3 * rgb2rgb.c, Software RGB to RGB convertor
4 * Written by Nick Kurshev.
9 #include "../mmx_defs.h"
12 static const uint64_t mask32 __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
13 static const uint64_t mask24l __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
14 static const uint64_t mask24h __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
15 static const uint64_t mask15b __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */
16 static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */
17 static const uint64_t mask15s __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
20 void rgb24to32(const uint8_t *src,uint8_t *dst,uint32_t src_size)
23 const uint8_t *s = src;
30 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
31 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
32 __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
33 if(mm_end == end) mm_end -= MMREG_SIZE*2;
42 "punpckldq %%mm1, %%mm0\n\t"
43 "punpckldq %%mm3, %%mm2\n\t"
44 "pand %%mm7, %%mm0\n\t"
45 "pand %%mm7, %%mm2\n\t"
46 MOVNTQ" %%mm0, %0\n\t"
54 __asm __volatile(SFENCE:::"memory");
55 __asm __volatile(EMMS:::"memory");
66 void rgb32to24(const uint8_t *src,uint8_t *dst,uint32_t src_size)
69 const uint8_t *s = src;
76 __asm __volatile(PREFETCH" %0"::"m"(*s):"memory");
77 mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
81 ::"m"(mask24l),"m"(mask24h):"memory");
82 if(mm_end == end) mm_end -= MMREG_SIZE*2;
89 "movq %%mm0, %%mm2\n\t"
90 "movq %%mm1, %%mm3\n\t"
93 "pand %%mm7, %%mm0\n\t"
94 "pand %%mm7, %%mm1\n\t"
95 "pand %%mm6, %%mm2\n\t"
96 "pand %%mm6, %%mm3\n\t"
97 "por %%mm2, %%mm0\n\t"
98 "por %%mm3, %%mm1\n\t"
99 MOVNTQ" %%mm0, %0\n\t"
107 __asm __volatile(SFENCE:::"memory");
108 __asm __volatile(EMMS:::"memory");
120 Original by Strepto/Astral
121 ported to gcc & bugfixed : A'rpi
122 MMX2, 3DNOW optimization by Nick Kurshev
123 32bit c version, and and&add trick by Michael Niedermayer
125 void rgb15to16(const uint8_t *src,uint8_t *dst,uint32_t src_size)
128 register const char* s=src+src_size;
129 register char* d=dst+src_size;
130 register int offs=-src_size;
131 __asm __volatile(PREFETCH" %0"::"m"(*(s+offs)));
140 "movq 8%1, %%mm2\n\t"
141 "movq %%mm0, %%mm1\n\t"
142 "movq %%mm2, %%mm3\n\t"
143 "pand %%mm4, %%mm0\n\t"
144 "pand %%mm4, %%mm2\n\t"
145 "paddw %%mm1, %%mm0\n\t"
146 "paddw %%mm3, %%mm2\n\t"
147 MOVNTQ" %%mm0, %0\n\t"
154 __asm __volatile(SFENCE:::"memory");
155 __asm __volatile(EMMS:::"memory");
158 const uint16_t *s1=( uint16_t * )src;
159 uint16_t *d1=( uint16_t * )dst;
160 uint16_t *e=((uint8_t *)s1)+src_size;
162 register int x=*( s1++ );
165 0111 1111 1110 0000=0x7FE0
166 00000000000001 1111=0x001F */
167 *( d1++ )=( x&0x001F )|( ( x&0x7FE0 )<<1 );
170 const uint32_t *s1=( uint32_t * )src;
171 uint32_t *d1=( uint32_t * )dst;
173 int size= src_size>>2;
174 for(i=0; i<size; i++)
176 register int x= s1[i];
177 // d1[i] = x + (x&0x7FE07FE0); //faster but need msbit =0 which might not allways be true
178 d1[i] = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
186 * Pallete is assumed to contain bgr32
188 void palette8torgb32(uint8_t *src, uint8_t *dst, int src_size, uint8_t *palette)
191 for(i=0; i<src_size; i++)
192 ((uint32_t *)dst)[i] = ((uint32_t *)palette)[ src[i] ];
196 * Pallete is assumed to contain bgr32
198 void palette8torgb24(uint8_t *src, uint8_t *dst, int src_size, uint8_t *palette)
202 writes 1 byte o much and might cause alignment issues on some architectures?
203 for(i=0; i<src_size; i++)
204 ((uint32_t *)(&dst[i*3])) = ((uint32_t *)palette)[ src[i] ];
206 for(i=0; i<src_size; i++)
209 dst[0]= palette[ src[i]*4+0 ];
210 dst[1]= palette[ src[i]*4+1 ];
211 dst[2]= palette[ src[i]*4+2 ];
216 void rgb32to16(uint8_t *src, uint8_t *dst, int src_size)
219 for(i=0; i<src_size; i+=4)
221 const int b= src[i+0];
222 const int g= src[i+1];
223 const int r= src[i+2];
225 ((uint16_t *)dst)[i]= (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
229 void rgb32to15(uint8_t *src, uint8_t *dst, int src_size)
232 for(i=0; i<src_size; i+=4)
234 const int b= src[i+0];
235 const int g= src[i+1];
236 const int r= src[i+2];
238 ((uint16_t *)dst)[i]= (b>>3) | ((g&0xF8)<<3) | ((r&0xF8)<<7);
244 * Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
246 void palette8torgb16(uint8_t *src, uint8_t *dst, int src_size, uint8_t *palette)
249 for(i=0; i<src_size; i++)
250 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
254 * Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
256 void palette8torgb15(uint8_t *src, uint8_t *dst, int src_size, uint8_t *palette)
259 for(i=0; i<src_size; i++)
260 ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];