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"
14 #include "../mangle.h"
17 #define CAN_COMPILE_X86_ASM
20 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
22 #ifdef CAN_COMPILE_X86_ASM
23 static const uint64_t mask32b __attribute__((aligned(8))) = 0x000000FF000000FFULL;
24 static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
25 static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
26 static const uint64_t mask32 __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
27 static const uint64_t mask24l __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
28 static const uint64_t mask24h __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
29 static const uint64_t mask24hh __attribute__((aligned(8))) = 0xffff000000000000ULL;
30 static const uint64_t mask24hhh __attribute__((aligned(8))) = 0xffffffff00000000ULL;
31 static const uint64_t mask24hhhh __attribute__((aligned(8))) = 0xffffffffffff0000ULL;
32 static const uint64_t mask15b __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
33 static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
34 static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
35 static const uint64_t red_16mask __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
36 static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL;
37 static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
38 static const uint64_t red_15mask __attribute__((aligned(8))) = 0x00007c000000f800ULL;
39 static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL;
40 static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
43 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
44 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
45 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
47 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
48 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
49 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
51 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
52 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
53 static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
56 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
57 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
58 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
59 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
61 static uint64_t __attribute__((aligned(8))) dither4[2]={
63 0x0200020002000200LL,};
65 static uint64_t __attribute__((aligned(8))) dither8[2]={
67 0x0004000400040004LL,};
71 #define RGB2YUV_SHIFT 8
72 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
73 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
74 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
75 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
76 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
77 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
78 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
79 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
80 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
82 //Note: we have C, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
88 #define RENAME(a) a ## _C
89 #include "rgb2rgb_template.c"
91 #ifdef CAN_COMPILE_X86_ASM
99 #define RENAME(a) a ## _MMX
100 #include "rgb2rgb_template.c"
108 #define RENAME(a) a ## _MMX2
109 #include "rgb2rgb_template.c"
117 #define RENAME(a) a ## _3DNow
118 #include "rgb2rgb_template.c"
120 #endif //CAN_COMPILE_X86_ASM
122 void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
124 #ifdef CAN_COMPILE_X86_ASM
125 // ordered per speed fasterst first
127 rgb24to32_MMX2(src, dst, src_size);
128 else if(gCpuCaps.has3DNow)
129 rgb24to32_3DNow(src, dst, src_size);
130 else if(gCpuCaps.hasMMX)
131 rgb24to32_MMX(src, dst, src_size);
133 rgb24to32_C(src, dst, src_size);
135 rgb24to32_C(src, dst, src_size);
139 void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
141 #ifdef CAN_COMPILE_X86_ASM
142 // ordered per speed fasterst first
144 rgb32to24_MMX2(src, dst, src_size);
145 else if(gCpuCaps.has3DNow)
146 rgb32to24_3DNow(src, dst, src_size);
147 else if(gCpuCaps.hasMMX)
148 rgb32to24_MMX(src, dst, src_size);
150 rgb32to24_C(src, dst, src_size);
152 rgb32to24_C(src, dst, src_size);
157 Original by Strepto/Astral
158 ported to gcc & bugfixed : A'rpi
159 MMX2, 3DNOW optimization by Nick Kurshev
160 32bit c version, and and&add trick by Michael Niedermayer
162 void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
164 #ifdef CAN_COMPILE_X86_ASM
165 // ordered per speed fasterst first
167 rgb15to16_MMX2(src, dst, src_size);
168 else if(gCpuCaps.has3DNow)
169 rgb15to16_3DNow(src, dst, src_size);
170 else if(gCpuCaps.hasMMX)
171 rgb15to16_MMX(src, dst, src_size);
173 rgb15to16_C(src, dst, src_size);
175 rgb15to16_C(src, dst, src_size);
180 * Pallete is assumed to contain bgr32
182 void palette8torgb32(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
185 for(i=0; i<num_pixels; i++)
186 ((unsigned *)dst)[i] = ((unsigned *)palette)[ src[i] ];
190 * Pallete is assumed to contain bgr32
192 void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
196 writes 1 byte o much and might cause alignment issues on some architectures?
197 for(i=0; i<num_pixels; i++)
198 ((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
200 for(i=0; i<num_pixels; i++)
203 dst[0]= palette[ src[i]*4+0 ];
204 dst[1]= palette[ src[i]*4+1 ];
205 dst[2]= palette[ src[i]*4+2 ];
210 void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
212 #ifdef CAN_COMPILE_X86_ASM
213 // ordered per speed fasterst first
215 rgb32to16_MMX2(src, dst, src_size);
216 else if(gCpuCaps.has3DNow)
217 rgb32to16_3DNow(src, dst, src_size);
218 else if(gCpuCaps.hasMMX)
219 rgb32to16_MMX(src, dst, src_size);
221 rgb32to16_C(src, dst, src_size);
223 rgb32to16_C(src, dst, src_size);
227 void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
229 #ifdef CAN_COMPILE_X86_ASM
230 // ordered per speed fasterst first
232 rgb32to15_MMX2(src, dst, src_size);
233 else if(gCpuCaps.has3DNow)
234 rgb32to15_3DNow(src, dst, src_size);
235 else if(gCpuCaps.hasMMX)
236 rgb32to15_MMX(src, dst, src_size);
238 rgb32to15_C(src, dst, src_size);
240 rgb32to15_C(src, dst, src_size);
244 void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
246 #ifdef CAN_COMPILE_X86_ASM
247 // ordered per speed fasterst first
249 rgb24to16_MMX2(src, dst, src_size);
250 else if(gCpuCaps.has3DNow)
251 rgb24to16_3DNow(src, dst, src_size);
252 else if(gCpuCaps.hasMMX)
253 rgb24to16_MMX(src, dst, src_size);
255 rgb24to16_C(src, dst, src_size);
257 rgb24to16_C(src, dst, src_size);
261 void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
263 #ifdef CAN_COMPILE_X86_ASM
264 // ordered per speed fasterst first
266 rgb24to15_MMX2(src, dst, src_size);
267 else if(gCpuCaps.has3DNow)
268 rgb24to15_3DNow(src, dst, src_size);
269 else if(gCpuCaps.hasMMX)
270 rgb24to15_MMX(src, dst, src_size);
272 rgb24to15_C(src, dst, src_size);
274 rgb24to15_C(src, dst, src_size);
279 * Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
281 void palette8torgb16(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
284 for(i=0; i<num_pixels; i++)
285 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
289 * Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
291 void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
294 for(i=0; i<num_pixels; i++)
295 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
298 void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
300 #ifdef CAN_COMPILE_X86_ASM
301 // ordered per speed fasterst first
303 rgb32tobgr32_MMX2(src, dst, src_size);
304 else if(gCpuCaps.has3DNow)
305 rgb32tobgr32_3DNow(src, dst, src_size);
306 else if(gCpuCaps.hasMMX)
307 rgb32tobgr32_MMX(src, dst, src_size);
309 rgb32tobgr32_C(src, dst, src_size);
311 rgb32tobgr32_C(src, dst, src_size);
317 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
318 * problem for anyone then tell me, and ill fix it)
320 void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
321 unsigned int width, unsigned int height,
322 unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
324 #ifdef CAN_COMPILE_X86_ASM
325 // ordered per speed fasterst first
327 yv12toyuy2_MMX2(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
328 else if(gCpuCaps.has3DNow)
329 yv12toyuy2_3DNow(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
330 else if(gCpuCaps.hasMMX)
331 yv12toyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
333 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
335 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
341 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
342 * problem for anyone then tell me, and ill fix it)
344 void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
345 unsigned int width, unsigned int height,
346 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
348 #ifdef CAN_COMPILE_X86_ASM
349 // ordered per speed fasterst first
351 yuy2toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
352 else if(gCpuCaps.has3DNow)
353 yuy2toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
354 else if(gCpuCaps.hasMMX)
355 yuy2toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
357 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
359 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
365 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
366 * problem for anyone then tell me, and ill fix it)
367 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
369 void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
370 unsigned int width, unsigned int height,
371 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
373 #ifdef CAN_COMPILE_X86_ASM
374 // ordered per speed fasterst first
376 uyvytoyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
377 else if(gCpuCaps.has3DNow)
378 uyvytoyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
379 else if(gCpuCaps.hasMMX)
380 uyvytoyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
382 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
384 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
390 * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
391 * problem for anyone then tell me, and ill fix it)
392 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
394 void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
395 unsigned int width, unsigned int height,
396 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
398 #ifdef CAN_COMPILE_X86_ASM
399 // ordered per speed fasterst first
401 rgb24toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
402 else if(gCpuCaps.has3DNow)
403 rgb24toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
404 else if(gCpuCaps.hasMMX)
405 rgb24toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
407 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
409 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
413 void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
414 int width, int height, int src1Stride, int src2Stride, int dstStride)
416 #ifdef CAN_COMPILE_X86_ASM
417 // ordered per speed fasterst first
419 interleaveBytes_MMX2(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
420 else if(gCpuCaps.has3DNow)
421 interleaveBytes_3DNow(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
422 else if(gCpuCaps.hasMMX)
423 interleaveBytes_MMX(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
425 interleaveBytes_C(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);
427 interleaveBytes_C(src1, src2, dst, width, height, src1Stride, src2Stride, dstStride);