/*
*
* rgb2rgb.c, Software RGB to RGB convertor
+ * pluralize by Software PAL8 to RGB convertor
+ * 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
+
+#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
+
+#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;
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 */
static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
+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;
+static const uint64_t red_15mask __attribute__((aligned(8))) = 0x00007c000000f800ULL;
+static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL;
+static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
+
+#ifdef FAST_BGR2YV12
+static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
+static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
+static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
+#else
+static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
+static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
+static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
+#endif
+static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
+static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
+static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
+
+#if 0
+static volatile uint64_t __attribute__((aligned(8))) b5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g6Dither;
+static volatile uint64_t __attribute__((aligned(8))) r5Dither;
+
+static uint64_t __attribute__((aligned(8))) dither4[2]={
+ 0x0103010301030103LL,
+ 0x0200020002000200LL,};
+
+static uint64_t __attribute__((aligned(8))) dither8[2]={
+ 0x0602060206020602LL,
+ 0x0004000400040004LL,};
+#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*2))*(MMREG_SIZE*2));
- __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
- 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, %%mm1\n\t"
- "movd 6%1, %%mm2\n\t"
- "movd 9%1, %%mm3\n\t"
- "punpckldq %%mm1, %%mm0\n\t"
- "punpckldq %%mm3, %%mm2\n\t"
- "pand %%mm7, %%mm0\n\t"
- "pand %%mm7, %%mm2\n\t"
- MOVNTQ" %%mm0, %0\n\t"
- MOVNTQ" %%mm2, 8%0"
- :"=m"(*dest)
- :"m"(*s)
- :"memory");
- dest += 16;
- s += 12;
- }
- __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*2))*(MMREG_SIZE*2));
- __asm __volatile(
- "movq %0, %%mm7\n\t"
- "movq %1, %%mm6"
- ::"m"(mask24l),"m"(mask24h):"memory");
- if(mm_end == end) mm_end -= MMREG_SIZE*2;
- while(s < mm_end)
- {
- __asm __volatile(
- PREFETCH" 32%1\n\t"
- "movq %1, %%mm0\n\t"
- "movq 8%1, %%mm1\n\t"
- "movq %%mm0, %%mm2\n\t"
- "movq %%mm1, %%mm3\n\t"
- "psrlq $8, %%mm2\n\t"
- "psrlq $8, %%mm3\n\t"
- "pand %%mm7, %%mm0\n\t"
- "pand %%mm7, %%mm1\n\t"
- "pand %%mm6, %%mm2\n\t"
- "pand %%mm6, %%mm3\n\t"
- "por %%mm2, %%mm0\n\t"
- "por %%mm3, %%mm1\n\t"
- MOVNTQ" %%mm0, %0\n\t"
- MOVNTQ" %%mm1, 6%0"
- :"=m"(*dest)
- :"m"(*s)
- :"memory");
- dest += 12;
- s += 16;
- }
- __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++;
- }
}
/*
*/
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");
+#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
-#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
+ rgb15to16_C(src, dst, src_size);
#endif
}
void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
{
- unsigned i,num_pixels=src_size/4;
- for(i=0; i<num_pixels; i+=4)
- {
- const int b= src[i+0];
- const int g= src[i+1];
- const int r= src[i+2];
-
- ((uint16_t *)dst)[i]= (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
- }
+#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
+ rgb32to16_C(src, dst, src_size);
+#endif
}
void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
{
- unsigned i,num_pixels=src_size/4;
- for(i=0; i<num_pixels; i+=4)
- {
- const int b= src[i+0];
- const int g= src[i+1];
- const int r= src[i+2];
-
- ((uint16_t *)dst)[i]= (b>>3) | ((g&0xF8)<<3) | ((r&0xF8)<<7);
- }
+#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
+ rgb32to15_C(src, dst, src_size);
+#endif
}
void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
{
- unsigned i,num_pixels=src_size/3;
- for(i=0; i<num_pixels; i+=3)
- {
- const int b= src[i+0];
- const int g= src[i+1];
- const int r= src[i+2];
-
- ((uint16_t *)dst)[i]= (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
- }
+#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
+ rgb24to16_C(src, dst, src_size);
+#endif
}
void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
{
- unsigned i,num_pixels=src_size/3;
- for(i=0; i<num_pixels; i+=3)
- {
- const int b= src[i+0];
- const int g= src[i+1];
- const int r= src[i+2];
-
- ((uint16_t *)dst)[i]= (b>>3) | ((g&0xF8)<<3) | ((r&0xF8)<<7);
- }
+#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
+ rgb24to15_C(src, dst, src_size);
+#endif
}
/**
for(i=0; i<num_pixels; i++)
((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
}
+
+void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
+{
+#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
+ rgb32tobgr32_C(src, dst, src_size);
+#endif
+}
+
/**
*
- * num_pixels must be a multiple of 16 for the MMX version
+ * 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)
*/
-void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, unsigned num_pixels)
+void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+ unsigned int width, unsigned int height,
+ unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
{
-#ifdef HAVE_MMX
- asm volatile(
- "xorl %%eax, %%eax \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"
- EMMS" \n\t"
- SFENCE
- ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (num_pixels>>1)
- : "memory", "%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;
- num_pixels>>=1;
- for(i=0; i<num_pixels; 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];
- }
+ yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
#endif
}
-void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, unsigned num_pixels)
+/**
+ *
+ * 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)
+ */
+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)
{
-#ifdef HAVE_MMX
- asm volatile(
- "xorl %%eax, %%eax \n\t"
- "pcmpeqw %%mm7, %%mm7 \n\t"
- "psrlw $8, %%mm7 \n\t" // FF,00,FF,00...
- "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)
+#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
+ yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
+#endif
+}
- MOVNTQ" %%mm0, (%3, %%eax) \n\t"
- MOVNTQ" %%mm2, (%2, %%eax) \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)
+{
+#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
+ uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
+#endif
+}
- "addl $8, %%eax \n\t"
- "cmpl %4, %%eax \n\t"
- " jb 1b \n\t"
- EMMS" \n\t"
- SFENCE
- ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (num_pixels>>1)
- : "memory", "%eax"
- );
+/**
+ *
+ * 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
- int i;
- num_pixels>>=1;
- for(i=0; i<num_pixels; 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];
- }
+ rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
#endif
-}
\ No newline at end of file
+}