3 * rgb2rgb.c, Software RGB to RGB convertor
4 * pluralize by Software PAL8 to RGB convertor
5 * Software YUV to YUV convertor
6 * Software YUV to RGB convertor
7 * Written by Nick Kurshev.
8 * palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
11 #include "../config.h"
13 #include "../cpudetect.h"
16 #define CAN_COMPILE_X86_ASM
19 #ifdef CAN_COMPILE_X86_ASM
20 static const uint64_t mask32b __attribute__((aligned(8))) = 0x000000FF000000FFULL;
21 static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
22 static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
23 static const uint64_t mask32 __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
24 static const uint64_t mask24l __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
25 static const uint64_t mask24h __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
26 static const uint64_t mask24hh __attribute__((aligned(8))) = 0xffff000000000000ULL;
27 static const uint64_t mask24hhh __attribute__((aligned(8))) = 0xffffffff00000000ULL;
28 static const uint64_t mask24hhhh __attribute__((aligned(8))) = 0xffffffffffff0000ULL;
29 static const uint64_t mask15b __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
30 static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
31 static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
32 static const uint64_t red_16mask __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
33 static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL;
34 static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
35 static const uint64_t red_15mask __attribute__((aligned(8))) = 0x00007c000000f800ULL;
36 static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL;
37 static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
39 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
40 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
41 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
42 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
44 static uint64_t __attribute__((aligned(8))) dither4[2]={
46 0x0200020002000200LL,};
48 static uint64_t __attribute__((aligned(8))) dither8[2]={
50 0x0004000400040004LL,};
54 #define RGB2YUV_SHIFT 8
55 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
56 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
57 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
58 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
59 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
60 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
61 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
62 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
63 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
65 //Note: we have C, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
71 #define RENAME(a) a ## _C
72 #include "rgb2rgb_template.c"
74 #ifdef CAN_COMPILE_X86_ASM
82 #define RENAME(a) a ## _MMX
83 #include "rgb2rgb_template.c"
91 #define RENAME(a) a ## _MMX2
92 #include "rgb2rgb_template.c"
100 #define RENAME(a) a ## _3DNow
101 #include "rgb2rgb_template.c"
103 #endif //CAN_COMPILE_X86_ASM
105 void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
107 #ifdef CAN_COMPILE_X86_ASM
108 // ordered per speed fasterst first
110 rgb24to32_MMX2(src, dst, src_size);
111 else if(gCpuCaps.has3DNow)
112 rgb24to32_3DNow(src, dst, src_size);
113 else if(gCpuCaps.hasMMX)
114 rgb24to32_MMX(src, dst, src_size);
116 rgb24to32_C(src, dst, src_size);
118 rgb24to32_C(src, dst, src_size);
122 void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
124 #ifdef CAN_COMPILE_X86_ASM
125 // ordered per speed fasterst first
127 rgb32to24_MMX2(src, dst, src_size);
128 else if(gCpuCaps.has3DNow)
129 rgb32to24_3DNow(src, dst, src_size);
130 else if(gCpuCaps.hasMMX)
131 rgb32to24_MMX(src, dst, src_size);
133 rgb32to24_C(src, dst, src_size);
135 rgb32to24_C(src, dst, src_size);
140 Original by Strepto/Astral
141 ported to gcc & bugfixed : A'rpi
142 MMX2, 3DNOW optimization by Nick Kurshev
143 32bit c version, and and&add trick by Michael Niedermayer
145 void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
147 #ifdef CAN_COMPILE_X86_ASM
148 // ordered per speed fasterst first
150 rgb15to16_MMX2(src, dst, src_size);
151 else if(gCpuCaps.has3DNow)
152 rgb15to16_3DNow(src, dst, src_size);
153 else if(gCpuCaps.hasMMX)
154 rgb15to16_MMX(src, dst, src_size);
156 rgb15to16_C(src, dst, src_size);
158 rgb15to16_C(src, dst, src_size);
163 * Pallete is assumed to contain bgr32
165 void palette8torgb32(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
168 for(i=0; i<num_pixels; i++)
169 ((unsigned *)dst)[i] = ((unsigned *)palette)[ src[i] ];
173 * Pallete is assumed to contain bgr32
175 void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
179 writes 1 byte o much and might cause alignment issues on some architectures?
180 for(i=0; i<num_pixels; i++)
181 ((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
183 for(i=0; i<num_pixels; i++)
186 dst[0]= palette[ src[i]*4+0 ];
187 dst[1]= palette[ src[i]*4+1 ];
188 dst[2]= palette[ src[i]*4+2 ];
193 void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
195 #ifdef CAN_COMPILE_X86_ASM
196 // ordered per speed fasterst first
198 rgb32to16_MMX2(src, dst, src_size);
199 else if(gCpuCaps.has3DNow)
200 rgb32to16_3DNow(src, dst, src_size);
201 else if(gCpuCaps.hasMMX)
202 rgb32to16_MMX(src, dst, src_size);
204 rgb32to16_C(src, dst, src_size);
206 rgb32to16_C(src, dst, src_size);
210 void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
212 #ifdef CAN_COMPILE_X86_ASM
213 // ordered per speed fasterst first
215 rgb32to15_MMX2(src, dst, src_size);
216 else if(gCpuCaps.has3DNow)
217 rgb32to15_3DNow(src, dst, src_size);
218 else if(gCpuCaps.hasMMX)
219 rgb32to15_MMX(src, dst, src_size);
221 rgb32to15_C(src, dst, src_size);
223 rgb32to15_C(src, dst, src_size);
227 void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
229 #ifdef CAN_COMPILE_X86_ASM
230 // ordered per speed fasterst first
232 rgb24to16_MMX2(src, dst, src_size);
233 else if(gCpuCaps.has3DNow)
234 rgb24to16_3DNow(src, dst, src_size);
235 else if(gCpuCaps.hasMMX)
236 rgb24to16_MMX(src, dst, src_size);
238 rgb24to16_C(src, dst, src_size);
240 rgb24to16_C(src, dst, src_size);
244 void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
246 #ifdef CAN_COMPILE_X86_ASM
247 // ordered per speed fasterst first
249 rgb24to15_MMX2(src, dst, src_size);
250 else if(gCpuCaps.has3DNow)
251 rgb24to15_3DNow(src, dst, src_size);
252 else if(gCpuCaps.hasMMX)
253 rgb24to15_MMX(src, dst, src_size);
255 rgb24to15_C(src, dst, src_size);
257 rgb24to15_C(src, dst, src_size);
262 * Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
264 void palette8torgb16(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
267 for(i=0; i<num_pixels; i++)
268 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
272 * Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
274 void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
277 for(i=0; i<num_pixels; i++)
278 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
281 void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
283 #ifdef CAN_COMPILE_X86_ASM
284 // ordered per speed fasterst first
286 rgb32tobgr32_MMX2(src, dst, src_size);
287 else if(gCpuCaps.has3DNow)
288 rgb32tobgr32_3DNow(src, dst, src_size);
289 else if(gCpuCaps.hasMMX)
290 rgb32tobgr32_MMX(src, dst, src_size);
292 rgb32tobgr32_C(src, dst, src_size);
294 rgb32tobgr32_C(src, dst, src_size);
300 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
301 * problem for anyone then tell me, and ill fix it)
303 void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
304 unsigned int width, unsigned int height,
305 unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
307 #ifdef CAN_COMPILE_X86_ASM
308 // ordered per speed fasterst first
310 yv12toyuy2_MMX2(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
311 else if(gCpuCaps.has3DNow)
312 yv12toyuy2_3DNow(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
313 else if(gCpuCaps.hasMMX)
314 yv12toyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
316 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
318 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
324 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
325 * problem for anyone then tell me, and ill fix it)
327 void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
328 unsigned int width, unsigned int height,
329 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
331 #ifdef CAN_COMPILE_X86_ASM
332 // ordered per speed fasterst first
334 yuy2toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
335 else if(gCpuCaps.has3DNow)
336 yuy2toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
337 else if(gCpuCaps.hasMMX)
338 yuy2toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
340 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
342 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
348 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
349 * problem for anyone then tell me, and ill fix it)
350 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
352 void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
353 unsigned int width, unsigned int height,
354 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
356 #ifdef CAN_COMPILE_X86_ASM
357 // ordered per speed fasterst first
359 uyvytoyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
360 else if(gCpuCaps.has3DNow)
361 uyvytoyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
362 else if(gCpuCaps.hasMMX)
363 uyvytoyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
365 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
367 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
373 * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
374 * problem for anyone then tell me, and ill fix it)
375 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
377 void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
378 unsigned int width, unsigned int height,
379 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
381 #ifdef CAN_COMPILE_X86_ASM
382 // ordered per speed fasterst first
384 rgb24toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
385 else if(gCpuCaps.has3DNow)
386 rgb24toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
387 else if(gCpuCaps.hasMMX)
388 rgb24toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
390 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
392 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);