sync with mplayer xp

- partial yvu9 support (copy only)
- rgb 15/16 -> 24/32 converters
- int->unsigned changes


git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@6493 b3059339-0415-0410-9bf9-f77b7e298cf2

8 files changed, 1123 insertions, 287 deletions

diff --git a/postproc/rgb2rgb.c b/postproc/rgb2rgb.c
index 91983bea0a..962a58945f 100644
--- a/postproc/rgb2rgb.c
+++ b/postproc/rgb2rgb.c
@@ -20,6 +20,8 @@
 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
 
 #ifdef CAN_COMPILE_X86_ASM
+static const uint64_t mmx_null  __attribute__((aligned(8))) = 0x0000000000000000ULL;
+static const uint64_t mmx_one   __attribute__((aligned(8))) = 0xFFFFFFFFFFFFFFFFULL;
 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;
@@ -35,6 +37,11 @@ static const uint64_t mask24hhhh  __attribute__((aligned(8))) = 0xffffffffffff00
 static const uint64_t mask15b  __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111  xxB */
 static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000  RGx */
 static const uint64_t mask15s  __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
+static const uint64_t mask15g  __attribute__((aligned(8))) = 0x03E003E003E003E0ULL;
+static const uint64_t mask15r  __attribute__((aligned(8))) = 0x7C007C007C007C00ULL;
+#define mask16b mask15b
+static const uint64_t mask16g  __attribute__((aligned(8))) = 0x07E007E007E007E0ULL;
+static const uint64_t mask16r  __attribute__((aligned(8))) = 0xF800F800F800F800ULL;
 static const uint64_t red_16mask  __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
 static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL;
 static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
@@ -137,10 +144,68 @@ void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb24to32_MMX(src, dst, src_size);
 	else
+#endif
 		rgb24to32_C(src, dst, src_size);
-#else
-		rgb24to32_C(src, dst, src_size);
+}
+
+void rgb15to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb15to24_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb15to24_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb15to24_MMX(src, dst, src_size);
+	else
 #endif
+		rgb15to24_C(src, dst, src_size);
+}
+
+void rgb16to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb16to24_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb16to24_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb16to24_MMX(src, dst, src_size);
+	else
+#endif
+		rgb16to24_C(src, dst, src_size);
+}
+
+void rgb15to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb15to32_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb15to32_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb15to32_MMX(src, dst, src_size);
+	else
+#endif
+		rgb15to32_C(src, dst, src_size);
+}
+
+void rgb16to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		rgb16to32_MMX2(src, dst, src_size);
+	else if(gCpuCaps.has3DNow)
+		rgb16to32_3DNow(src, dst, src_size);
+	else if(gCpuCaps.hasMMX)
+		rgb16to32_MMX(src, dst, src_size);
+	else
+#endif
+		rgb16to32_C(src, dst, src_size);
 }
 
 void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
@@ -154,10 +219,8 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned 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
+		rgb32to24_C(src, dst, src_size);
 }
 
 /*
@@ -177,10 +240,8 @@ void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb15to16_MMX(src, dst, src_size);
 	else
-		rgb15to16_C(src, dst, src_size);
-#else
-		rgb15to16_C(src, dst, src_size);
 #endif
+		rgb15to16_C(src, dst, src_size);
 }
 
 /**
@@ -242,10 +303,8 @@ void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb32to16_MMX(src, dst, src_size);
 	else
-		rgb32to16_C(src, dst, src_size);
-#else
-		rgb32to16_C(src, dst, src_size);
 #endif
+		rgb32to16_C(src, dst, src_size);
 }
 
 void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
@@ -259,10 +318,8 @@ void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb32to15_MMX(src, dst, src_size);
 	else
-		rgb32to15_C(src, dst, src_size);
-#else
-		rgb32to15_C(src, dst, src_size);
 #endif
+		rgb32to15_C(src, dst, src_size);
 }
 
 void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
@@ -276,10 +333,8 @@ void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb24to16_MMX(src, dst, src_size);
 	else
-		rgb24to16_C(src, dst, src_size);
-#else
-		rgb24to16_C(src, dst, src_size);
 #endif
+		rgb24to16_C(src, dst, src_size);
 }
 
 void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
@@ -293,10 +348,8 @@ void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb24to15_MMX(src, dst, src_size);
 	else
-		rgb24to15_C(src, dst, src_size);
-#else
-		rgb24to15_C(src, dst, src_size);
 #endif
+		rgb24to15_C(src, dst, src_size);
 }
 
 /**
@@ -330,10 +383,8 @@ void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb32tobgr32_MMX(src, dst, src_size);
 	else
-		rgb32tobgr32_C(src, dst, src_size);
-#else
-		rgb32tobgr32_C(src, dst, src_size);
 #endif
+		rgb32tobgr32_C(src, dst, src_size);
 }
 
 void rgb24tobgr24(const uint8_t *src, uint8_t *dst, unsigned int src_size)
@@ -347,10 +398,8 @@ void rgb24tobgr24(const uint8_t *src, uint8_t *dst, unsigned int src_size)
 	else if(gCpuCaps.hasMMX)
 		rgb24tobgr24_MMX(src, dst, src_size);
 	else
-		rgb24tobgr24_C(src, dst, src_size);
-#else
-		rgb24tobgr24_C(src, dst, src_size);
 #endif
+		rgb24tobgr24_C(src, dst, src_size);
 }
 
 /**
@@ -371,10 +420,8 @@ void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, u
 	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
-		yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 #endif
+		yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 }
 
 /**
@@ -394,10 +441,8 @@ void yuv422ptoyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc
 	else if(gCpuCaps.hasMMX)
 		yuv422ptoyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 	else
-		yuv422ptoyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
-#else
-		yuv422ptoyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 #endif
+		yuv422ptoyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
 }
 
 /**
@@ -418,10 +463,8 @@ void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
 	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
-		yuy2toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 #endif
+		yuy2toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 }
 
 /**
@@ -488,14 +531,13 @@ void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst
 	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
+		rgb24toyv12_C(src, ydst, udst, vdst, width,  height, lumStride, chromStride, srcStride);
 }
 
 void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
-		     int width, int height, int src1Stride, int src2Stride, int dstStride)
+		     unsigned width, unsigned height, unsigned src1Stride,
+		     unsigned src2Stride, unsigned dstStride)
 {
 #ifdef CAN_COMPILE_X86_ASM
 	// ordered per speed fasterst first
@@ -506,8 +548,6 @@ void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
 	else if(gCpuCaps.hasMMX)
 		interleaveBytes_MMX(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
 	else
-		interleaveBytes_C(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
-#else
-		interleaveBytes_C(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
 #endif
+		interleaveBytes_C(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
 }
diff --git a/postproc/rgb2rgb.h b/postproc/rgb2rgb.h
index fb4f04590d..9fb6da6ef1 100644
--- a/postproc/rgb2rgb.h
+++ b/postproc/rgb2rgb.h
@@ -10,12 +10,16 @@
 #define RGB2RGB_INCLUDED
 
 extern void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb24to16(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb24to15(const uint8_t *src,uint8_t *dst,unsigned src_size);
 extern void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size);
-extern void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size);
 extern void rgb32to16(const uint8_t *src,uint8_t *dst,unsigned src_size);
 extern void rgb32to15(const uint8_t *src,uint8_t *dst,unsigned src_size);
-extern void rgb24to16(const uint8_t *src,uint8_t *dst,unsigned src_size);
-extern void rgb24to15(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb15to24(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb15to32(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb16to24(const uint8_t *src,uint8_t *dst,unsigned src_size);
+extern void rgb16to32(const uint8_t *src,uint8_t *dst,unsigned src_size);
 extern void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned src_size);
 extern void rgb24tobgr24(const uint8_t *src, uint8_t *dst, unsigned src_size);
 
@@ -39,7 +43,8 @@ extern void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_
 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride);
 
 extern void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
-			    int width, int height, int src1Stride, int src2Stride, int dstStride);
+			    unsigned width, unsigned height, unsigned src1Stride,
+			    unsigned src2Stride, unsigned dstStride);
 	
 
 #define MODE_RGB  0x1
@@ -47,11 +52,11 @@ extern void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
 
 typedef void (* yuv2rgb_fun) (uint8_t * image, uint8_t * py,
 			      uint8_t * pu, uint8_t * pv,
-			      int h_size, int v_size,
-			      int rgb_stride, int y_stride, int uv_stride);
+			      unsigned h_size, unsigned v_size,
+			      unsigned rgb_stride, unsigned y_stride, unsigned uv_stride);
 
 extern yuv2rgb_fun yuv2rgb;
 
-void yuv2rgb_init (int bpp, int mode);
+void yuv2rgb_init (unsigned bpp, int mode);
 
 #endif
diff --git a/postproc/rgb2rgb_template.c b/postproc/rgb2rgb_template.c
index 9d59eabc70..015e7f2d56 100644
--- a/postproc/rgb2rgb_template.c
+++ b/postproc/rgb2rgb_template.c
@@ -8,6 +8,13 @@
  *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
  */
 
+#include <stddef.h>
+#include <inttypes.h> /* for __WORDSIZE */
+
+#ifndef __WORDSIZE
+#warning You have misconfigured system and probably will lose performance!
+#endif
+
 #undef PREFETCH
 #undef MOVNTQ
 #undef EMMS
@@ -56,13 +63,13 @@ static inline void RENAME(rgb24to32)(const uint8_t *src,uint8_t *dst,unsigned sr
   const uint8_t *s = src;
   const uint8_t *end;
 #ifdef HAVE_MMX
-  const uint8_t *mm_end;
+  uint8_t *mm_end;
 #endif
   end = s + src_size;
 #ifdef HAVE_MMX
   __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
-  mm_end = end - 23;
   __asm __volatile("movq	%0, %%mm7"::"m"(mask32):"memory");
+  mm_end = (uint8_t*)((((unsigned long)end)/24)*24);
   while(s < mm_end)
   {
     __asm __volatile(
@@ -107,12 +114,12 @@ static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,unsigned sr
   const uint8_t *s = src;
   const uint8_t *end;
 #ifdef HAVE_MMX
-  const uint8_t *mm_end;
+  uint8_t *mm_end;
 #endif
   end = s + src_size;
 #ifdef HAVE_MMX
   __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
-  mm_end = end - 31;
+  mm_end = (uint8_t*)((((unsigned long)end)/32)*32);
   while(s < mm_end)
   {
     __asm __volatile(
@@ -186,15 +193,16 @@ static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,unsigned sr
 */
 static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size)
 {
+  register const uint8_t* s=src;
+  register uint8_t* d=dst;
+  register const uint8_t *end;
+  uint8_t *mm_end;
+  end = s + src_size;
 #ifdef HAVE_MMX
-  register int offs=15-src_size;
-  register const char* s=src-offs;
-  register char* d=dst-offs;
-  __asm __volatile(PREFETCH"	%0"::"m"(*(s+offs)));
-  __asm __volatile(
-	"movq	%0, %%mm4\n\t"
-	::"m"(mask15s));
-  while(offs<0)
+  __asm __volatile(PREFETCH"	%0"::"m"(*s));
+  __asm __volatile("movq	%0, %%mm4"::"m"(mask15s));
+  mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
+  while(s<mm_end)
   {
 	__asm __volatile(
 		PREFETCH"	32%1\n\t"
@@ -208,40 +216,28 @@ static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned sr
 		"paddw	%%mm3, %%mm2\n\t"
 		MOVNTQ"	%%mm0, %0\n\t"
 		MOVNTQ"	%%mm2, 8%0"
-		:"=m"(*(d+offs))
-		:"m"(*(s+offs))
+		:"=m"(*d)
+		:"m"(*s)
 		);
-	offs+=16;
+	d+=16;
+	s+=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 );
-   }
-#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
 #endif
+    mm_end = (uint8_t*)((((unsigned long)end)/4)*4);
+    while(s < mm_end)
+    {
+	register unsigned x= *((uint32_t *)s);
+	*((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
+	d+=4;
+	s+=4;
+    }
+    if(s < end)
+    {
+	register unsigned short x= *((uint16_t *)s);
+	*((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
+    }
 }
 
 static inline void RENAME(bgr24torgb24)(const uint8_t *src, uint8_t *dst, unsigned src_size)
@@ -257,17 +253,20 @@ static inline void RENAME(bgr24torgb24)(const uint8_t *src, uint8_t *dst, unsign
 
 static inline void RENAME(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;
+	const uint8_t *end;
+#ifdef HAVE_MMX
+	const uint8_t *mm_end;
+#endif
 	uint16_t *d = (uint16_t *)dst;
 	end = s + src_size;
-	mm_end = end - 15;
+#ifdef HAVE_MMX
 	__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));
+	mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
 	while(s < mm_end)
 	{
 	    __asm __volatile(
@@ -303,43 +302,35 @@ static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, unsigned
 		d += 4;
 		s += 16;
 	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
 	while(s < end)
 	{
 		const int b= *s++;
 		const int g= *s++;
 		const int r= *s++;
-                s++;
 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
+		s++;
 	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
-#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);
-	}
-#endif
 }
 
 static inline void RENAME(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;
+	const uint8_t *end;
+#ifdef HAVE_MMX
+	const uint8_t *mm_end;
+#endif
 	uint16_t *d = (uint16_t *)dst;
 	end = s + src_size;
-	mm_end = end - 15;
+#ifdef HAVE_MMX
 	__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));
+	mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
 	while(s < mm_end)
 	{
 	    __asm __volatile(
@@ -375,43 +366,35 @@ static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, unsigned
 		d += 4;
 		s += 16;
 	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
 	while(s < end)
 	{
 		const int b= *s++;
 		const int g= *s++;
 		const int r= *s++;
-		s++;
 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
+		s++;
 	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
-#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);
-	}
-#endif
 }
 
 static inline void RENAME(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;
+	const uint8_t *end;
+#ifdef HAVE_MMX
+	const uint8_t *mm_end;
+#endif
 	uint16_t *d = (uint16_t *)dst;
 	end = s + src_size;
-	mm_end = end - 11;
+#ifdef HAVE_MMX
 	__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));
+	mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
 	while(s < mm_end)
 	{
 	    __asm __volatile(
@@ -447,6 +430,9 @@ static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, unsigned
 		d += 4;
 		s += 12;
 	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
 	while(s < end)
 	{
 		const int b= *s++;
@@ -454,35 +440,24 @@ static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, unsigned
 		const int r= *s++;
 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
 	}
-	__asm __volatile(SFENCE:::"memory");
-	__asm __volatile(EMMS:::"memory");
-#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);
-	}
-#endif
 }
 
 static inline void RENAME(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;
+	const uint8_t *end;
+#ifdef HAVE_MMX
+	const uint8_t *mm_end;
+#endif
 	uint16_t *d = (uint16_t *)dst;
 	end = s + src_size;
-	mm_end = end -11;
+#ifdef HAVE_MMX
 	__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));
+	mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
 	while(s < mm_end)
 	{
 	    __asm __volatile(
@@ -518,6 +493,9 @@ static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, unsigned
 		d += 4;
 		s += 12;
 	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
 	while(s < end)
 	{
 		const int b= *s++;
@@ -525,25 +503,448 @@ static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, unsigned
 		const int r= *s++;
 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
 	}
+}
+
+/*
+  I use here less accurate approximation by simply
+ left-shifting the input
+  value and filling the low order bits with
+ zeroes. This method improves png's
+  compression but this scheme cannot reproduce white exactly, since it does not
+  generate an all-ones maximum value; the net effect is to darken the
+  image slightly.
+
+  The better method should be "left bit replication":
+
+   4 3 2 1 0
+   ---------
+   1 1 0 1 1
+
+   7 6 5 4 3  2 1 0
+   ----------------
+   1 1 0 1 1  1 1 0
+   |=======|  |===|
+       |      Leftmost Bits Repeated to Fill Open Bits
+       |
+   Original Bits
+*/
+static inline void RENAME(rgb15to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)
+{
+	const uint16_t *end;
+#ifdef HAVE_MMX
+	const uint16_t *mm_end;
+#endif
+	uint8_t *d = (uint8_t *)dst;
+	const uint16_t *s = (uint16_t *)src;
+	end = s + src_size/2;
+#ifdef HAVE_MMX
+	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
+	mm_end = (uint16_t*)((((unsigned long)end)/8)*8);
+	while(s < mm_end)
+	{
+	    __asm __volatile(
+		PREFETCH" 32%1\n\t"
+		"movq	%1, %%mm0\n\t"
+		"movq	%1, %%mm1\n\t"
+		"movq	%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$2, %%mm1\n\t"
+		"psrlq	$7, %%mm2\n\t"
+		"movq	%%mm0, %%mm3\n\t"
+		"movq	%%mm1, %%mm4\n\t"
+		"movq	%%mm2, %%mm5\n\t"
+		"punpcklwd %5, %%mm0\n\t"
+		"punpcklwd %5, %%mm1\n\t"
+		"punpcklwd %5, %%mm2\n\t"
+		"punpckhwd %5, %%mm3\n\t"
+		"punpckhwd %5, %%mm4\n\t"
+		"punpckhwd %5, %%mm5\n\t"
+		"psllq	$8, %%mm1\n\t"
+		"psllq	$16, %%mm2\n\t"
+		"por	%%mm1, %%mm0\n\t"
+		"por	%%mm2, %%mm0\n\t"
+		"psllq	$8, %%mm4\n\t"
+		"psllq	$16, %%mm5\n\t"
+		"por	%%mm4, %%mm3\n\t"
+		"por	%%mm5, %%mm3\n\t"
+
+		"movq	%%mm0, %%mm6\n\t"
+		"movq	%%mm3, %%mm7\n\t"
+		
+		"movq	8%1, %%mm0\n\t"
+		"movq	8%1, %%mm1\n\t"
+		"movq	8%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$2, %%mm1\n\t"
+		"psrlq	$7, %%mm2\n\t"
+		"movq	%%mm0, %%mm3\n\t"
+		"movq	%%mm1, %%mm4\n\t"
+		"movq	%%mm2, %%mm5\n\t"
+		"punpcklwd %5, %%mm0\n\t"
+		"punpcklwd %5, %%mm1\n\t"
+		"punpcklwd %5, %%mm2\n\t"
+		"punpckhwd %5, %%mm3\n\t"
+		"punpckhwd %5, %%mm4\n\t"
+		"punpckhwd %5, %%mm5\n\t"
+		"psllq	$8, %%mm1\n\t"
+		"psllq	$16, %%mm2\n\t"
+		"por	%%mm1, %%mm0\n\t"
+		"por	%%mm2, %%mm0\n\t"
+		"psllq	$8, %%mm4\n\t"
+		"psllq	$16, %%mm5\n\t"
+		"por	%%mm4, %%mm3\n\t"
+		"por	%%mm5, %%mm3\n\t"
+
+		:"=m"(*d)
+		:"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
+		:"memory");
+	    /* Borrowed 32 to 24 */
+	    __asm __volatile(
+		"movq	%%mm0, %%mm4\n\t"
+		"movq	%%mm3, %%mm5\n\t"
+		"movq	%%mm6, %%mm0\n\t"
+		"movq	%%mm7, %%mm1\n\t"
+		
+		"movq	%%mm4, %%mm6\n\t"
+		"movq	%%mm5, %%mm7\n\t"
+		"movq	%%mm0, %%mm2\n\t"
+		"movq	%%mm1, %%mm3\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"(*d)
+		:"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
+		:"memory");
+		d += 24;
+		s += 8;
+	}
 	__asm __volatile(SFENCE:::"memory");
 	__asm __volatile(EMMS:::"memory");
-#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++)
+#endif
+	while(s < end)
+	{
+		register uint16_t bgr;
+		bgr = *s++;
+		*d++ = (bgr&0x1F)<<3;
+		*d++ = (bgr&0x3E0)>>2;
+		*d++ = (bgr&0x7C00)>>7;
+	}
+}
+
+static inline void RENAME(rgb16to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)
+{
+	const uint16_t *end;
+#ifdef HAVE_MMX
+	const uint16_t *mm_end;
+#endif
+	uint8_t *d = (uint8_t *)dst;
+	const uint16_t *s = (const uint16_t *)src;
+	end = s + src_size/2;
+#ifdef HAVE_MMX
+	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
+	mm_end = (uint16_t*)((((unsigned long)end)/8)*8);
+	while(s < mm_end)
+	{
+	    __asm __volatile(
+		PREFETCH" 32%1\n\t"
+		"movq	%1, %%mm0\n\t"
+		"movq	%1, %%mm1\n\t"
+		"movq	%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$3, %%mm1\n\t"
+		"psrlq	$8, %%mm2\n\t"
+		"movq	%%mm0, %%mm3\n\t"
+		"movq	%%mm1, %%mm4\n\t"
+		"movq	%%mm2, %%mm5\n\t"
+		"punpcklwd %5, %%mm0\n\t"
+		"punpcklwd %5, %%mm1\n\t"
+		"punpcklwd %5, %%mm2\n\t"
+		"punpckhwd %5, %%mm3\n\t"
+		"punpckhwd %5, %%mm4\n\t"
+		"punpckhwd %5, %%mm5\n\t"
+		"psllq	$8, %%mm1\n\t"
+		"psllq	$16, %%mm2\n\t"
+		"por	%%mm1, %%mm0\n\t"
+		"por	%%mm2, %%mm0\n\t"
+		"psllq	$8, %%mm4\n\t"
+		"psllq	$16, %%mm5\n\t"
+		"por	%%mm4, %%mm3\n\t"
+		"por	%%mm5, %%mm3\n\t"
+		
+		"movq	%%mm0, %%mm6\n\t"
+		"movq	%%mm3, %%mm7\n\t"
+
+		"movq	8%1, %%mm0\n\t"
+		"movq	8%1, %%mm1\n\t"
+		"movq	8%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$3, %%mm1\n\t"
+		"psrlq	$8, %%mm2\n\t"
+		"movq	%%mm0, %%mm3\n\t"
+		"movq	%%mm1, %%mm4\n\t"
+		"movq	%%mm2, %%mm5\n\t"
+		"punpcklwd %5, %%mm0\n\t"
+		"punpcklwd %5, %%mm1\n\t"
+		"punpcklwd %5, %%mm2\n\t"
+		"punpckhwd %5, %%mm3\n\t"
+		"punpckhwd %5, %%mm4\n\t"
+		"punpckhwd %5, %%mm5\n\t"
+		"psllq	$8, %%mm1\n\t"
+		"psllq	$16, %%mm2\n\t"
+		"por	%%mm1, %%mm0\n\t"
+		"por	%%mm2, %%mm0\n\t"
+		"psllq	$8, %%mm4\n\t"
+		"psllq	$16, %%mm5\n\t"
+		"por	%%mm4, %%mm3\n\t"
+		"por	%%mm5, %%mm3\n\t"
+		:"=m"(*d)
+		:"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)		
+		:"memory");
+	    /* Borrowed 32 to 24 */
+	    __asm __volatile(
+		"movq	%%mm0, %%mm4\n\t"
+		"movq	%%mm3, %%mm5\n\t"
+		"movq	%%mm6, %%mm0\n\t"
+		"movq	%%mm7, %%mm1\n\t"
+		
+		"movq	%%mm4, %%mm6\n\t"
+		"movq	%%mm5, %%mm7\n\t"
+		"movq	%%mm0, %%mm2\n\t"
+		"movq	%%mm1, %%mm3\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"(*d)
+		:"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
+		:"memory");
+		d += 24;
+		s += 8;
+	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
+	while(s < end)
+	{
+		register uint16_t bgr;
+		bgr = *s++;
+		*d++ = (bgr&0x1F)<<3;
+		*d++ = (bgr&0x7E0)>>3;
+		*d++ = (bgr&0xF800)>>8;
+	}
+}
+
+static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)
+{
+	const uint16_t *end;
+#ifdef HAVE_MMX
+	const uint16_t *mm_end;
+#endif
+	uint8_t *d = (uint8_t *)dst;
+	const uint16_t *s = (const uint16_t *)src;
+	end = s + src_size/2;
+#ifdef HAVE_MMX
+	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
+	__asm __volatile("pxor	%%mm7,%%mm7\n\t":::"memory");
+	mm_end = (uint16_t*)((((unsigned long)end)/4)*4);
+	while(s < mm_end)
+	{
+	    __asm __volatile(
+		PREFETCH" 32%1\n\t"
+		"movq	%1, %%mm0\n\t"
+		"movq	%1, %%mm1\n\t"
+		"movq	%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$2, %%mm1\n\t"
+		"psrlq	$7, %%mm2\n\t"
+		"movq	%%mm0, %%mm3\n\t"
+		"movq	%%mm1, %%mm4\n\t"
+		"movq	%%mm2, %%mm5\n\t"
+		"punpcklwd %%mm7, %%mm0\n\t"
+		"punpcklwd %%mm7, %%mm1\n\t"
+		"punpcklwd %%mm7, %%mm2\n\t"
+		"punpckhwd %%mm7, %%mm3\n\t"
+		"punpckhwd %%mm7, %%mm4\n\t"
+		"punpckhwd %%mm7, %%mm5\n\t"
+		"psllq	$8, %%mm1\n\t"
+		"psllq	$16, %%mm2\n\t"
+		"por	%%mm1, %%mm0\n\t"
+		"por	%%mm2, %%mm0\n\t"
+		"psllq	$8, %%mm4\n\t"
+		"psllq	$16, %%mm5\n\t"
+		"por	%%mm4, %%mm3\n\t"
+		"por	%%mm5, %%mm3\n\t"
+		MOVNTQ"	%%mm0, %0\n\t"
+		MOVNTQ"	%%mm3, 8%0\n\t"
+		:"=m"(*d)
+		:"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r)
+		:"memory");
+		d += 16;
+		s += 4;
+	}
+	__asm __volatile(SFENCE:::"memory");
+	__asm __volatile(EMMS:::"memory");
+#endif
+	while(s < end)
 	{
-		const int b= src[i+0];
-		const int g= src[i+1];
-		const int r= src[i+2];
+		register uint16_t bgr;
+		bgr = *s++;
+		*d++ = (bgr&0x1F)<<3;
+		*d++ = (bgr&0x3E0)>>2;
+		*d++ = (bgr&0x7C00)>>7;
+		*d++ = 0;
+	}
+}
 
-		d[j]= (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
+static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)
+{
+	const uint16_t *end;
+#ifdef HAVE_MMX
+	const uint16_t *mm_end;
+#endif
+	uint8_t *d = (uint8_t *)dst;
+	const uint16_t *s = (uint16_t *)src;
+	end = s + src_size/2;
+#ifdef HAVE_MMX
+	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");
+	__asm __volatile("pxor	%%mm7,%%mm7\n\t":::"memory");
+	mm_end = (uint16_t*)((((unsigned long)end)/4)*4);
+	while(s < mm_end)
+	{
+	    __asm __volatile(
+		PREFETCH" 32%1\n\t"
+		"movq	%1, %%mm0\n\t"
+		"movq	%1, %%mm1\n\t"
+		"movq	%1, %%mm2\n\t"
+		"pand	%2, %%mm0\n\t"
+		"pand	%3, %%mm1\n\t"
+		"pand	%4, %%mm2\n\t"
+		"psllq	$3, %%mm0\n\t"
+		"psrlq	$3, %%mm1\n\t"
+		"psrlq	$8, %%mm2\n\t"