+/*
+ *
+ * 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 & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
+ */
#include <inttypes.h>
#include "../config.h"
#include "rgb2rgb.h"
-#include "mmx.h"
+#include "../cpudetect.h"
-#ifdef HAVE_3DNOW
-#define PREFETCH "prefetch"
-#define PREFETCHW "prefetchw"
-#elif HAVE_MMX2
-#define PREFETCH "prefetchnta"
-#define PREFETCHW "prefetcht0"
+#ifdef ARCH_X86
+#define CAN_COMPILE_X86_ASM
#endif
-#ifdef HAVE_3DNOW
-#define EMMS "femms"
+#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
-#define EMMS "emms"
+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"
-#ifdef HAVE_MMX2
-#define MOVNTQ "movntq"
+//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)
+{
+#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
+}
+
+void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#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
+}
+
+/*
+ Original by Strepto/Astral
+ ported to gcc & bugfixed : A'rpi
+ MMX2, 3DNOW optimization by Nick Kurshev
+ 32bit c version, and and&add trick by Michael Niedermayer
+*/
+void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
+{
+#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
+ rgb15to16_C(src, dst, src_size);
+#endif
+}
+
+/**
+ * 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 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)
+{
+#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
-#define MOVNTQ "movq"
-#endif
-
-void rgb24to32(uint8_t *src,uint8_t *dst,uint32_t src_size)
-{
- uint8_t *dest = dst;
- uint8_t *s = src;
- uint8_t *end;
-#ifdef HAVE_MMX
- const uint64_t mask32 = 0x00FFFFFF00FFFFFFULL;
- uint8_t *mm_end;
-#endif
- end = s + src_size;
-#ifdef HAVE_MMX
-#ifdef PREFETCH
- __asm __volatile(
- PREFETCH" %0\n\t"
- PREFETCH" 64%0\n\t"
- PREFETCHW" %1\n\t"
- PREFETCHW" 64%1\n\t"::"m"(*s),"m"(*dest):"memory");
-#endif
- mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
- __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
- while(s < mm_end)
- {
-#ifdef PREFETCH
- __asm __volatile(
- PREFETCH" 128%0\n\t"
- PREFETCHW" 128%1"
- ::"m"(*s),"m"(*dest):"memory");
-#endif
- __asm __volatile(
- "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(EMMS:::"memory");
-#endif
- while(s < end)
- {
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = 0;
- }
-}
-
-/* TODO: MMX optimization */
-void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
-{
- uint8_t *dest = dst;
- uint8_t *s = src;
- uint8_t *end;
- end = s + src_size;
- while(s < end)
- {
- *dest++ = *s++;
- *dest++ = *s++;
- *dest++ = *s++;
- s++;
- }
-}
-
-/* Original by Strepto/Astral
- ported to gcc & bugfixed : A'rpi */
-void rgb15to16(uint8_t *src,uint8_t *dst,uint32_t src_size)
-{
-#ifdef HAVE_MMX
- static uint64_t mask_b = 0x001F001F001F001FLL; // 00000000 00011111 xxB
- static uint64_t mask_rg = 0x7FE07FE07FE07FE0LL; // 01111111 11100000 RGx
- register char* s=src+src_size;
- register char* d=dst+src_size;
- register int offs=-src_size;
- movq_m2r (mask_b, mm4);
- movq_m2r (mask_rg, mm5);
- while(offs<0){
- movq_m2r (*(s+offs), mm0);
- movq_r2r (mm0, mm1);
-
- movq_m2r (*(s+8+offs), mm2);
- movq_r2r (mm2, mm3);
-
- pand_r2r (mm4, mm0);
- pand_r2r (mm5, mm1);
-
- psllq_i2r(1,mm1);
- pand_r2r (mm4, mm2);
-
- pand_r2r (mm5, mm3);
- por_r2r (mm1, mm0);
-
- psllq_i2r(1,mm3);
- movq_r2m (mm0,*(d+offs));
-
- por_r2r (mm3,mm2);
- movq_r2m (mm2,*(d+8+offs));
-
- offs+=16;
- }
- emms();
+ rgb32to15_C(src, dst, src_size);
+#endif
+}
+
+void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
+{
+#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)
+{
+#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
+}
+
+/**
+ * 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)
+{
+#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
+}
+
+/**
+ *
+ * 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 int width, unsigned int height,
+ unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
+{
+#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
+ yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
+#endif
+}
+
+/**
+ *
+ * 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 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
+}
+
+/**
+ *
+ * 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
+}
+
+/**
+ *
+ * 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
- 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 );
- }
+ rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
#endif
}