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 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
21 #ifdef CAN_COMPILE_X86_ASM
22 static const uint64_t mask32b __attribute__((aligned(8))) = 0x000000FF000000FFULL;
23 static const uint64_t mask32g __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
24 static const uint64_t mask32r __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
25 static const uint64_t mask32 __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
26 static const uint64_t mask24l __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
27 static const uint64_t mask24h __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
28 static const uint64_t mask24hh __attribute__((aligned(8))) = 0xffff000000000000ULL;
29 static const uint64_t mask24hhh __attribute__((aligned(8))) = 0xffffffff00000000ULL;
30 static const uint64_t mask24hhhh __attribute__((aligned(8))) = 0xffffffffffff0000ULL;
31 static const uint64_t mask15b __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
32 static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
33 static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
34 static const uint64_t red_16mask __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
35 static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL;
36 static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
37 static const uint64_t red_15mask __attribute__((aligned(8))) = 0x00007c000000f800ULL;
38 static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL;
39 static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
42 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
43 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
44 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
46 static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
47 static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
48 static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
50 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
51 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
52 static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
55 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
56 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
57 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
58 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
60 static uint64_t __attribute__((aligned(8))) dither4[2]={
62 0x0200020002000200LL,};
64 static uint64_t __attribute__((aligned(8))) dither8[2]={
66 0x0004000400040004LL,};
70 #define RGB2YUV_SHIFT 8
71 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
72 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
73 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
74 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
75 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
76 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
77 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
78 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
79 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
81 //Note: we have C, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
87 #define RENAME(a) a ## _C
88 #include "rgb2rgb_template.c"
90 #ifdef CAN_COMPILE_X86_ASM
98 #define RENAME(a) a ## _MMX
99 #include "rgb2rgb_template.c"
107 #define RENAME(a) a ## _MMX2
108 #include "rgb2rgb_template.c"
116 #define RENAME(a) a ## _3DNow
117 #include "rgb2rgb_template.c"
119 #endif //CAN_COMPILE_X86_ASM
121 void rgb24to32(const uint8_t *src,uint8_t *dst,unsigned src_size)
123 #ifdef CAN_COMPILE_X86_ASM
124 // ordered per speed fasterst first
126 rgb24to32_MMX2(src, dst, src_size);
127 else if(gCpuCaps.has3DNow)
128 rgb24to32_3DNow(src, dst, src_size);
129 else if(gCpuCaps.hasMMX)
130 rgb24to32_MMX(src, dst, src_size);
132 rgb24to32_C(src, dst, src_size);
134 rgb24to32_C(src, dst, src_size);
138 void rgb32to24(const uint8_t *src,uint8_t *dst,unsigned src_size)
140 #ifdef CAN_COMPILE_X86_ASM
141 // ordered per speed fasterst first
143 rgb32to24_MMX2(src, dst, src_size);
144 else if(gCpuCaps.has3DNow)
145 rgb32to24_3DNow(src, dst, src_size);
146 else if(gCpuCaps.hasMMX)
147 rgb32to24_MMX(src, dst, src_size);
149 rgb32to24_C(src, dst, src_size);
151 rgb32to24_C(src, dst, src_size);
156 Original by Strepto/Astral
157 ported to gcc & bugfixed : A'rpi
158 MMX2, 3DNOW optimization by Nick Kurshev
159 32bit c version, and and&add trick by Michael Niedermayer
161 void rgb15to16(const uint8_t *src,uint8_t *dst,unsigned src_size)
163 #ifdef CAN_COMPILE_X86_ASM
164 // ordered per speed fasterst first
166 rgb15to16_MMX2(src, dst, src_size);
167 else if(gCpuCaps.has3DNow)
168 rgb15to16_3DNow(src, dst, src_size);
169 else if(gCpuCaps.hasMMX)
170 rgb15to16_MMX(src, dst, src_size);
172 rgb15to16_C(src, dst, src_size);
174 rgb15to16_C(src, dst, src_size);
179 * Pallete is assumed to contain bgr32
181 void palette8torgb32(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
184 for(i=0; i<num_pixels; i++)
185 ((unsigned *)dst)[i] = ((unsigned *)palette)[ src[i] ];
189 * Pallete is assumed to contain bgr32
191 void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
195 writes 1 byte o much and might cause alignment issues on some architectures?
196 for(i=0; i<num_pixels; i++)
197 ((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
199 for(i=0; i<num_pixels; i++)
202 dst[0]= palette[ src[i]*4+0 ];
203 dst[1]= palette[ src[i]*4+1 ];
204 dst[2]= palette[ src[i]*4+2 ];
209 void rgb32to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
211 #ifdef CAN_COMPILE_X86_ASM
212 // ordered per speed fasterst first
214 rgb32to16_MMX2(src, dst, src_size);
215 else if(gCpuCaps.has3DNow)
216 rgb32to16_3DNow(src, dst, src_size);
217 else if(gCpuCaps.hasMMX)
218 rgb32to16_MMX(src, dst, src_size);
220 rgb32to16_C(src, dst, src_size);
222 rgb32to16_C(src, dst, src_size);
226 void rgb32to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
228 #ifdef CAN_COMPILE_X86_ASM
229 // ordered per speed fasterst first
231 rgb32to15_MMX2(src, dst, src_size);
232 else if(gCpuCaps.has3DNow)
233 rgb32to15_3DNow(src, dst, src_size);
234 else if(gCpuCaps.hasMMX)
235 rgb32to15_MMX(src, dst, src_size);
237 rgb32to15_C(src, dst, src_size);
239 rgb32to15_C(src, dst, src_size);
243 void rgb24to16(const uint8_t *src, uint8_t *dst, unsigned src_size)
245 #ifdef CAN_COMPILE_X86_ASM
246 // ordered per speed fasterst first
248 rgb24to16_MMX2(src, dst, src_size);
249 else if(gCpuCaps.has3DNow)
250 rgb24to16_3DNow(src, dst, src_size);
251 else if(gCpuCaps.hasMMX)
252 rgb24to16_MMX(src, dst, src_size);
254 rgb24to16_C(src, dst, src_size);
256 rgb24to16_C(src, dst, src_size);
260 void rgb24to15(const uint8_t *src, uint8_t *dst, unsigned src_size)
262 #ifdef CAN_COMPILE_X86_ASM
263 // ordered per speed fasterst first
265 rgb24to15_MMX2(src, dst, src_size);
266 else if(gCpuCaps.has3DNow)
267 rgb24to15_3DNow(src, dst, src_size);
268 else if(gCpuCaps.hasMMX)
269 rgb24to15_MMX(src, dst, src_size);
271 rgb24to15_C(src, dst, src_size);
273 rgb24to15_C(src, dst, src_size);
278 * Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
280 void palette8torgb16(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
283 for(i=0; i<num_pixels; i++)
284 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
288 * Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
290 void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette)
293 for(i=0; i<num_pixels; i++)
294 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
297 void rgb32tobgr32(const uint8_t *src, uint8_t *dst, unsigned int src_size)
299 #ifdef CAN_COMPILE_X86_ASM
300 // ordered per speed fasterst first
302 rgb32tobgr32_MMX2(src, dst, src_size);
303 else if(gCpuCaps.has3DNow)
304 rgb32tobgr32_3DNow(src, dst, src_size);
305 else if(gCpuCaps.hasMMX)
306 rgb32tobgr32_MMX(src, dst, src_size);
308 rgb32tobgr32_C(src, dst, src_size);
310 rgb32tobgr32_C(src, dst, src_size);
316 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
317 * problem for anyone then tell me, and ill fix it)
319 void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
320 unsigned int width, unsigned int height,
321 unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)
323 #ifdef CAN_COMPILE_X86_ASM
324 // ordered per speed fasterst first
326 yv12toyuy2_MMX2(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
327 else if(gCpuCaps.has3DNow)
328 yv12toyuy2_3DNow(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
329 else if(gCpuCaps.hasMMX)
330 yv12toyuy2_MMX(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
332 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
334 yv12toyuy2_C(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride);
340 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
341 * problem for anyone then tell me, and ill fix it)
343 void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
344 unsigned int width, unsigned int height,
345 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
347 #ifdef CAN_COMPILE_X86_ASM
348 // ordered per speed fasterst first
350 yuy2toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
351 else if(gCpuCaps.has3DNow)
352 yuy2toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
353 else if(gCpuCaps.hasMMX)
354 yuy2toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
356 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
358 yuy2toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
364 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
365 * problem for anyone then tell me, and ill fix it)
366 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
368 void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
369 unsigned int width, unsigned int height,
370 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
372 #ifdef CAN_COMPILE_X86_ASM
373 // ordered per speed fasterst first
375 uyvytoyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
376 else if(gCpuCaps.has3DNow)
377 uyvytoyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
378 else if(gCpuCaps.hasMMX)
379 uyvytoyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
381 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
383 uyvytoyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
389 * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
390 * problem for anyone then tell me, and ill fix it)
391 * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
393 void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
394 unsigned int width, unsigned int height,
395 unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)
397 #ifdef CAN_COMPILE_X86_ASM
398 // ordered per speed fasterst first
400 rgb24toyv12_MMX2(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
401 else if(gCpuCaps.has3DNow)
402 rgb24toyv12_3DNow(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
403 else if(gCpuCaps.hasMMX)
404 rgb24toyv12_MMX(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
406 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);
408 rgb24toyv12_C(src, ydst, udst, vdst, width, height, lumStride, chromStride, srcStride);