2504
|
1 #include <inttypes.h>
|
|
2 #include "../config.h"
|
|
3 #include "rgb2rgb.h"
|
2506
|
4 #include "mmx.h"
|
2504
|
5
|
2511
|
6 #ifdef HAVE_3DNOW
|
|
7 #define PREFETCH "prefetch"
|
|
8 #define PREFETCHW "prefetchw"
|
|
9 #elif HAVE_MMX2
|
|
10 #define PREFETCH "prefetchnta"
|
|
11 #define PREFETCHW "prefetcht0"
|
|
12 #endif
|
|
13
|
|
14 #ifdef HAVE_3DNOW
|
|
15 #define EMMS "femms"
|
|
16 #else
|
|
17 #define EMMS "emms"
|
|
18 #endif
|
|
19
|
|
20 #ifdef HAVE_MMX2
|
|
21 #define MOVNTQ "movntq"
|
|
22 #else
|
|
23 #define MOVNTQ "movq"
|
|
24 #endif
|
|
25
|
2504
|
26 void rgb24to32(uint8_t *src,uint8_t *dst,uint32_t src_size)
|
|
27 {
|
2508
|
28 uint8_t *dest = dst;
|
2504
|
29 uint8_t *s = src;
|
|
30 uint8_t *end;
|
2510
|
31 #ifdef HAVE_MMX
|
|
32 const uint64_t mask32 = 0x00FFFFFF00FFFFFFULL;
|
|
33 uint8_t *mm_end;
|
|
34 #endif
|
2504
|
35 end = s + src_size;
|
2510
|
36 #ifdef HAVE_MMX
|
2511
|
37 #ifdef PREFETCH
|
|
38 __asm __volatile(
|
|
39 PREFETCH" %0\n\t"
|
2512
|
40 ::"m"(*s):"memory");
|
2511
|
41 #endif
|
2510
|
42 mm_end = (uint8_t*)((((unsigned long)end)/16)*16);
|
|
43 __asm __volatile("movq %0, %%mm7"::"m"(mask32):"memory");
|
|
44 while(s < mm_end)
|
|
45 {
|
2511
|
46 #ifdef PREFETCH
|
|
47 __asm __volatile(
|
2512
|
48 PREFETCH" 32%0\n\t"
|
|
49 ::"m"(*s):"memory");
|
2511
|
50 #endif
|
2510
|
51 __asm __volatile(
|
|
52 "movd %1, %%mm0\n\t"
|
|
53 "movd 3%1, %%mm1\n\t"
|
|
54 "movd 6%1, %%mm2\n\t"
|
|
55 "movd 9%1, %%mm3\n\t"
|
|
56 "punpckldq %%mm1, %%mm0\n\t"
|
|
57 "punpckldq %%mm3, %%mm2\n\t"
|
|
58 "pand %%mm7, %%mm0\n\t"
|
|
59 "pand %%mm7, %%mm2\n\t"
|
2511
|
60 MOVNTQ" %%mm0, %0\n\t"
|
|
61 MOVNTQ" %%mm2, 8%0"
|
2510
|
62 :"=m"(*dest)
|
|
63 :"m"(*s)
|
|
64 :"memory");
|
|
65 dest += 16;
|
|
66 s += 12;
|
|
67 }
|
2511
|
68 __asm __volatile(EMMS:::"memory");
|
2510
|
69 #endif
|
2504
|
70 while(s < end)
|
|
71 {
|
2508
|
72 *dest++ = *s++;
|
|
73 *dest++ = *s++;
|
|
74 *dest++ = *s++;
|
|
75 *dest++ = 0;
|
2504
|
76 }
|
|
77 }
|
2505
|
78
|
2510
|
79 /* TODO: MMX optimization */
|
2505
|
80 void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
|
|
81 {
|
|
82 uint8_t *dest = dst;
|
|
83 uint8_t *s = src;
|
|
84 uint8_t *end;
|
|
85 end = s + src_size;
|
|
86 while(s < end)
|
|
87 {
|
|
88 *dest++ = *s++;
|
|
89 *dest++ = *s++;
|
|
90 *dest++ = *s++;
|
|
91 s++;
|
|
92 }
|
|
93 }
|
2506
|
94
|
|
95 /* Original by Strepto/Astral
|
|
96 ported to gcc & bugfixed : A'rpi */
|
|
97 void rgb15to16(uint8_t *src,uint8_t *dst,uint32_t src_size)
|
|
98 {
|
|
99 #ifdef HAVE_MMX
|
|
100 static uint64_t mask_b = 0x001F001F001F001FLL; // 00000000 00011111 xxB
|
|
101 static uint64_t mask_rg = 0x7FE07FE07FE07FE0LL; // 01111111 11100000 RGx
|
|
102 register char* s=src+src_size;
|
|
103 register char* d=dst+src_size;
|
|
104 register int offs=-src_size;
|
|
105 movq_m2r (mask_b, mm4);
|
|
106 movq_m2r (mask_rg, mm5);
|
|
107 while(offs<0){
|
|
108 movq_m2r (*(s+offs), mm0);
|
|
109 movq_r2r (mm0, mm1);
|
|
110
|
|
111 movq_m2r (*(s+8+offs), mm2);
|
|
112 movq_r2r (mm2, mm3);
|
|
113
|
|
114 pand_r2r (mm4, mm0);
|
|
115 pand_r2r (mm5, mm1);
|
|
116
|
|
117 psllq_i2r(1,mm1);
|
|
118 pand_r2r (mm4, mm2);
|
|
119
|
|
120 pand_r2r (mm5, mm3);
|
|
121 por_r2r (mm1, mm0);
|
|
122
|
|
123 psllq_i2r(1,mm3);
|
|
124 movq_r2m (mm0,*(d+offs));
|
|
125
|
|
126 por_r2r (mm3,mm2);
|
|
127 movq_r2m (mm2,*(d+8+offs));
|
|
128
|
|
129 offs+=16;
|
|
130 }
|
|
131 emms();
|
|
132 #else
|
|
133 uint16_t *s1=( uint16_t * )src;
|
|
134 uint16_t *d1=( uint16_t * )dst;
|
|
135 uint16_t *e=((uint8_t *)s1)+src_size;
|
|
136 while( s1<e ){
|
|
137 register int x=*( s1++ );
|
|
138 /* rrrrrggggggbbbbb
|
|
139 0rrrrrgggggbbbbb
|
|
140 0111 1111 1110 0000=0x7FE0
|
|
141 00000000000001 1111=0x001F */
|
|
142 *( d1++ )=( x&0x001F )|( ( x&0x7FE0 )<<1 );
|
|
143 }
|
|
144 #endif
|
|
145 }
|