xorl %%eax, %%eax		\n\t"
-			".balign 16			\n\t"
-			"1:				\n\t"
-			PREFETCH" 64(%0, %%eax, 4)	\n\t"
-			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)
-			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)
-			"movq 16(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(8)
-			"movq 24(%0, %%eax, 4), %%mm3	\n\t" // YUYV YUYV(12)
-			"pand %%mm7, %%mm0		\n\t" // Y0Y0 Y0Y0(0)
-			"pand %%mm7, %%mm1		\n\t" // Y0Y0 Y0Y0(4)
-			"pand %%mm7, %%mm2		\n\t" // Y0Y0 Y0Y0(8)
-			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(12)
-			"packuswb %%mm1, %%mm0		\n\t" // YYYY YYYY(0)
-			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(8)
-
-			MOVNTQ" %%mm0, (%1, %%eax, 2)	\n\t"
-			MOVNTQ" %%mm2, 8(%1, %%eax, 2)	\n\t"
-
-			"addl $8, %%eax			\n\t"
-			"cmpl %4, %%eax			\n\t"
-			" jb 1b				\n\t"
-
-			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)
-			: "memory", "%eax"
-		);
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		yuy2toyv12_MMX2(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.has3DNow)
+		yuy2toyv12_3DNow(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.hasMMX)
+		yuy2toyv12_MMX(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else
+		yuy2toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 #else
-		int i;
-		for(i=0; i<chromWidth; i++)
-		{
-			ydst[2*i+0] 	= src[4*i+0];
-			udst[i] 	= src[4*i+1];
-			ydst[2*i+1] 	= src[4*i+2];
-			vdst[i] 	= src[4*i+3];
-		}
-		ydst += lumStride;
-		src  += srcStride;
-
-		for(i=0; i<chromWidth; i++)
-		{
-			ydst[2*i+0] 	= src[4*i+0];
-			ydst[2*i+1] 	= src[4*i+2];
-		}
-#endif
-		udst += chromStride;
-		vdst += chromStride;
-		ydst += lumStride;
-		src  += srcStride;
-	}
-#ifdef HAVE_MMX
-asm volatile(   EMMS" \n\t"
-        	SFENCE" \n\t"
-        	:::"memory");
+		yuy2toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 #endif
 }
 
@@ -821,125 +347,48 @@ asm volatile(   EMMS" \n\t"
  *
  * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
  * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
  */
 void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
 	unsigned int width, unsigned int height,
 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
 {
-	int y;
-	const int chromWidth= width>>1;
-	for(y=0; y<height; y+=2)
-	{
-#ifdef HAVE_MMX
-		asm volatile(
-			"xorl %%eax, %%eax		\n\t"
-			"pcmpeqw %%mm7, %%mm7		\n\t"
-			"psrlw $8, %%mm7		\n\t" // FF,00,FF,00...
-			".balign 16			\n\t"
-			"1:				\n\t"
-			PREFETCH" 64(%0, %%eax, 4)	\n\t"
-			"movq (%0, %%eax, 4), %%mm0	\n\t" // UYVY UYVY(0)
-			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // UYVY UYVY(4)
-			"movq %%mm0, %%mm2		\n\t" // UYVY UYVY(0)
-			"movq %%mm1, %%mm3		\n\t" // UYVY UYVY(4)
-			"pand %%mm7, %%mm0		\n\t" // U0V0 U0V0(0)
-			"pand %%mm7, %%mm1		\n\t" // U0V0 U0V0(4)
-			"psrlw $8, %%mm2		\n\t" // Y0Y0 Y0Y0(0)
-			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(4)
-			"packuswb %%mm1, %%mm0		\n\t" // UVUV UVUV(0)
-			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(0)
-
-			MOVNTQ" %%mm2, (%1, %%eax, 2)	\n\t"
-
-			"movq 16(%0, %%eax, 4), %%mm1	\n\t" // UYVY UYVY(8)
-			"movq 24(%0, %%eax, 4), %%mm2	\n\t" // UYVY UYVY(12)
-			"movq %%mm1, %%mm3		\n\t" // UYVY UYVY(8)
-			"movq %%mm2, %%mm4		\n\t" // UYVY UYVY(12)
-			"pand %%mm7, %%mm1		\n\t" // U0V0 U0V0(8)
-			"pand %%mm7, %%mm2		\n\t" // U0V0 U0V0(12)
-			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(8)
-			"psrlw $8, %%mm4		\n\t" // Y0Y0 Y0Y0(12)
-			"packuswb %%mm2, %%mm1		\n\t" // UVUV UVUV(8)
-			"packuswb %%mm4, %%mm3		\n\t" // YYYY YYYY(8)
-
-			MOVNTQ" %%mm3, 8(%1, %%eax, 2)	\n\t"
-
-			"movq %%mm0, %%mm2		\n\t" // UVUV UVUV(0)
-			"movq %%mm1, %%mm3		\n\t" // UVUV UVUV(8)
-			"psrlw $8, %%mm0		\n\t" // V0V0 V0V0(0)
-			"psrlw $8, %%mm1		\n\t" // V0V0 V0V0(8)
-			"pand %%mm7, %%mm2		\n\t" // U0U0 U0U0(0)
-			"pand %%mm7, %%mm3		\n\t" // U0U0 U0U0(8)
-			"packuswb %%mm1, %%mm0		\n\t" // VVVV VVVV(0)
-			"packuswb %%mm3, %%mm2		\n\t" // UUUU UUUU(0)
-
-			MOVNTQ" %%mm0, (%3, %%eax)	\n\t"
-			MOVNTQ" %%mm2, (%2, %%eax)	\n\t"
-
-			"addl $8, %%eax			\n\t"
-			"cmpl %4, %%eax			\n\t"
-			" jb 1b				\n\t"
-			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)
-			: "memory", "%eax"
-		);
-
-		ydst += lumStride;
-		src  += srcStride;
-
-		asm volatile(
-			"xorl %%eax, %%eax		\n\t"
-			".balign 16			\n\t"
-			"1:				\n\t"
-			PREFETCH" 64(%0, %%eax, 4)	\n\t"
-			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)
-			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)
-			"movq 16(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(8)
-			"movq 24(%0, %%eax, 4), %%mm3	\n\t" // YUYV YUYV(12)
-			"psrlw $8, %%mm0		\n\t" // Y0Y0 Y0Y0(0)
-			"psrlw $8, %%mm1		\n\t" // Y0Y0 Y0Y0(4)
-			"psrlw $8, %%mm2		\n\t" // Y0Y0 Y0Y0(8)
-			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(12)
-			"packuswb %%mm1, %%mm0		\n\t" // YYYY YYYY(0)
-			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(8)
-
-			MOVNTQ" %%mm0, (%1, %%eax, 2)	\n\t"
-			MOVNTQ" %%mm2, 8(%1, %%eax, 2)	\n\t"
-
-			"addl $8, %%eax			\n\t"
-			"cmpl %4, %%eax			\n\t"
-			" jb 1b				\n\t"
-
-			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)
-			: "memory", "%eax"
-		);
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		uyvytoyv12_MMX2(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.has3DNow)
+		uyvytoyv12_3DNow(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.hasMMX)
+		uyvytoyv12_MMX(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else
+		uyvytoyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 #else
-		int i;
-		for(i=0; i<chromWidth; i++)
-		{
-			udst[i] 	= src[4*i+0];
-			ydst[2*i+0] 	= src[4*i+1];
-			vdst[i] 	= src[4*i+2];
-			ydst[2*i+1] 	= src[4*i+3];
-		}
-		ydst += lumStride;
-		src  += srcStride;
-
-		for(i=0; i<chromWidth; i++)
-		{
-			ydst[2*i+0] 	= src[4*i+1];
-			ydst[2*i+1] 	= src[4*i+3];
-		}
-#endif
-		udst += chromStride;
-		vdst += chromStride;
-		ydst += lumStride;
-		src  += srcStride;
-	}
-#ifdef HAVE_MMX
-asm volatile(   EMMS" \n\t"
-        	SFENCE" \n\t"
-        	:::"memory");
+		uyvytoyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 #endif
 }
 
-
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
+ */
+void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+	unsigned int width, unsigned int height,
+	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb24toyv12_MMX2(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.has3DNow)
+		rgb24toyv12_3DNow(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else if(gCpuCaps.hasMMX)
+		rgb24toyv12_MMX(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+	else
+		rgb24toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+#else
+		rgb24toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
+#endif
+}
diff --git a/postproc/rgb2rgb_template.c b/postproc/rgb2rgb_template.c
index d345854d95..ff3ee83b44 100644
--- a/postproc/rgb2rgb_template.c
+++ b/postproc/rgb2rgb_template.c
@@ -5,49 +5,52 @@
  *               Software YUV to YUV convertor
  *               Software YUV to RGB convertor
  *  Written by Nick Kurshev.
- *  palette stuff & yuv stuff by Michael
+ *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
  */
-#include <inttypes.h>
-#include "../config.h"
-#include "rgb2rgb.h"
-#include "../mmx_defs.h"
 
-#ifdef HAVE_MMX
-static const uint64_t mask32b  __attribute__((aligned(8))) = 0x000000FF000000FFULL;
-static const uint64_t mask32g  __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
-static const uint64_t mask32r  __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
-static const uint64_t mask32   __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
-static const uint64_t mask24l  __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
-static const uint64_t mask24h  __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
-static const uint64_t mask24hh  __attribute__((aligned(8))) = 0xffff000000000000ULL;
-static const uint64_t mask24hhh  __attribute__((aligned(8))) = 0xffffffff00000000ULL;
-static const uint64_t mask24hhhh  __attribute__((aligned(8))) = 0xffffffffffff0000ULL;
-static const uint64_t mask15b  __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111  xxB */
-static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000  RGx */</