Skip to content

L
libswscale
Commit cb389e29 authored by michael's avatar michael
Browse files
Options

runtime cpu detection 

rgb24toyv12 in C


git-svn-id: file:///var/local/repositories/mplayer/trunk/postproc@3133 b3059339-0415-0410-9bf9-f77b7e298cf2
parent 86a0253c
Hide whitespace changes
Inline Side-by-side
Showing with 320 additions and 857 deletions
rgb2rgb.c
View file @ cb389e29
...@@ -5,14 +5,18 @@ ...@@ -5,14 +5,18 @@
* Software YUV to YUV convertor * Software YUV to YUV convertor
* Software YUV to RGB convertor * Software YUV to RGB convertor
* Written by Nick Kurshev. * 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 <inttypes.h>
#include "../config.h" #include "../config.h"
#include "rgb2rgb.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 mask32b __attribute__((aligned(8))) = 0x000000FF000000FFULL;
static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL; static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL; static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
...@@ -47,133 +51,89 @@ static uint64_t __attribute__((aligned(8))) dither8[2]={ ...@@ -47,133 +51,89 @@ static uint64_t __attribute__((aligned(8))) dither8[2]={
#endif #endif
#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) void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
uint8_t *dest = dst; #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end; if(gCpuCaps.hasMMX2)
#ifdef HAVE_MMX rgb24to32_MMX2(src, dst, src_size);
uint8_t *mm_end; else if(gCpuCaps.has3DNow)
#endif rgb24to32_3DNow(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.hasMMX)
#ifdef HAVE_MMX rgb24to32_MMX(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*s):"memory"); else
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4)); rgb24to32_C(src, dst, src_size);
__asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory"); #else
if(mm_end == end) mm_end -= MMREG_SIZE*4; rgb24to32_C(src, dst, src_size);
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");
#endif #endif
while(s < end)
{
*dest++ = *s++;
*dest++ = *s++;
*dest++ = *s++;
*dest++ = 0;
}
} }
void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size) void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
uint8_t *dest = dst; #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end; if(gCpuCaps.hasMMX2)
#ifdef HAVE_MMX rgb32to24_MMX2(src, dst, src_size);
uint8_t *mm_end; else if(gCpuCaps.has3DNow)
#endif rgb32to24_3DNow(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.hasMMX)
#ifdef HAVE_MMX rgb32to24_MMX(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*s):"memory"); else
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*4))*(MMREG_SIZE*4)); rgb32to24_C(src, dst, src_size);
while(s < mm_end) #else
{ rgb32to24_C(src, dst, src_size);
__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");
#endif #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) ...@@ -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) void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef CAN_COMPILE_X86_ASM
register const char* s=src+src_size; // ordered per speed fasterst first
register char* d=dst+src_size; if(gCpuCaps.hasMMX2)
register int offs=-src_size; rgb15to16_MMX2(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*(s+offs))); else if(gCpuCaps.has3DNow)
__asm __volatile( rgb15to16_3DNow(src, dst, src_size);
"movq %0, %%mm4\n\t" else if(gCpuCaps.hasMMX)
::"m"(mask15s)); rgb15to16_MMX(src, dst, src_size);
while(offs<0) else
{ rgb15to16_C(src, dst, src_size);
__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 );
}
#else #else
const unsigned *s1=( unsigned * )src; rgb15to16_C(src, dst, src_size);
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 #endif
} }
...@@ -275,287 +192,69 @@ void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons ...@@ -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) void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end,*mm_end; if(gCpuCaps.hasMMX2)
uint16_t *d = (uint16_t *)dst; rgb32to16_MMX2(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.has3DNow)
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); rgb32to16_3DNow(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); else if(gCpuCaps.hasMMX)
__asm __volatile( rgb32to16_MMX(src, dst, src_size);
"movq %0, %%mm7\n\t" else
"movq %1, %%mm6\n\t" rgb32to16_C(src, dst, src_size);
::"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");
#else #else
unsigned j,i,num_pixels=src_size/4; rgb32to16_C(src, dst, src_size);
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 #endif
} }
void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size) void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end,*mm_end; if(gCpuCaps.hasMMX2)
uint16_t *d = (uint16_t *)dst; rgb32to15_MMX2(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.has3DNow)
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); rgb32to15_3DNow(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); else if(gCpuCaps.hasMMX)
__asm __volatile( rgb32to15_MMX(src, dst, src_size);
"movq %0, %%mm7\n\t" else
"movq %1, %%mm6\n\t" rgb32to15_C(src, dst, src_size);
::"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");
#else #else
unsigned j,i,num_pixels=src_size/4; rgb32to15_C(src, dst, src_size);
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 #endif
} }
void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size) void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end,*mm_end; if(gCpuCaps.hasMMX2)
uint16_t *d = (uint16_t *)dst; rgb24to16_MMX2(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.has3DNow)
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); rgb24to16_3DNow(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); else if(gCpuCaps.hasMMX)
__asm __volatile( rgb24to16_MMX(src, dst, src_size);
"movq %0, %%mm7\n\t" else
"movq %1, %%mm6\n\t" rgb24to16_C(src, dst, src_size);
::"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");
#else #else
unsigned j,i,num_pixels=src_size/3; rgb24to16_C(src, dst, src_size);
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 #endif
} }
void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size) void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef CAN_COMPILE_X86_ASM
const uint8_t *s = src; // ordered per speed fasterst first
const uint8_t *end,*mm_end; if(gCpuCaps.hasMMX2)
uint16_t *d = (uint16_t *)dst; rgb24to15_MMX2(src, dst, src_size);
end = s + src_size; else if(gCpuCaps.has3DNow)
mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2)); rgb24to15_3DNow(src, dst, src_size);
__asm __volatile(PREFETCH" %0"::"m"(*src):"memory"); else if(gCpuCaps.hasMMX)
__asm __volatile( rgb24to15_MMX(src, dst, src_size);
"movq %0, %%mm7\n\t" else
"movq %1, %%mm6\n\t" rgb24to15_C(src, dst, src_size);
::"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");
#else #else
unsigned j,i,num_pixels=src_size/3; rgb24to15_C(src, dst, src_size);
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);
}
#endif #endif
} }
...@@ -581,41 +280,18 @@ void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons ...@@ -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) void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
{ {
int num_pixels= src_size >> 2; #ifdef CAN_COMPILE_X86_ASM
#ifdef HAVE_MMX // ordered per speed fasterst first
asm volatile ( if(gCpuCaps.hasMMX2)
"xorl %%eax, %%eax \n\t" rgb32tobgr32_MMX2(src, dst, src_size);
".balign 16 \n\t" else if(gCpuCaps.has3DNow)
"1: \n\t" rgb32tobgr32_3DNow(src, dst, src_size);
PREFETCH" 32(%0, %%eax) \n\t" else if(gCpuCaps.hasMMX)
"movq (%0, %%eax), %%mm0 \n\t" rgb32tobgr32_MMX(src, dst, src_size);
"movq %%mm0, %%mm1 \n\t" else
"movq %%mm0, %%mm2 \n\t" rgb32tobgr32_C(src, dst, src_size);
"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");
#else #else
int i; rgb32tobgr32_C(src, dst, src_size);
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];
}
#endif #endif
} }
...@@ -628,67 +304,18 @@ void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, u ...@@ -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 width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int dstStride) unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
{ {
int y; #ifdef CAN_COMPILE_X86_ASM
const int chromWidth= width>>1; // ordered per speed fasterst first
for(y=0; y<height; y++) if(gCpuCaps.hasMMX2)
{ yv12toyuy2_MMX2(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
#ifdef HAVE_MMX else if(gCpuCaps.has3DNow)
//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway) yv12toyuy2_3DNow(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
asm volatile( else if(gCpuCaps.hasMMX)
"xorl %%eax, %%eax \n\t" yv12toyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
".balign 16 \n\t" else
"1: \n\t" yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
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"
);
#else #else
int i; yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
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");
#endif #endif
} }
...@@ -701,119 +328,18 @@ void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, ...@@ -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 width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int srcStride) unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
{ {
int y; #ifdef CAN_COMPILE_X86_ASM
const int chromWidth= width>>1; // ordered per speed fasterst first
for(y=0; y<height; y+=2) if(gCpuCaps.hasMMX2)
{ yuy2toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
#ifdef HAVE_MMX else if(gCpuCaps.has3DNow)
asm volatile( yuy2toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"xorl %%eax, %%eax \n\t" else if(gCpuCaps.hasMMX)
"pcmpeqw %%mm7, %%mm7 \n\t" yuy2toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"psrlw $8, %%mm7 \n\t" // FF,00,FF,00... else
".balign 16 \n\t" yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"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"
);
#else #else
int i; yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
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");
#endif #endif
} }
...@@ -821,125 +347,48 @@ asm volatile( EMMS" \n\t" ...@@ -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 * 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) * 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, void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
unsigned int width, unsigned int height, unsigned int width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int srcStride) unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
{ {
int y; #ifdef CAN_COMPILE_X86_ASM
const int chromWidth= width>>1; // ordered per speed fasterst first
for(y=0; y<height; y+=2) if(gCpuCaps.hasMMX2)
{ uyvytoyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
#ifdef HAVE_MMX else if(gCpuCaps.has3DNow)
asm volatile( uyvytoyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"xorl %%eax, %%eax \n\t" else if(gCpuCaps.hasMMX)
"pcmpeqw %%mm7, %%mm7 \n\t" uyvytoyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"psrlw $8, %%mm7 \n\t" // FF,00,FF,00... else
".balign 16 \n\t" uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
"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"
);
#else #else
int i; uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
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");
#endif #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
}
rgb2rgb_template.c
View file @ cb389e29
...@@ -5,49 +5,52 @@ ...@@ -5,49 +5,52 @@
* Software YUV to YUV convertor * Software YUV to YUV convertor
* Software YUV to RGB convertor * Software YUV to RGB convertor
* Written by Nick Kurshev. * 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 #undef PREFETCH
static const uint64_t mask32b __attribute__((aligned(8))) = 0x000000FF000000FFULL; #undef MOVNTQ
static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL; #undef EMMS
static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL; #undef SFENCE
static const uint64_t mask32 __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL; #undef MMREG_SIZE
static const uint64_t mask24l __attribute__((aligned(8))) = 0x0000000000FFFFFFULL; #undef PREFETCHW
static const uint64_t mask24h __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL; #undef PAVGB
static const uint64_t mask24hh __attribute__((aligned(8))) = 0xffff000000000000ULL;
static const uint64_t mask24hhh __attribute__((aligned(8))) = 0xffffffff00000000ULL; #ifdef HAVE_SSE2
static const uint64_t mask24hhhh __attribute__((aligned(8))) = 0xffffffffffff0000ULL; #define MMREG_SIZE 16
static const uint64_t mask15b __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */ #else
static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */ #define MMREG_SIZE 8
static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL; #endif
static const uint64_t red_16mask __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL; #ifdef HAVE_3DNOW
static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL; #define PREFETCH "prefetch"
static const uint64_t red_15mask __attribute__((aligned(8))) = 0x00007c000000f800ULL; #define PREFETCHW "prefetchw"
static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL; #define PAVGB "pavgusb"
static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL; #elif defined ( HAVE_MMX2 )
#if 0 #define PREFETCH "prefetchnta"
static volatile uint64_t __attribute__((aligned(8))) b5Dither; #define PREFETCHW "prefetcht0"
static volatile uint64_t __attribute__((aligned(8))) g5Dither; #define PAVGB "pavgb"
static volatile uint64_t __attribute__((aligned(8))) g6Dither; #else
static volatile uint64_t __attribute__((aligned(8))) r5Dither; #define PREFETCH "/nop"
#define PREFETCHW "/nop"
static uint64_t __attribute__((aligned(8))) dither4[2]={
0x0103010301030103LL,
0x0200020002000200LL,};
static uint64_t __attribute__((aligned(8))) dither8[2]={
0x0602060206020602LL,
0x0004000400040004LL,};
#endif #endif
#ifdef HAVE_3DNOW
/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
#define EMMS "femms"
#else
#define EMMS "emms"
#endif #endif
void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size) #ifdef HAVE_MMX2
#define MOVNTQ "movntq"
#define SFENCE "sfence"
#else
#define MOVNTQ "movq"
#define SFENCE "/nop"
#endif
static inline void RENAME(rgb24to32)(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
uint8_t *dest = dst; uint8_t *dest = dst;
const uint8_t *s = src; const uint8_t *s = src;
...@@ -99,7 +102,7 @@ void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size) ...@@ -99,7 +102,7 @@ void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
} }
} }
void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size) static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
uint8_t *dest = dst; uint8_t *dest = dst;
const uint8_t *s = src; const uint8_t *s = src;
...@@ -153,7 +156,7 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size) ...@@ -153,7 +156,7 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
"por %%mm3, %%mm1\n\t" "por %%mm3, %%mm1\n\t"
"pand %6, %%mm5\n\t" "pand %6, %%mm5\n\t"
"por %%mm5, %%mm4\n\t" "por %%mm5, %%mm4\n\t"
MOVNTQ" %%mm0, %0\n\t" MOVNTQ" %%mm0, %0\n\t"
MOVNTQ" %%mm1, 8%0\n\t" MOVNTQ" %%mm1, 8%0\n\t"
MOVNTQ" %%mm4, 16%0" MOVNTQ" %%mm4, 16%0"
...@@ -182,7 +185,7 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size) ...@@ -182,7 +185,7 @@ void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
MMX2, 3DNOW optimization by Nick Kurshev MMX2, 3DNOW optimization by Nick Kurshev
32bit c version, and and&add trick by Michael Niedermayer 32bit c version, and and&add trick by Michael Niedermayer
*/ */
void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size) static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef HAVE_MMX
register const char* s=src+src_size; register const char* s=src+src_size;
...@@ -242,38 +245,7 @@ void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size) ...@@ -242,38 +245,7 @@ void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
#endif #endif
} }
/** static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)
* Pallete is assumed to contain bgr32
*/
void palette8torgb32(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
{
unsigned i;
for(i=0; i<num_pixels; i++)
((unsigned *)dst)[i] = ((unsigned *)palette)[ src[i] ];
}
/**
* Pallete is assumed to contain bgr32
*/
void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
{
unsigned i;
/*
writes 1 byte o much and might cause alignment issues on some architectures?
for(i=0; i<num_pixels; i++)
((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
*/
for(i=0; i<num_pixels; i++)
{
//FIXME slow?
dst[0]= palette[ src[i]*4+0 ];
dst[1]= palette[ src[i]*4+1 ];
dst[2]= palette[ src[i]*4+2 ];
dst+= 3;
}
}
void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef HAVE_MMX
const uint8_t *s = src; const uint8_t *s = src;
...@@ -344,7 +316,7 @@ void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size) ...@@ -344,7 +316,7 @@ void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
#endif #endif
} }
void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size) static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef HAVE_MMX
const uint8_t *s = src; const uint8_t *s = src;
...@@ -415,7 +387,7 @@ void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size) ...@@ -415,7 +387,7 @@ void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
#endif #endif
} }
void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size) static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef HAVE_MMX
const uint8_t *s = src; const uint8_t *s = src;
...@@ -487,7 +459,7 @@ void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size) ...@@ -487,7 +459,7 @@ void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
#endif #endif
} }
void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size) static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)
{ {
#ifdef HAVE_MMX #ifdef HAVE_MMX
const uint8_t *s = src; const uint8_t *s = src;
...@@ -559,27 +531,7 @@ void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size) ...@@ -559,27 +531,7 @@ void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
#endif #endif
} }
/** static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, unsigned int src_size)
* Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
*/
void palette8torgb16(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
{
unsigned i;
for(i=0; i<num_pixels; i++)
((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
}
/**
* Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
*/
void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
{
unsigned i;
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)
{ {
int num_pixels= src_size >> 2; int num_pixels= src_size >> 2;
#ifdef HAVE_MMX #ifdef HAVE_MMX
...@@ -624,7 +576,7 @@ void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size) ...@@ -624,7 +576,7 @@ void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
* height should be a multiple of 2 and width should be a multiple of 16 (if this is a * 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) * 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, static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
unsigned int width, unsigned int height, unsigned int width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int dstStride) unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
{ {
...@@ -697,7 +649,7 @@ asm( EMMS" \n\t" ...@@ -697,7 +649,7 @@ asm( EMMS" \n\t"
* height should be a multiple of 2 and width should be a multiple of 16 (if this is a * 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) * 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, static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
unsigned int width, unsigned int height, unsigned int width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int srcStride) unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
{ {
...@@ -821,8 +773,9 @@ asm volatile( EMMS" \n\t" ...@@ -821,8 +773,9 @@ asm volatile( EMMS" \n\t"
* *
* height should be a multiple of 2 and width should be a multiple of 16 (if this is a * 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) * 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, static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
unsigned int width, unsigned int height, unsigned int width, unsigned int height,
unsigned int lumStride, unsigned int chromStride, unsigned int srcStride) unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
{ {
...@@ -942,4 +895,65 @@ asm volatile( EMMS" \n\t" ...@@ -942,4 +895,65 @@ asm volatile( EMMS" \n\t"
#endif #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
*/
static inline void RENAME(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)
{
int y;
const int chromWidth= width>>1;
for(y=0; y<height; y+=2)
{
int i;
for(i=0; i<chromWidth; i++)
{
unsigned int b= src[6*i+0];
unsigned int g= src[6*i+1];
unsigned int r= src[6*i+2];
unsigned int Y = RY*r + GY*g + BY*b + 16;
unsigned int V = RV*r + GV*g + BV*b + 128;
unsigned int U = RU*r + GU*g + BU*b + 128;
udst[i] = U;
vdst[i] = V;
ydst[2*i] = Y;
b= src[6*i+3];
g= src[6*i+4];
r= src[6*i+5];
Y = RY*r + GY*g + BY*b + 16;
ydst[2*i+1] = Y;
}
ydst += lumStride;
src += srcStride;
for(i=0; i<chromWidth; i++)
{
unsigned int b= src[6*i+0];
unsigned int g= src[6*i+1];
unsigned int r= src[6*i+2];
unsigned int Y = RY*r + GY*g + BY*b + 16;
ydst[2*i] = Y;
b= src[6*i+3];
g= src[6*i+4];
r= src[6*i+5];
Y = RY*r + GY*g + BY*b + 16;
ydst[2*i+1] = Y;
}
udst += chromStride;
vdst += chromStride;
ydst += lumStride;
src += srcStride;
}
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment