| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * software RGB to RGB converter | ||
| 3 | * pluralize by software PAL8 to RGB converter | ||
| 4 | * software YUV to YUV converter | ||
| 5 | * software YUV to RGB converter | ||
| 6 | * Written by Nick Kurshev. | ||
| 7 | * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at) | ||
| 8 | * | ||
| 9 | * This file is part of FFmpeg. | ||
| 10 | * | ||
| 11 | * FFmpeg is free software; you can redistribute it and/or | ||
| 12 | * modify it under the terms of the GNU Lesser General Public | ||
| 13 | * License as published by the Free Software Foundation; either | ||
| 14 | * version 2.1 of the License, or (at your option) any later version. | ||
| 15 | * | ||
| 16 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 19 | * Lesser General Public License for more details. | ||
| 20 | * | ||
| 21 | * You should have received a copy of the GNU Lesser General Public | ||
| 22 | * License along with FFmpeg; if not, write to the Free Software | ||
| 23 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 24 | */ | ||
| 25 | |||
| 26 | #include <stdint.h> | ||
| 27 | |||
| 28 | #include "config.h" | ||
| 29 | #include "libavutil/attributes.h" | ||
| 30 | #include "libavutil/x86/cpu.h" | ||
| 31 | #include "libavutil/cpu.h" | ||
| 32 | #include "libavutil/bswap.h" | ||
| 33 | #include "libavutil/mem_internal.h" | ||
| 34 | |||
| 35 | #include "libswscale/rgb2rgb.h" | ||
| 36 | #include "libswscale/swscale.h" | ||
| 37 | #include "libswscale/swscale_internal.h" | ||
| 38 | |||
| 39 | #if HAVE_INLINE_ASM | ||
| 40 | #include "libavutil/x86/asm.h" | ||
| 41 | |||
| 42 | DECLARE_ASM_CONST(8, uint64_t, mmx_ff) = 0x00000000000000FFULL; | ||
| 43 | DECLARE_ASM_CONST(8, uint64_t, mmx_null) = 0x0000000000000000ULL; | ||
| 44 | DECLARE_ASM_CONST(8, uint64_t, mask32a) = 0xFF000000FF000000ULL; | ||
| 45 | DECLARE_ASM_CONST(8, uint64_t, mask3216br) = 0x00F800F800F800F8ULL; | ||
| 46 | DECLARE_ASM_CONST(8, uint64_t, mask3216g) = 0x0000FC000000FC00ULL; | ||
| 47 | DECLARE_ASM_CONST(8, uint64_t, mask3215g) = 0x0000F8000000F800ULL; | ||
| 48 | DECLARE_ASM_CONST(8, uint64_t, mul3216) = 0x2000000420000004ULL; | ||
| 49 | DECLARE_ASM_CONST(8, uint64_t, mul3215) = 0x2000000820000008ULL; | ||
| 50 | DECLARE_ASM_CONST(8, uint64_t, mask24b) = 0x00FF0000FF0000FFULL; | ||
| 51 | DECLARE_ASM_CONST(8, uint64_t, mask24g) = 0xFF0000FF0000FF00ULL; | ||
| 52 | DECLARE_ASM_CONST(8, uint64_t, mask24r) = 0x0000FF0000FF0000ULL; | ||
| 53 | DECLARE_ASM_CONST(8, uint64_t, mask24l) = 0x0000000000FFFFFFULL; | ||
| 54 | DECLARE_ASM_CONST(8, uint64_t, mask24h) = 0x0000FFFFFF000000ULL; | ||
| 55 | DECLARE_ASM_CONST(8, uint64_t, mask15b) = 0x001F001F001F001FULL; /* 00000000 00011111 xxB */ | ||
| 56 | DECLARE_ASM_CONST(8, uint64_t, mask15rg) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000 RGx */ | ||
| 57 | DECLARE_ASM_CONST(8, uint64_t, mask15s) = 0xFFE0FFE0FFE0FFE0ULL; | ||
| 58 | DECLARE_ASM_CONST(8, uint64_t, mask15g) = 0x03E003E003E003E0ULL; | ||
| 59 | DECLARE_ASM_CONST(8, uint64_t, mask15r) = 0x7C007C007C007C00ULL; | ||
| 60 | #define mask16b mask15b | ||
| 61 | DECLARE_ASM_CONST(8, uint64_t, mask16g) = 0x07E007E007E007E0ULL; | ||
| 62 | DECLARE_ASM_CONST(8, uint64_t, mask16r) = 0xF800F800F800F800ULL; | ||
| 63 | #define red_16mask mask3215g | ||
| 64 | DECLARE_ASM_CONST(8, uint64_t, green_16mask) = 0x000007e0000007e0ULL; | ||
| 65 | DECLARE_ASM_CONST(8, uint64_t, blue_16mask) = 0x0000001f0000001fULL; | ||
| 66 | DECLARE_ASM_CONST(8, uint64_t, red_15mask) = 0x00007c0000007c00ULL; | ||
| 67 | DECLARE_ASM_CONST(8, uint64_t, green_15mask) = 0x000003e0000003e0ULL; | ||
| 68 | #define blue_15mask blue_16mask | ||
| 69 | DECLARE_ASM_CONST(8, uint64_t, mul15_mid) = 0x4200420042004200ULL; | ||
| 70 | DECLARE_ASM_CONST(8, uint64_t, mul15_hi) = 0x0210021002100210ULL; | ||
| 71 | DECLARE_ASM_CONST(8, uint64_t, mul16_mid) = 0x2080208020802080ULL; | ||
| 72 | |||
| 73 | #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 74 | #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 75 | #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 76 | #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 77 | #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 78 | #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 79 | #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 80 | #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 81 | #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5)) | ||
| 82 | |||
| 83 | // MMXEXT versions | ||
| 84 | #define PREFETCH "prefetchnta" | ||
| 85 | #define PAVGB "pavgb" | ||
| 86 | #define MOVNTQ "movntq" | ||
| 87 | #define SFENCE "sfence" | ||
| 88 | |||
| 89 | #define EMMS "emms" | ||
| 90 | |||
| 91 | 8 | static inline void rgb24tobgr32_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 92 | { | ||
| 93 | 8 | uint8_t *dest = dst; | |
| 94 | 8 | const uint8_t *s = src; | |
| 95 | const uint8_t *end; | ||
| 96 | const uint8_t *mm_end; | ||
| 97 | 8 | end = s + src_size; | |
| 98 | 8 | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 99 | 8 | mm_end = end - 23; | |
| 100 | 8 | __asm__ volatile("movq %0, %%mm7"::"m"(mask32a):"memory"); | |
| 101 |
2/2✓ Branch 0 taken 251 times.
✓ Branch 1 taken 8 times.
|
259 | while (s < mm_end) { |
| 102 | 251 | __asm__ volatile( | |
| 103 | PREFETCH" 32(%1) \n\t" | ||
| 104 | "movd (%1), %%mm0 \n\t" | ||
| 105 | "punpckldq 3(%1), %%mm0 \n\t" | ||
| 106 | "movd 6(%1), %%mm1 \n\t" | ||
| 107 | "punpckldq 9(%1), %%mm1 \n\t" | ||
| 108 | "movd 12(%1), %%mm2 \n\t" | ||
| 109 | "punpckldq 15(%1), %%mm2 \n\t" | ||
| 110 | "movd 18(%1), %%mm3 \n\t" | ||
| 111 | "punpckldq 21(%1), %%mm3 \n\t" | ||
| 112 | "por %%mm7, %%mm0 \n\t" | ||
| 113 | "por %%mm7, %%mm1 \n\t" | ||
| 114 | "por %%mm7, %%mm2 \n\t" | ||
| 115 | "por %%mm7, %%mm3 \n\t" | ||
| 116 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 117 | MOVNTQ" %%mm1, 8(%0) \n\t" | ||
| 118 | MOVNTQ" %%mm2, 16(%0) \n\t" | ||
| 119 | MOVNTQ" %%mm3, 24(%0)" | ||
| 120 | :: "r"(dest), "r"(s) | ||
| 121 | :"memory"); | ||
| 122 | 251 | dest += 32; | |
| 123 | 251 | s += 24; | |
| 124 | } | ||
| 125 | 8 | __asm__ volatile(SFENCE:::"memory"); | |
| 126 | 8 | __asm__ volatile(EMMS:::"memory"); | |
| 127 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 8 times.
|
20 | while (s < end) { |
| 128 | 12 | *dest++ = *s++; | |
| 129 | 12 | *dest++ = *s++; | |
| 130 | 12 | *dest++ = *s++; | |
| 131 | 12 | *dest++ = 255; | |
| 132 | } | ||
| 133 | 8 | } | |
| 134 | |||
| 135 | #define STORE_BGR24_MMX \ | ||
| 136 | "psrlq $8, %%mm2 \n\t" \ | ||
| 137 | "psrlq $8, %%mm3 \n\t" \ | ||
| 138 | "psrlq $8, %%mm6 \n\t" \ | ||
| 139 | "psrlq $8, %%mm7 \n\t" \ | ||
| 140 | "pand "MANGLE(mask24l)", %%mm0\n\t" \ | ||
| 141 | "pand "MANGLE(mask24l)", %%mm1\n\t" \ | ||
| 142 | "pand "MANGLE(mask24l)", %%mm4\n\t" \ | ||
| 143 | "pand "MANGLE(mask24l)", %%mm5\n\t" \ | ||
| 144 | "pand "MANGLE(mask24h)", %%mm2\n\t" \ | ||
| 145 | "pand "MANGLE(mask24h)", %%mm3\n\t" \ | ||
| 146 | "pand "MANGLE(mask24h)", %%mm6\n\t" \ | ||
| 147 | "pand "MANGLE(mask24h)", %%mm7\n\t" \ | ||
| 148 | "por %%mm2, %%mm0 \n\t" \ | ||
| 149 | "por %%mm3, %%mm1 \n\t" \ | ||
| 150 | "por %%mm6, %%mm4 \n\t" \ | ||
| 151 | "por %%mm7, %%mm5 \n\t" \ | ||
| 152 | \ | ||
| 153 | "movq %%mm1, %%mm2 \n\t" \ | ||
| 154 | "movq %%mm4, %%mm3 \n\t" \ | ||
| 155 | "psllq $48, %%mm2 \n\t" \ | ||
| 156 | "psllq $32, %%mm3 \n\t" \ | ||
| 157 | "por %%mm2, %%mm0 \n\t" \ | ||
| 158 | "psrlq $16, %%mm1 \n\t" \ | ||
| 159 | "psrlq $32, %%mm4 \n\t" \ | ||
| 160 | "psllq $16, %%mm5 \n\t" \ | ||
| 161 | "por %%mm3, %%mm1 \n\t" \ | ||
| 162 | "por %%mm5, %%mm4 \n\t" \ | ||
| 163 | \ | ||
| 164 | MOVNTQ" %%mm0, (%0) \n\t" \ | ||
| 165 | MOVNTQ" %%mm1, 8(%0) \n\t" \ | ||
| 166 | MOVNTQ" %%mm4, 16(%0)" | ||
| 167 | |||
| 168 | |||
| 169 | 9 | static inline void rgb32tobgr24_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 170 | { | ||
| 171 | 9 | uint8_t *dest = dst; | |
| 172 | 9 | const uint8_t *s = src; | |
| 173 | const uint8_t *end; | ||
| 174 | const uint8_t *mm_end; | ||
| 175 | 9 | end = s + src_size; | |
| 176 | 9 | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 177 | 9 | mm_end = end - 31; | |
| 178 |
2/2✓ Branch 0 taken 253 times.
✓ Branch 1 taken 9 times.
|
262 | while (s < mm_end) { |
| 179 | 253 | __asm__ volatile( | |
| 180 | PREFETCH" 32(%1) \n\t" | ||
| 181 | "movq (%1), %%mm0 \n\t" | ||
| 182 | "movq 8(%1), %%mm1 \n\t" | ||
| 183 | "movq 16(%1), %%mm4 \n\t" | ||
| 184 | "movq 24(%1), %%mm5 \n\t" | ||
| 185 | "movq %%mm0, %%mm2 \n\t" | ||
| 186 | "movq %%mm1, %%mm3 \n\t" | ||
| 187 | "movq %%mm4, %%mm6 \n\t" | ||
| 188 | "movq %%mm5, %%mm7 \n\t" | ||
| 189 | STORE_BGR24_MMX | ||
| 190 | :: "r"(dest), "r"(s) | ||
| 191 | NAMED_CONSTRAINTS_ADD(mask24l,mask24h) | ||
| 192 | :"memory"); | ||
| 193 | 253 | dest += 24; | |
| 194 | 253 | s += 32; | |
| 195 | } | ||
| 196 | 9 | __asm__ volatile(SFENCE:::"memory"); | |
| 197 | 9 | __asm__ volatile(EMMS:::"memory"); | |
| 198 |
2/2✓ Branch 0 taken 11 times.
✓ Branch 1 taken 9 times.
|
20 | while (s < end) { |
| 199 | 11 | *dest++ = *s++; | |
| 200 | 11 | *dest++ = *s++; | |
| 201 | 11 | *dest++ = *s++; | |
| 202 | 11 | s++; | |
| 203 | } | ||
| 204 | 9 | } | |
| 205 | |||
| 206 | /* | ||
| 207 | original by Strepto/Astral | ||
| 208 | ported to gcc & bugfixed: A'rpi | ||
| 209 | MMXEXT, 3DNOW optimization by Nick Kurshev | ||
| 210 | 32-bit C version, and and&add trick by Michael Niedermayer | ||
| 211 | */ | ||
| 212 | ✗ | static inline void rgb15to16_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 213 | { | ||
| 214 | ✗ | register const uint8_t* s=src; | |
| 215 | ✗ | register uint8_t* d=dst; | |
| 216 | register const uint8_t *end; | ||
| 217 | const uint8_t *mm_end; | ||
| 218 | ✗ | end = s + src_size; | |
| 219 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s)); | |
| 220 | ✗ | __asm__ volatile("movq %0, %%mm4"::"m"(mask15s)); | |
| 221 | ✗ | mm_end = end - 15; | |
| 222 | ✗ | while (s<mm_end) { | |
| 223 | ✗ | __asm__ volatile( | |
| 224 | PREFETCH" 32(%1) \n\t" | ||
| 225 | "movq (%1), %%mm0 \n\t" | ||
| 226 | "movq 8(%1), %%mm2 \n\t" | ||
| 227 | "movq %%mm0, %%mm1 \n\t" | ||
| 228 | "movq %%mm2, %%mm3 \n\t" | ||
| 229 | "pand %%mm4, %%mm0 \n\t" | ||
| 230 | "pand %%mm4, %%mm2 \n\t" | ||
| 231 | "paddw %%mm1, %%mm0 \n\t" | ||
| 232 | "paddw %%mm3, %%mm2 \n\t" | ||
| 233 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 234 | MOVNTQ" %%mm2, 8(%0)" | ||
| 235 | :: "r"(d), "r"(s) | ||
| 236 | ); | ||
| 237 | ✗ | d+=16; | |
| 238 | ✗ | s+=16; | |
| 239 | } | ||
| 240 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 241 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 242 | ✗ | mm_end = end - 3; | |
| 243 | ✗ | while (s < mm_end) { | |
| 244 | ✗ | register unsigned x= *((const uint32_t *)s); | |
| 245 | ✗ | *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0); | |
| 246 | ✗ | d+=4; | |
| 247 | ✗ | s+=4; | |
| 248 | } | ||
| 249 | ✗ | if (s < end) { | |
| 250 | ✗ | register unsigned short x= *((const uint16_t *)s); | |
| 251 | ✗ | *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0); | |
| 252 | } | ||
| 253 | ✗ | } | |
| 254 | |||
| 255 | ✗ | static inline void rgb16to15_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 256 | { | ||
| 257 | ✗ | register const uint8_t* s=src; | |
| 258 | ✗ | register uint8_t* d=dst; | |
| 259 | register const uint8_t *end; | ||
| 260 | const uint8_t *mm_end; | ||
| 261 | ✗ | end = s + src_size; | |
| 262 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s)); | |
| 263 | ✗ | __asm__ volatile("movq %0, %%mm7"::"m"(mask15rg)); | |
| 264 | ✗ | __asm__ volatile("movq %0, %%mm6"::"m"(mask15b)); | |
| 265 | ✗ | mm_end = end - 15; | |
| 266 | ✗ | while (s<mm_end) { | |
| 267 | ✗ | __asm__ volatile( | |
| 268 | PREFETCH" 32(%1) \n\t" | ||
| 269 | "movq (%1), %%mm0 \n\t" | ||
| 270 | "movq 8(%1), %%mm2 \n\t" | ||
| 271 | "movq %%mm0, %%mm1 \n\t" | ||
| 272 | "movq %%mm2, %%mm3 \n\t" | ||
| 273 | "psrlq $1, %%mm0 \n\t" | ||
| 274 | "psrlq $1, %%mm2 \n\t" | ||
| 275 | "pand %%mm7, %%mm0 \n\t" | ||
| 276 | "pand %%mm7, %%mm2 \n\t" | ||
| 277 | "pand %%mm6, %%mm1 \n\t" | ||
| 278 | "pand %%mm6, %%mm3 \n\t" | ||
| 279 | "por %%mm1, %%mm0 \n\t" | ||
| 280 | "por %%mm3, %%mm2 \n\t" | ||
| 281 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 282 | MOVNTQ" %%mm2, 8(%0)" | ||
| 283 | :: "r"(d), "r"(s) | ||
| 284 | ); | ||
| 285 | ✗ | d+=16; | |
| 286 | ✗ | s+=16; | |
| 287 | } | ||
| 288 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 289 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 290 | ✗ | mm_end = end - 3; | |
| 291 | ✗ | while (s < mm_end) { | |
| 292 | ✗ | register uint32_t x= *((const uint32_t*)s); | |
| 293 | ✗ | *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F); | |
| 294 | ✗ | s+=4; | |
| 295 | ✗ | d+=4; | |
| 296 | } | ||
| 297 | ✗ | if (s < end) { | |
| 298 | ✗ | register uint16_t x= *((const uint16_t*)s); | |
| 299 | ✗ | *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F); | |
| 300 | } | ||
| 301 | ✗ | } | |
| 302 | |||
| 303 | ✗ | static inline void rgb32to16_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 304 | { | ||
| 305 | ✗ | const uint8_t *s = src; | |
| 306 | const uint8_t *end; | ||
| 307 | const uint8_t *mm_end; | ||
| 308 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 309 | ✗ | end = s + src_size; | |
| 310 | ✗ | mm_end = end - 15; | |
| 311 | ✗ | __asm__ volatile( | |
| 312 | "movq %3, %%mm5 \n\t" | ||
| 313 | "movq %4, %%mm6 \n\t" | ||
| 314 | "movq %5, %%mm7 \n\t" | ||
| 315 | "jmp 2f \n\t" | ||
| 316 | ".p2align 4 \n\t" | ||
| 317 | "1: \n\t" | ||
| 318 | PREFETCH" 32(%1) \n\t" | ||
| 319 | "movd (%1), %%mm0 \n\t" | ||
| 320 | "movd 4(%1), %%mm3 \n\t" | ||
| 321 | "punpckldq 8(%1), %%mm0 \n\t" | ||
| 322 | "punpckldq 12(%1), %%mm3 \n\t" | ||
| 323 | "movq %%mm0, %%mm1 \n\t" | ||
| 324 | "movq %%mm3, %%mm4 \n\t" | ||
| 325 | "pand %%mm6, %%mm0 \n\t" | ||
| 326 | "pand %%mm6, %%mm3 \n\t" | ||
| 327 | "pmaddwd %%mm7, %%mm0 \n\t" | ||
| 328 | "pmaddwd %%mm7, %%mm3 \n\t" | ||
| 329 | "pand %%mm5, %%mm1 \n\t" | ||
| 330 | "pand %%mm5, %%mm4 \n\t" | ||
| 331 | "por %%mm1, %%mm0 \n\t" | ||
| 332 | "por %%mm4, %%mm3 \n\t" | ||
| 333 | "psrld $5, %%mm0 \n\t" | ||
| 334 | "pslld $11, %%mm3 \n\t" | ||
| 335 | "por %%mm3, %%mm0 \n\t" | ||
| 336 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 337 | "add $16, %1 \n\t" | ||
| 338 | "add $8, %0 \n\t" | ||
| 339 | "2: \n\t" | ||
| 340 | "cmp %2, %1 \n\t" | ||
| 341 | " jb 1b \n\t" | ||
| 342 | : "+r" (d), "+r"(s) | ||
| 343 | : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216) | ||
| 344 | ); | ||
| 345 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 346 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 347 | ✗ | while (s < end) { | |
| 348 | ✗ | register int rgb = *(const uint32_t*)s; s += 4; | |
| 349 | ✗ | *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8); | |
| 350 | } | ||
| 351 | ✗ | } | |
| 352 | |||
| 353 | ✗ | static inline void rgb32tobgr16_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 354 | { | ||
| 355 | ✗ | const uint8_t *s = src; | |
| 356 | const uint8_t *end; | ||
| 357 | const uint8_t *mm_end; | ||
| 358 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 359 | ✗ | end = s + src_size; | |
| 360 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 361 | ✗ | __asm__ volatile( | |
| 362 | "movq %0, %%mm7 \n\t" | ||
| 363 | "movq %1, %%mm6 \n\t" | ||
| 364 | ::"m"(red_16mask),"m"(green_16mask)); | ||
| 365 | ✗ | mm_end = end - 15; | |
| 366 | ✗ | while (s < mm_end) { | |
| 367 | ✗ | __asm__ volatile( | |
| 368 | PREFETCH" 32(%1) \n\t" | ||
| 369 | "movd (%1), %%mm0 \n\t" | ||
| 370 | "movd 4(%1), %%mm3 \n\t" | ||
| 371 | "punpckldq 8(%1), %%mm0 \n\t" | ||
| 372 | "punpckldq 12(%1), %%mm3 \n\t" | ||
| 373 | "movq %%mm0, %%mm1 \n\t" | ||
| 374 | "movq %%mm0, %%mm2 \n\t" | ||
| 375 | "movq %%mm3, %%mm4 \n\t" | ||
| 376 | "movq %%mm3, %%mm5 \n\t" | ||
| 377 | "psllq $8, %%mm0 \n\t" | ||
| 378 | "psllq $8, %%mm3 \n\t" | ||
| 379 | "pand %%mm7, %%mm0 \n\t" | ||
| 380 | "pand %%mm7, %%mm3 \n\t" | ||
| 381 | "psrlq $5, %%mm1 \n\t" | ||
| 382 | "psrlq $5, %%mm4 \n\t" | ||
| 383 | "pand %%mm6, %%mm1 \n\t" | ||
| 384 | "pand %%mm6, %%mm4 \n\t" | ||
| 385 | "psrlq $19, %%mm2 \n\t" | ||
| 386 | "psrlq $19, %%mm5 \n\t" | ||
| 387 | "pand %2, %%mm2 \n\t" | ||
| 388 | "pand %2, %%mm5 \n\t" | ||
| 389 | "por %%mm1, %%mm0 \n\t" | ||
| 390 | "por %%mm4, %%mm3 \n\t" | ||
| 391 | "por %%mm2, %%mm0 \n\t" | ||
| 392 | "por %%mm5, %%mm3 \n\t" | ||
| 393 | "psllq $16, %%mm3 \n\t" | ||
| 394 | "por %%mm3, %%mm0 \n\t" | ||
| 395 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 396 | :: "r"(d),"r"(s),"m"(blue_16mask):"memory"); | ||
| 397 | ✗ | d += 4; | |
| 398 | ✗ | s += 16; | |
| 399 | } | ||
| 400 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 401 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 402 | ✗ | while (s < end) { | |
| 403 | ✗ | register int rgb = *(const uint32_t*)s; s += 4; | |
| 404 | ✗ | *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19); | |
| 405 | } | ||
| 406 | ✗ | } | |
| 407 | |||
| 408 | ✗ | static inline void rgb32to15_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 409 | { | ||
| 410 | ✗ | const uint8_t *s = src; | |
| 411 | const uint8_t *end; | ||
| 412 | const uint8_t *mm_end; | ||
| 413 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 414 | ✗ | end = s + src_size; | |
| 415 | ✗ | mm_end = end - 15; | |
| 416 | ✗ | __asm__ volatile( | |
| 417 | "movq %3, %%mm5 \n\t" | ||
| 418 | "movq %4, %%mm6 \n\t" | ||
| 419 | "movq %5, %%mm7 \n\t" | ||
| 420 | "jmp 2f \n\t" | ||
| 421 | ".p2align 4 \n\t" | ||
| 422 | "1: \n\t" | ||
| 423 | PREFETCH" 32(%1) \n\t" | ||
| 424 | "movd (%1), %%mm0 \n\t" | ||
| 425 | "movd 4(%1), %%mm3 \n\t" | ||
| 426 | "punpckldq 8(%1), %%mm0 \n\t" | ||
| 427 | "punpckldq 12(%1), %%mm3 \n\t" | ||
| 428 | "movq %%mm0, %%mm1 \n\t" | ||
| 429 | "movq %%mm3, %%mm4 \n\t" | ||
| 430 | "pand %%mm6, %%mm0 \n\t" | ||
| 431 | "pand %%mm6, %%mm3 \n\t" | ||
| 432 | "pmaddwd %%mm7, %%mm0 \n\t" | ||
| 433 | "pmaddwd %%mm7, %%mm3 \n\t" | ||
| 434 | "pand %%mm5, %%mm1 \n\t" | ||
| 435 | "pand %%mm5, %%mm4 \n\t" | ||
| 436 | "por %%mm1, %%mm0 \n\t" | ||
| 437 | "por %%mm4, %%mm3 \n\t" | ||
| 438 | "psrld $6, %%mm0 \n\t" | ||
| 439 | "pslld $10, %%mm3 \n\t" | ||
| 440 | "por %%mm3, %%mm0 \n\t" | ||
| 441 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 442 | "add $16, %1 \n\t" | ||
| 443 | "add $8, %0 \n\t" | ||
| 444 | "2: \n\t" | ||
| 445 | "cmp %2, %1 \n\t" | ||
| 446 | " jb 1b \n\t" | ||
| 447 | : "+r" (d), "+r"(s) | ||
| 448 | : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215) | ||
| 449 | ); | ||
| 450 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 451 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 452 | ✗ | while (s < end) { | |
| 453 | ✗ | register int rgb = *(const uint32_t*)s; s += 4; | |
| 454 | ✗ | *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9); | |
| 455 | } | ||
| 456 | ✗ | } | |
| 457 | |||
| 458 | ✗ | static inline void rgb32tobgr15_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 459 | { | ||
| 460 | ✗ | const uint8_t *s = src; | |
| 461 | const uint8_t *end; | ||
| 462 | const uint8_t *mm_end; | ||
| 463 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 464 | ✗ | end = s + src_size; | |
| 465 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 466 | ✗ | __asm__ volatile( | |
| 467 | "movq %0, %%mm7 \n\t" | ||
| 468 | "movq %1, %%mm6 \n\t" | ||
| 469 | ::"m"(red_15mask),"m"(green_15mask)); | ||
| 470 | ✗ | mm_end = end - 15; | |
| 471 | ✗ | while (s < mm_end) { | |
| 472 | ✗ | __asm__ volatile( | |
| 473 | PREFETCH" 32(%1) \n\t" | ||
| 474 | "movd (%1), %%mm0 \n\t" | ||
| 475 | "movd 4(%1), %%mm3 \n\t" | ||
| 476 | "punpckldq 8(%1), %%mm0 \n\t" | ||
| 477 | "punpckldq 12(%1), %%mm3 \n\t" | ||
| 478 | "movq %%mm0, %%mm1 \n\t" | ||
| 479 | "movq %%mm0, %%mm2 \n\t" | ||
| 480 | "movq %%mm3, %%mm4 \n\t" | ||
| 481 | "movq %%mm3, %%mm5 \n\t" | ||
| 482 | "psllq $7, %%mm0 \n\t" | ||
| 483 | "psllq $7, %%mm3 \n\t" | ||
| 484 | "pand %%mm7, %%mm0 \n\t" | ||
| 485 | "pand %%mm7, %%mm3 \n\t" | ||
| 486 | "psrlq $6, %%mm1 \n\t" | ||
| 487 | "psrlq $6, %%mm4 \n\t" | ||
| 488 | "pand %%mm6, %%mm1 \n\t" | ||
| 489 | "pand %%mm6, %%mm4 \n\t" | ||
| 490 | "psrlq $19, %%mm2 \n\t" | ||
| 491 | "psrlq $19, %%mm5 \n\t" | ||
| 492 | "pand %2, %%mm2 \n\t" | ||
| 493 | "pand %2, %%mm5 \n\t" | ||
| 494 | "por %%mm1, %%mm0 \n\t" | ||
| 495 | "por %%mm4, %%mm3 \n\t" | ||
| 496 | "por %%mm2, %%mm0 \n\t" | ||
| 497 | "por %%mm5, %%mm3 \n\t" | ||
| 498 | "psllq $16, %%mm3 \n\t" | ||
| 499 | "por %%mm3, %%mm0 \n\t" | ||
| 500 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 501 | ::"r"(d),"r"(s),"m"(blue_15mask):"memory"); | ||
| 502 | ✗ | d += 4; | |
| 503 | ✗ | s += 16; | |
| 504 | } | ||
| 505 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 506 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 507 | ✗ | while (s < end) { | |
| 508 | ✗ | register int rgb = *(const uint32_t*)s; s += 4; | |
| 509 | ✗ | *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19); | |
| 510 | } | ||
| 511 | ✗ | } | |
| 512 | |||
| 513 | ✗ | static inline void rgb24tobgr16_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 514 | { | ||
| 515 | ✗ | const uint8_t *s = src; | |
| 516 | const uint8_t *end; | ||
| 517 | const uint8_t *mm_end; | ||
| 518 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 519 | ✗ | end = s + src_size; | |
| 520 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 521 | ✗ | __asm__ volatile( | |
| 522 | "movq %0, %%mm7 \n\t" | ||
| 523 | "movq %1, %%mm6 \n\t" | ||
| 524 | ::"m"(red_16mask),"m"(green_16mask)); | ||
| 525 | ✗ | mm_end = end - 11; | |
| 526 | ✗ | while (s < mm_end) { | |
| 527 | ✗ | __asm__ volatile( | |
| 528 | PREFETCH" 32(%1) \n\t" | ||
| 529 | "movd (%1), %%mm0 \n\t" | ||
| 530 | "movd 3(%1), %%mm3 \n\t" | ||
| 531 | "punpckldq 6(%1), %%mm0 \n\t" | ||
| 532 | "punpckldq 9(%1), %%mm3 \n\t" | ||
| 533 | "movq %%mm0, %%mm1 \n\t" | ||
| 534 | "movq %%mm0, %%mm2 \n\t" | ||
| 535 | "movq %%mm3, %%mm4 \n\t" | ||
| 536 | "movq %%mm3, %%mm5 \n\t" | ||
| 537 | "psrlq $3, %%mm0 \n\t" | ||
| 538 | "psrlq $3, %%mm3 \n\t" | ||
| 539 | "pand %2, %%mm0 \n\t" | ||
| 540 | "pand %2, %%mm3 \n\t" | ||
| 541 | "psrlq $5, %%mm1 \n\t" | ||
| 542 | "psrlq $5, %%mm4 \n\t" | ||
| 543 | "pand %%mm6, %%mm1 \n\t" | ||
| 544 | "pand %%mm6, %%mm4 \n\t" | ||
| 545 | "psrlq $8, %%mm2 \n\t" | ||
| 546 | "psrlq $8, %%mm5 \n\t" | ||
| 547 | "pand %%mm7, %%mm2 \n\t" | ||
| 548 | "pand %%mm7, %%mm5 \n\t" | ||
| 549 | "por %%mm1, %%mm0 \n\t" | ||
| 550 | "por %%mm4, %%mm3 \n\t" | ||
| 551 | "por %%mm2, %%mm0 \n\t" | ||
| 552 | "por %%mm5, %%mm3 \n\t" | ||
| 553 | "psllq $16, %%mm3 \n\t" | ||
| 554 | "por %%mm3, %%mm0 \n\t" | ||
| 555 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 556 | ::"r"(d),"r"(s),"m"(blue_16mask):"memory"); | ||
| 557 | ✗ | d += 4; | |
| 558 | ✗ | s += 12; | |
| 559 | } | ||
| 560 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 561 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 562 | ✗ | while (s < end) { | |
| 563 | ✗ | const int b = *s++; | |
| 564 | ✗ | const int g = *s++; | |
| 565 | ✗ | const int r = *s++; | |
| 566 | ✗ | *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8); | |
| 567 | } | ||
| 568 | ✗ | } | |
| 569 | |||
| 570 | ✗ | static inline void rgb24to16_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 571 | { | ||
| 572 | ✗ | const uint8_t *s = src; | |
| 573 | const uint8_t *end; | ||
| 574 | const uint8_t *mm_end; | ||
| 575 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 576 | ✗ | end = s + src_size; | |
| 577 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 578 | ✗ | __asm__ volatile( | |
| 579 | "movq %0, %%mm7 \n\t" | ||
| 580 | "movq %1, %%mm6 \n\t" | ||
| 581 | ::"m"(red_16mask),"m"(green_16mask)); | ||
| 582 | ✗ | mm_end = end - 15; | |
| 583 | ✗ | while (s < mm_end) { | |
| 584 | ✗ | __asm__ volatile( | |
| 585 | PREFETCH" 32(%1) \n\t" | ||
| 586 | "movd (%1), %%mm0 \n\t" | ||
| 587 | "movd 3(%1), %%mm3 \n\t" | ||
| 588 | "punpckldq 6(%1), %%mm0 \n\t" | ||
| 589 | "punpckldq 9(%1), %%mm3 \n\t" | ||
| 590 | "movq %%mm0, %%mm1 \n\t" | ||
| 591 | "movq %%mm0, %%mm2 \n\t" | ||
| 592 | "movq %%mm3, %%mm4 \n\t" | ||
| 593 | "movq %%mm3, %%mm5 \n\t" | ||
| 594 | "psllq $8, %%mm0 \n\t" | ||
| 595 | "psllq $8, %%mm3 \n\t" | ||
| 596 | "pand %%mm7, %%mm0 \n\t" | ||
| 597 | "pand %%mm7, %%mm3 \n\t" | ||
| 598 | "psrlq $5, %%mm1 \n\t" | ||
| 599 | "psrlq $5, %%mm4 \n\t" | ||
| 600 | "pand %%mm6, %%mm1 \n\t" | ||
| 601 | "pand %%mm6, %%mm4 \n\t" | ||
| 602 | "psrlq $19, %%mm2 \n\t" | ||
| 603 | "psrlq $19, %%mm5 \n\t" | ||
| 604 | "pand %2, %%mm2 \n\t" | ||
| 605 | "pand %2, %%mm5 \n\t" | ||
| 606 | "por %%mm1, %%mm0 \n\t" | ||
| 607 | "por %%mm4, %%mm3 \n\t" | ||
| 608 | "por %%mm2, %%mm0 \n\t" | ||
| 609 | "por %%mm5, %%mm3 \n\t" | ||
| 610 | "psllq $16, %%mm3 \n\t" | ||
| 611 | "por %%mm3, %%mm0 \n\t" | ||
| 612 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 613 | ::"r"(d),"r"(s),"m"(blue_16mask):"memory"); | ||
| 614 | ✗ | d += 4; | |
| 615 | ✗ | s += 12; | |
| 616 | } | ||
| 617 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 618 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 619 | ✗ | while (s < end) { | |
| 620 | ✗ | const int r = *s++; | |
| 621 | ✗ | const int g = *s++; | |
| 622 | ✗ | const int b = *s++; | |
| 623 | ✗ | *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8); | |
| 624 | } | ||
| 625 | ✗ | } | |
| 626 | |||
| 627 | ✗ | static inline void rgb24tobgr15_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 628 | { | ||
| 629 | ✗ | const uint8_t *s = src; | |
| 630 | const uint8_t *end; | ||
| 631 | const uint8_t *mm_end; | ||
| 632 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 633 | ✗ | end = s + src_size; | |
| 634 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 635 | ✗ | __asm__ volatile( | |
| 636 | "movq %0, %%mm7 \n\t" | ||
| 637 | "movq %1, %%mm6 \n\t" | ||
| 638 | ::"m"(red_15mask),"m"(green_15mask)); | ||
| 639 | ✗ | mm_end = end - 11; | |
| 640 | ✗ | while (s < mm_end) { | |
| 641 | ✗ | __asm__ volatile( | |
| 642 | PREFETCH" 32(%1) \n\t" | ||
| 643 | "movd (%1), %%mm0 \n\t" | ||
| 644 | "movd 3(%1), %%mm3 \n\t" | ||
| 645 | "punpckldq 6(%1), %%mm0 \n\t" | ||
| 646 | "punpckldq 9(%1), %%mm3 \n\t" | ||
| 647 | "movq %%mm0, %%mm1 \n\t" | ||
| 648 | "movq %%mm0, %%mm2 \n\t" | ||
| 649 | "movq %%mm3, %%mm4 \n\t" | ||
| 650 | "movq %%mm3, %%mm5 \n\t" | ||
| 651 | "psrlq $3, %%mm0 \n\t" | ||
| 652 | "psrlq $3, %%mm3 \n\t" | ||
| 653 | "pand %2, %%mm0 \n\t" | ||
| 654 | "pand %2, %%mm3 \n\t" | ||
| 655 | "psrlq $6, %%mm1 \n\t" | ||
| 656 | "psrlq $6, %%mm4 \n\t" | ||
| 657 | "pand %%mm6, %%mm1 \n\t" | ||
| 658 | "pand %%mm6, %%mm4 \n\t" | ||
| 659 | "psrlq $9, %%mm2 \n\t" | ||
| 660 | "psrlq $9, %%mm5 \n\t" | ||
| 661 | "pand %%mm7, %%mm2 \n\t" | ||
| 662 | "pand %%mm7, %%mm5 \n\t" | ||
| 663 | "por %%mm1, %%mm0 \n\t" | ||
| 664 | "por %%mm4, %%mm3 \n\t" | ||
| 665 | "por %%mm2, %%mm0 \n\t" | ||
| 666 | "por %%mm5, %%mm3 \n\t" | ||
| 667 | "psllq $16, %%mm3 \n\t" | ||
| 668 | "por %%mm3, %%mm0 \n\t" | ||
| 669 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 670 | ::"r"(d),"r"(s),"m"(blue_15mask):"memory"); | ||
| 671 | ✗ | d += 4; | |
| 672 | ✗ | s += 12; | |
| 673 | } | ||
| 674 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 675 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 676 | ✗ | while (s < end) { | |
| 677 | ✗ | const int b = *s++; | |
| 678 | ✗ | const int g = *s++; | |
| 679 | ✗ | const int r = *s++; | |
| 680 | ✗ | *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7); | |
| 681 | } | ||
| 682 | ✗ | } | |
| 683 | |||
| 684 | ✗ | static inline void rgb24to15_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 685 | { | ||
| 686 | ✗ | const uint8_t *s = src; | |
| 687 | const uint8_t *end; | ||
| 688 | const uint8_t *mm_end; | ||
| 689 | ✗ | uint16_t *d = (uint16_t *)dst; | |
| 690 | ✗ | end = s + src_size; | |
| 691 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*src):"memory"); | |
| 692 | ✗ | __asm__ volatile( | |
| 693 | "movq %0, %%mm7 \n\t" | ||
| 694 | "movq %1, %%mm6 \n\t" | ||
| 695 | ::"m"(red_15mask),"m"(green_15mask)); | ||
| 696 | ✗ | mm_end = end - 15; | |
| 697 | ✗ | while (s < mm_end) { | |
| 698 | ✗ | __asm__ volatile( | |
| 699 | PREFETCH" 32(%1) \n\t" | ||
| 700 | "movd (%1), %%mm0 \n\t" | ||
| 701 | "movd 3(%1), %%mm3 \n\t" | ||
| 702 | "punpckldq 6(%1), %%mm0 \n\t" | ||
| 703 | "punpckldq 9(%1), %%mm3 \n\t" | ||
| 704 | "movq %%mm0, %%mm1 \n\t" | ||
| 705 | "movq %%mm0, %%mm2 \n\t" | ||
| 706 | "movq %%mm3, %%mm4 \n\t" | ||
| 707 | "movq %%mm3, %%mm5 \n\t" | ||
| 708 | "psllq $7, %%mm0 \n\t" | ||
| 709 | "psllq $7, %%mm3 \n\t" | ||
| 710 | "pand %%mm7, %%mm0 \n\t" | ||
| 711 | "pand %%mm7, %%mm3 \n\t" | ||
| 712 | "psrlq $6, %%mm1 \n\t" | ||
| 713 | "psrlq $6, %%mm4 \n\t" | ||
| 714 | "pand %%mm6, %%mm1 \n\t" | ||
| 715 | "pand %%mm6, %%mm4 \n\t" | ||
| 716 | "psrlq $19, %%mm2 \n\t" | ||
| 717 | "psrlq $19, %%mm5 \n\t" | ||
| 718 | "pand %2, %%mm2 \n\t" | ||
| 719 | "pand %2, %%mm5 \n\t" | ||
| 720 | "por %%mm1, %%mm0 \n\t" | ||
| 721 | "por %%mm4, %%mm3 \n\t" | ||
| 722 | "por %%mm2, %%mm0 \n\t" | ||
| 723 | "por %%mm5, %%mm3 \n\t" | ||
| 724 | "psllq $16, %%mm3 \n\t" | ||
| 725 | "por %%mm3, %%mm0 \n\t" | ||
| 726 | MOVNTQ" %%mm0, (%0) \n\t" | ||
| 727 | ::"r"(d),"r"(s),"m"(blue_15mask):"memory"); | ||
| 728 | ✗ | d += 4; | |
| 729 | ✗ | s += 12; | |
| 730 | } | ||
| 731 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 732 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 733 | ✗ | while (s < end) { | |
| 734 | ✗ | const int r = *s++; | |
| 735 | ✗ | const int g = *s++; | |
| 736 | ✗ | const int b = *s++; | |
| 737 | ✗ | *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7); | |
| 738 | } | ||
| 739 | ✗ | } | |
| 740 | |||
| 741 | ✗ | static inline void rgb15tobgr24_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 742 | { | ||
| 743 | const uint16_t *end; | ||
| 744 | const uint16_t *mm_end; | ||
| 745 | ✗ | uint8_t *d = dst; | |
| 746 | ✗ | const uint16_t *s = (const uint16_t*)src; | |
| 747 | ✗ | end = s + src_size/2; | |
| 748 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 749 | ✗ | mm_end = end - 7; | |
| 750 | ✗ | while (s < mm_end) { | |
| 751 | ✗ | __asm__ volatile( | |
| 752 | PREFETCH" 32(%1) \n\t" | ||
| 753 | "movq (%1), %%mm0 \n\t" | ||
| 754 | "movq (%1), %%mm1 \n\t" | ||
| 755 | "movq (%1), %%mm2 \n\t" | ||
| 756 | "pand %2, %%mm0 \n\t" | ||
| 757 | "pand %3, %%mm1 \n\t" | ||
| 758 | "pand %4, %%mm2 \n\t" | ||
| 759 | "psllq $5, %%mm0 \n\t" | ||
| 760 | "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t" | ||
| 761 | "pmulhw "MANGLE(mul15_mid)", %%mm1 \n\t" | ||
| 762 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 763 | "movq %%mm0, %%mm3 \n\t" | ||
| 764 | "movq %%mm1, %%mm4 \n\t" | ||
| 765 | "movq %%mm2, %%mm5 \n\t" | ||
| 766 | "punpcklwd %5, %%mm0 \n\t" | ||
| 767 | "punpcklwd %5, %%mm1 \n\t" | ||
| 768 | "punpcklwd %5, %%mm2 \n\t" | ||
| 769 | "punpckhwd %5, %%mm3 \n\t" | ||
| 770 | "punpckhwd %5, %%mm4 \n\t" | ||
| 771 | "punpckhwd %5, %%mm5 \n\t" | ||
| 772 | "psllq $8, %%mm1 \n\t" | ||
| 773 | "psllq $16, %%mm2 \n\t" | ||
| 774 | "por %%mm1, %%mm0 \n\t" | ||
| 775 | "por %%mm2, %%mm0 \n\t" | ||
| 776 | "psllq $8, %%mm4 \n\t" | ||
| 777 | "psllq $16, %%mm5 \n\t" | ||
| 778 | "por %%mm4, %%mm3 \n\t" | ||
| 779 | "por %%mm5, %%mm3 \n\t" | ||
| 780 | |||
| 781 | "movq %%mm0, %%mm6 \n\t" | ||
| 782 | "movq %%mm3, %%mm7 \n\t" | ||
| 783 | |||
| 784 | "movq 8(%1), %%mm0 \n\t" | ||
| 785 | "movq 8(%1), %%mm1 \n\t" | ||
| 786 | "movq 8(%1), %%mm2 \n\t" | ||
| 787 | "pand %2, %%mm0 \n\t" | ||
| 788 | "pand %3, %%mm1 \n\t" | ||
| 789 | "pand %4, %%mm2 \n\t" | ||
| 790 | "psllq $5, %%mm0 \n\t" | ||
| 791 | "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t" | ||
| 792 | "pmulhw "MANGLE(mul15_mid)", %%mm1 \n\t" | ||
| 793 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 794 | "movq %%mm0, %%mm3 \n\t" | ||
| 795 | "movq %%mm1, %%mm4 \n\t" | ||
| 796 | "movq %%mm2, %%mm5 \n\t" | ||
| 797 | "punpcklwd %5, %%mm0 \n\t" | ||
| 798 | "punpcklwd %5, %%mm1 \n\t" | ||
| 799 | "punpcklwd %5, %%mm2 \n\t" | ||
| 800 | "punpckhwd %5, %%mm3 \n\t" | ||
| 801 | "punpckhwd %5, %%mm4 \n\t" | ||
| 802 | "punpckhwd %5, %%mm5 \n\t" | ||
| 803 | "psllq $8, %%mm1 \n\t" | ||
| 804 | "psllq $16, %%mm2 \n\t" | ||
| 805 | "por %%mm1, %%mm0 \n\t" | ||
| 806 | "por %%mm2, %%mm0 \n\t" | ||
| 807 | "psllq $8, %%mm4 \n\t" | ||
| 808 | "psllq $16, %%mm5 \n\t" | ||
| 809 | "por %%mm4, %%mm3 \n\t" | ||
| 810 | "por %%mm5, %%mm3 \n\t" | ||
| 811 | |||
| 812 | :"=m"(*d) | ||
| 813 | :"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null) | ||
| 814 | NAMED_CONSTRAINTS_ADD(mul15_mid,mul15_hi) | ||
| 815 | :"memory"); | ||
| 816 | /* borrowed 32 to 24 */ | ||
| 817 | ✗ | __asm__ volatile( | |
| 818 | "movq %%mm0, %%mm4 \n\t" | ||
| 819 | "movq %%mm3, %%mm5 \n\t" | ||
| 820 | "movq %%mm6, %%mm0 \n\t" | ||
| 821 | "movq %%mm7, %%mm1 \n\t" | ||
| 822 | |||
| 823 | "movq %%mm4, %%mm6 \n\t" | ||
| 824 | "movq %%mm5, %%mm7 \n\t" | ||
| 825 | "movq %%mm0, %%mm2 \n\t" | ||
| 826 | "movq %%mm1, %%mm3 \n\t" | ||
| 827 | |||
| 828 | STORE_BGR24_MMX | ||
| 829 | |||
| 830 | :: "r"(d), "m"(*s) | ||
| 831 | NAMED_CONSTRAINTS_ADD(mask24l,mask24h) | ||
| 832 | :"memory"); | ||
| 833 | ✗ | d += 24; | |
| 834 | ✗ | s += 8; | |
| 835 | } | ||
| 836 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 837 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 838 | ✗ | while (s < end) { | |
| 839 | register uint16_t bgr; | ||
| 840 | ✗ | bgr = *s++; | |
| 841 | ✗ | *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2); | |
| 842 | ✗ | *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7); | |
| 843 | ✗ | *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12); | |
| 844 | } | ||
| 845 | ✗ | } | |
| 846 | |||
| 847 | ✗ | static inline void rgb16tobgr24_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 848 | { | ||
| 849 | const uint16_t *end; | ||
| 850 | const uint16_t *mm_end; | ||
| 851 | ✗ | uint8_t *d = (uint8_t *)dst; | |
| 852 | ✗ | const uint16_t *s = (const uint16_t *)src; | |
| 853 | ✗ | end = s + src_size/2; | |
| 854 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 855 | ✗ | mm_end = end - 7; | |
| 856 | ✗ | while (s < mm_end) { | |
| 857 | ✗ | __asm__ volatile( | |
| 858 | PREFETCH" 32(%1) \n\t" | ||
| 859 | "movq (%1), %%mm0 \n\t" | ||
| 860 | "movq (%1), %%mm1 \n\t" | ||
| 861 | "movq (%1), %%mm2 \n\t" | ||
| 862 | "pand %2, %%mm0 \n\t" | ||
| 863 | "pand %3, %%mm1 \n\t" | ||
| 864 | "pand %4, %%mm2 \n\t" | ||
| 865 | "psllq $5, %%mm0 \n\t" | ||
| 866 | "psrlq $1, %%mm2 \n\t" | ||
| 867 | "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t" | ||
| 868 | "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t" | ||
| 869 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 870 | "movq %%mm0, %%mm3 \n\t" | ||
| 871 | "movq %%mm1, %%mm4 \n\t" | ||
| 872 | "movq %%mm2, %%mm5 \n\t" | ||
| 873 | "punpcklwd %5, %%mm0 \n\t" | ||
| 874 | "punpcklwd %5, %%mm1 \n\t" | ||
| 875 | "punpcklwd %5, %%mm2 \n\t" | ||
| 876 | "punpckhwd %5, %%mm3 \n\t" | ||
| 877 | "punpckhwd %5, %%mm4 \n\t" | ||
| 878 | "punpckhwd %5, %%mm5 \n\t" | ||
| 879 | "psllq $8, %%mm1 \n\t" | ||
| 880 | "psllq $16, %%mm2 \n\t" | ||
| 881 | "por %%mm1, %%mm0 \n\t" | ||
| 882 | "por %%mm2, %%mm0 \n\t" | ||
| 883 | "psllq $8, %%mm4 \n\t" | ||
| 884 | "psllq $16, %%mm5 \n\t" | ||
| 885 | "por %%mm4, %%mm3 \n\t" | ||
| 886 | "por %%mm5, %%mm3 \n\t" | ||
| 887 | |||
| 888 | "movq %%mm0, %%mm6 \n\t" | ||
| 889 | "movq %%mm3, %%mm7 \n\t" | ||
| 890 | |||
| 891 | "movq 8(%1), %%mm0 \n\t" | ||
| 892 | "movq 8(%1), %%mm1 \n\t" | ||
| 893 | "movq 8(%1), %%mm2 \n\t" | ||
| 894 | "pand %2, %%mm0 \n\t" | ||
| 895 | "pand %3, %%mm1 \n\t" | ||
| 896 | "pand %4, %%mm2 \n\t" | ||
| 897 | "psllq $5, %%mm0 \n\t" | ||
| 898 | "psrlq $1, %%mm2 \n\t" | ||
| 899 | "pmulhw "MANGLE(mul15_mid)", %%mm0 \n\t" | ||
| 900 | "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t" | ||
| 901 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 902 | "movq %%mm0, %%mm3 \n\t" | ||
| 903 | "movq %%mm1, %%mm4 \n\t" | ||
| 904 | "movq %%mm2, %%mm5 \n\t" | ||
| 905 | "punpcklwd %5, %%mm0 \n\t" | ||
| 906 | "punpcklwd %5, %%mm1 \n\t" | ||
| 907 | "punpcklwd %5, %%mm2 \n\t" | ||
| 908 | "punpckhwd %5, %%mm3 \n\t" | ||
| 909 | "punpckhwd %5, %%mm4 \n\t" | ||
| 910 | "punpckhwd %5, %%mm5 \n\t" | ||
| 911 | "psllq $8, %%mm1 \n\t" | ||
| 912 | "psllq $16, %%mm2 \n\t" | ||
| 913 | "por %%mm1, %%mm0 \n\t" | ||
| 914 | "por %%mm2, %%mm0 \n\t" | ||
| 915 | "psllq $8, %%mm4 \n\t" | ||
| 916 | "psllq $16, %%mm5 \n\t" | ||
| 917 | "por %%mm4, %%mm3 \n\t" | ||
| 918 | "por %%mm5, %%mm3 \n\t" | ||
| 919 | :"=m"(*d) | ||
| 920 | :"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null) | ||
| 921 | NAMED_CONSTRAINTS_ADD(mul15_mid,mul16_mid,mul15_hi) | ||
| 922 | :"memory"); | ||
| 923 | /* borrowed 32 to 24 */ | ||
| 924 | ✗ | __asm__ volatile( | |
| 925 | "movq %%mm0, %%mm4 \n\t" | ||
| 926 | "movq %%mm3, %%mm5 \n\t" | ||
| 927 | "movq %%mm6, %%mm0 \n\t" | ||
| 928 | "movq %%mm7, %%mm1 \n\t" | ||
| 929 | |||
| 930 | "movq %%mm4, %%mm6 \n\t" | ||
| 931 | "movq %%mm5, %%mm7 \n\t" | ||
| 932 | "movq %%mm0, %%mm2 \n\t" | ||
| 933 | "movq %%mm1, %%mm3 \n\t" | ||
| 934 | |||
| 935 | STORE_BGR24_MMX | ||
| 936 | |||
| 937 | :: "r"(d), "m"(*s) | ||
| 938 | NAMED_CONSTRAINTS_ADD(mask24l,mask24h) | ||
| 939 | :"memory"); | ||
| 940 | ✗ | d += 24; | |
| 941 | ✗ | s += 8; | |
| 942 | } | ||
| 943 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 944 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 945 | ✗ | while (s < end) { | |
| 946 | register uint16_t bgr; | ||
| 947 | ✗ | bgr = *s++; | |
| 948 | ✗ | *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2); | |
| 949 | ✗ | *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9); | |
| 950 | ✗ | *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13); | |
| 951 | } | ||
| 952 | ✗ | } | |
| 953 | |||
| 954 | /* | ||
| 955 | * mm0 = 00 B3 00 B2 00 B1 00 B0 | ||
| 956 | * mm1 = 00 G3 00 G2 00 G1 00 G0 | ||
| 957 | * mm2 = 00 R3 00 R2 00 R1 00 R0 | ||
| 958 | * mm6 = FF FF FF FF FF FF FF FF | ||
| 959 | * mm7 = 00 00 00 00 00 00 00 00 | ||
| 960 | */ | ||
| 961 | #define PACK_RGB32 \ | ||
| 962 | "packuswb %%mm7, %%mm0 \n\t" /* 00 00 00 00 B3 B2 B1 B0 */ \ | ||
| 963 | "packuswb %%mm7, %%mm1 \n\t" /* 00 00 00 00 G3 G2 G1 G0 */ \ | ||
| 964 | "packuswb %%mm7, %%mm2 \n\t" /* 00 00 00 00 R3 R2 R1 R0 */ \ | ||
| 965 | "punpcklbw %%mm1, %%mm0 \n\t" /* G3 B3 G2 B2 G1 B1 G0 B0 */ \ | ||
| 966 | "punpcklbw %%mm6, %%mm2 \n\t" /* FF R3 FF R2 FF R1 FF R0 */ \ | ||
| 967 | "movq %%mm0, %%mm3 \n\t" \ | ||
| 968 | "punpcklwd %%mm2, %%mm0 \n\t" /* FF R1 G1 B1 FF R0 G0 B0 */ \ | ||
| 969 | "punpckhwd %%mm2, %%mm3 \n\t" /* FF R3 G3 B3 FF R2 G2 B2 */ \ | ||
| 970 | MOVNTQ" %%mm0, (%0) \n\t" \ | ||
| 971 | MOVNTQ" %%mm3, 8(%0) \n\t" \ | ||
| 972 | |||
| 973 | ✗ | static inline void rgb15to32_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 974 | { | ||
| 975 | const uint16_t *end; | ||
| 976 | const uint16_t *mm_end; | ||
| 977 | ✗ | uint8_t *d = dst; | |
| 978 | ✗ | const uint16_t *s = (const uint16_t *)src; | |
| 979 | ✗ | end = s + src_size/2; | |
| 980 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 981 | ✗ | __asm__ volatile("pxor %%mm7,%%mm7 \n\t":::"memory"); | |
| 982 | ✗ | __asm__ volatile("pcmpeqd %%mm6,%%mm6 \n\t":::"memory"); | |
| 983 | ✗ | mm_end = end - 3; | |
| 984 | ✗ | while (s < mm_end) { | |
| 985 | ✗ | __asm__ volatile( | |
| 986 | PREFETCH" 32(%1) \n\t" | ||
| 987 | "movq (%1), %%mm0 \n\t" | ||
| 988 | "movq (%1), %%mm1 \n\t" | ||
| 989 | "movq (%1), %%mm2 \n\t" | ||
| 990 | "pand %2, %%mm0 \n\t" | ||
| 991 | "pand %3, %%mm1 \n\t" | ||
| 992 | "pand %4, %%mm2 \n\t" | ||
| 993 | "psllq $5, %%mm0 \n\t" | ||
| 994 | "pmulhw %5, %%mm0 \n\t" | ||
| 995 | "pmulhw %5, %%mm1 \n\t" | ||
| 996 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 997 | PACK_RGB32 | ||
| 998 | ::"r"(d),"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r) ,"m"(mul15_mid) | ||
| 999 | NAMED_CONSTRAINTS_ADD(mul15_hi) | ||
| 1000 | :"memory"); | ||
| 1001 | ✗ | d += 16; | |
| 1002 | ✗ | s += 4; | |
| 1003 | } | ||
| 1004 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 1005 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 1006 | ✗ | while (s < end) { | |
| 1007 | register uint16_t bgr; | ||
| 1008 | ✗ | bgr = *s++; | |
| 1009 | ✗ | *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2); | |
| 1010 | ✗ | *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7); | |
| 1011 | ✗ | *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12); | |
| 1012 | ✗ | *d++ = 255; | |
| 1013 | } | ||
| 1014 | ✗ | } | |
| 1015 | |||
| 1016 | ✗ | static inline void rgb16to32_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 1017 | { | ||
| 1018 | const uint16_t *end; | ||
| 1019 | const uint16_t *mm_end; | ||
| 1020 | ✗ | uint8_t *d = dst; | |
| 1021 | ✗ | const uint16_t *s = (const uint16_t*)src; | |
| 1022 | ✗ | end = s + src_size/2; | |
| 1023 | ✗ | __asm__ volatile(PREFETCH" %0"::"m"(*s):"memory"); | |
| 1024 | ✗ | __asm__ volatile("pxor %%mm7,%%mm7 \n\t":::"memory"); | |
| 1025 | ✗ | __asm__ volatile("pcmpeqd %%mm6,%%mm6 \n\t":::"memory"); | |
| 1026 | ✗ | mm_end = end - 3; | |
| 1027 | ✗ | while (s < mm_end) { | |
| 1028 | ✗ | __asm__ volatile( | |
| 1029 | PREFETCH" 32(%1) \n\t" | ||
| 1030 | "movq (%1), %%mm0 \n\t" | ||
| 1031 | "movq (%1), %%mm1 \n\t" | ||
| 1032 | "movq (%1), %%mm2 \n\t" | ||
| 1033 | "pand %2, %%mm0 \n\t" | ||
| 1034 | "pand %3, %%mm1 \n\t" | ||
| 1035 | "pand %4, %%mm2 \n\t" | ||
| 1036 | "psllq $5, %%mm0 \n\t" | ||
| 1037 | "psrlq $1, %%mm2 \n\t" | ||
| 1038 | "pmulhw %5, %%mm0 \n\t" | ||
| 1039 | "pmulhw "MANGLE(mul16_mid)", %%mm1 \n\t" | ||
| 1040 | "pmulhw "MANGLE(mul15_hi)", %%mm2 \n\t" | ||
| 1041 | PACK_RGB32 | ||
| 1042 | ::"r"(d),"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mul15_mid) | ||
| 1043 | NAMED_CONSTRAINTS_ADD(mul16_mid,mul15_hi) | ||
| 1044 | :"memory"); | ||
| 1045 | ✗ | d += 16; | |
| 1046 | ✗ | s += 4; | |
| 1047 | } | ||
| 1048 | ✗ | __asm__ volatile(SFENCE:::"memory"); | |
| 1049 | ✗ | __asm__ volatile(EMMS:::"memory"); | |
| 1050 | ✗ | while (s < end) { | |
| 1051 | register uint16_t bgr; | ||
| 1052 | ✗ | bgr = *s++; | |
| 1053 | ✗ | *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2); | |
| 1054 | ✗ | *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9); | |
| 1055 | ✗ | *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13); | |
| 1056 | ✗ | *d++ = 255; | |
| 1057 | } | ||
| 1058 | ✗ | } | |
| 1059 | |||
| 1060 | 7 | static inline void rgb24tobgr24_mmxext(const uint8_t *src, uint8_t *dst, int src_size) | |
| 1061 | { | ||
| 1062 | 7 | x86_reg mmx_size= 23 - src_size; | |
| 1063 | 7 | __asm__ volatile ( | |
| 1064 | "test %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 1065 | "jns 2f \n\t" | ||
| 1066 | "movq "MANGLE(mask24r)", %%mm5 \n\t" | ||
| 1067 | "movq "MANGLE(mask24g)", %%mm6 \n\t" | ||
| 1068 | "movq "MANGLE(mask24b)", %%mm7 \n\t" | ||
| 1069 | ".p2align 4 \n\t" | ||
| 1070 | "1: \n\t" | ||
| 1071 | PREFETCH" 32(%1, %%"FF_REG_a") \n\t" | ||
| 1072 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" // BGR BGR BG | ||
| 1073 | "movq (%1, %%"FF_REG_a"), %%mm1 \n\t" // BGR BGR BG | ||
| 1074 | "movq 2(%1, %%"FF_REG_a"), %%mm2 \n\t" // R BGR BGR B | ||
| 1075 | "psllq $16, %%mm0 \n\t" // 00 BGR BGR | ||
| 1076 | "pand %%mm5, %%mm0 \n\t" | ||
| 1077 | "pand %%mm6, %%mm1 \n\t" | ||
| 1078 | "pand %%mm7, %%mm2 \n\t" | ||
| 1079 | "por %%mm0, %%mm1 \n\t" | ||
| 1080 | "por %%mm2, %%mm1 \n\t" | ||
| 1081 | "movq 6(%1, %%"FF_REG_a"), %%mm0 \n\t" // BGR BGR BG | ||
| 1082 | MOVNTQ" %%mm1,(%2, %%"FF_REG_a") \n\t" // RGB RGB RG | ||
| 1083 | "movq 8(%1, %%"FF_REG_a"), %%mm1 \n\t" // R BGR BGR B | ||
| 1084 | "movq 10(%1, %%"FF_REG_a"), %%mm2 \n\t" // GR BGR BGR | ||
| 1085 | "pand %%mm7, %%mm0 \n\t" | ||
| 1086 | "pand %%mm5, %%mm1 \n\t" | ||
| 1087 | "pand %%mm6, %%mm2 \n\t" | ||
| 1088 | "por %%mm0, %%mm1 \n\t" | ||
| 1089 | "por %%mm2, %%mm1 \n\t" | ||
| 1090 | "movq 14(%1, %%"FF_REG_a"), %%mm0 \n\t" // R BGR BGR B | ||
| 1091 | MOVNTQ" %%mm1, 8(%2, %%"FF_REG_a")\n\t" // B RGB RGB R | ||
| 1092 | "movq 16(%1, %%"FF_REG_a"), %%mm1 \n\t" // GR BGR BGR | ||
| 1093 | "movq 18(%1, %%"FF_REG_a"), %%mm2 \n\t" // BGR BGR BG | ||
| 1094 | "pand %%mm6, %%mm0 \n\t" | ||
| 1095 | "pand %%mm7, %%mm1 \n\t" | ||
| 1096 | "pand %%mm5, %%mm2 \n\t" | ||
| 1097 | "por %%mm0, %%mm1 \n\t" | ||
| 1098 | "por %%mm2, %%mm1 \n\t" | ||
| 1099 | MOVNTQ" %%mm1, 16(%2, %%"FF_REG_a") \n\t" | ||
| 1100 | "add $24, %%"FF_REG_a" \n\t" | ||
| 1101 | " js 1b \n\t" | ||
| 1102 | "2: \n\t" | ||
| 1103 | : "+a" (mmx_size) | ||
| 1104 | 7 | : "r" (src-mmx_size), "r"(dst-mmx_size) | |
| 1105 | NAMED_CONSTRAINTS_ADD(mask24r,mask24g,mask24b) | ||
| 1106 | ); | ||
| 1107 | |||
| 1108 | 7 | __asm__ volatile(SFENCE:::"memory"); | |
| 1109 | 7 | __asm__ volatile(EMMS:::"memory"); | |
| 1110 | |||
| 1111 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 4 times.
|
7 | if (mmx_size==23) return; //finished, was multiple of 8 |
| 1112 | |||
| 1113 | 4 | src+= src_size; | |
| 1114 | 4 | dst+= src_size; | |
| 1115 | 4 | src_size= 23-mmx_size; | |
| 1116 | 4 | src-= src_size; | |
| 1117 | 4 | dst-= src_size; | |
| 1118 |
2/2✓ Branch 0 taken 11 times.
✓ Branch 1 taken 4 times.
|
15 | for (unsigned i = 0; i < src_size; i +=3) { |
| 1119 | register uint8_t x; | ||
| 1120 | 11 | x = src[i + 2]; | |
| 1121 | 11 | dst[i + 1] = src[i + 1]; | |
| 1122 | 11 | dst[i + 2] = src[i + 0]; | |
| 1123 | 11 | dst[i + 0] = x; | |
| 1124 | } | ||
| 1125 | } | ||
| 1126 | |||
| 1127 | ✗ | static inline void yuvPlanartoyuy2_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1128 | int width, int height, | ||
| 1129 | int lumStride, int chromStride, int dstStride, int vertLumPerChroma) | ||
| 1130 | { | ||
| 1131 | ✗ | const x86_reg chromWidth= width>>1; | |
| 1132 | ✗ | for (int y = 0; y < height; y++) { | |
| 1133 | //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway) | ||
| 1134 | ✗ | __asm__ volatile( | |
| 1135 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 1136 | ".p2align 4 \n\t" | ||
| 1137 | "1: \n\t" | ||
| 1138 | PREFETCH" 32(%1, %%"FF_REG_a", 2) \n\t" | ||
| 1139 | PREFETCH" 32(%2, %%"FF_REG_a") \n\t" | ||
| 1140 | PREFETCH" 32(%3, %%"FF_REG_a") \n\t" | ||
| 1141 | "movq (%2, %%"FF_REG_a"), %%mm0 \n\t" // U(0) | ||
| 1142 | "movq %%mm0, %%mm2 \n\t" // U(0) | ||
| 1143 | "movq (%3, %%"FF_REG_a"), %%mm1 \n\t" // V(0) | ||
| 1144 | "punpcklbw %%mm1, %%mm0 \n\t" // UVUV UVUV(0) | ||
| 1145 | "punpckhbw %%mm1, %%mm2 \n\t" // UVUV UVUV(8) | ||
| 1146 | |||
| 1147 | "movq (%1, %%"FF_REG_a",2), %%mm3 \n\t" // Y(0) | ||
| 1148 | "movq 8(%1, %%"FF_REG_a",2), %%mm5 \n\t" // Y(8) | ||
| 1149 | "movq %%mm3, %%mm4 \n\t" // Y(0) | ||
| 1150 | "movq %%mm5, %%mm6 \n\t" // Y(8) | ||
| 1151 | "punpcklbw %%mm0, %%mm3 \n\t" // YUYV YUYV(0) | ||
| 1152 | "punpckhbw %%mm0, %%mm4 \n\t" // YUYV YUYV(4) | ||
| 1153 | "punpcklbw %%mm2, %%mm5 \n\t" // YUYV YUYV(8) | ||
| 1154 | "punpckhbw %%mm2, %%mm6 \n\t" // YUYV YUYV(12) | ||
| 1155 | |||
| 1156 | MOVNTQ" %%mm3, (%0, %%"FF_REG_a", 4) \n\t" | ||
| 1157 | MOVNTQ" %%mm4, 8(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1158 | MOVNTQ" %%mm5, 16(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1159 | MOVNTQ" %%mm6, 24(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1160 | |||
| 1161 | "add $8, %%"FF_REG_a" \n\t" | ||
| 1162 | "cmp %4, %%"FF_REG_a" \n\t" | ||
| 1163 | " jb 1b \n\t" | ||
| 1164 | ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth) | ||
| 1165 | : "%"FF_REG_a | ||
| 1166 | ); | ||
| 1167 | ✗ | if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { | |
| 1168 | ✗ | usrc += chromStride; | |
| 1169 | ✗ | vsrc += chromStride; | |
| 1170 | } | ||
| 1171 | ✗ | ysrc += lumStride; | |
| 1172 | ✗ | dst += dstStride; | |
| 1173 | } | ||
| 1174 | ✗ | __asm__(EMMS" \n\t" | |
| 1175 | SFENCE" \n\t" | ||
| 1176 | :::"memory"); | ||
| 1177 | ✗ | } | |
| 1178 | |||
| 1179 | /** | ||
| 1180 | * Height should be a multiple of 2 and width should be a multiple of 16. | ||
| 1181 | * (If this is a problem for anyone then tell me, and I will fix it.) | ||
| 1182 | */ | ||
| 1183 | ✗ | static inline void yv12toyuy2_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1184 | int width, int height, | ||
| 1185 | int lumStride, int chromStride, int dstStride) | ||
| 1186 | { | ||
| 1187 | //FIXME interpolate chroma | ||
| 1188 | ✗ | yuvPlanartoyuy2_mmxext(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2); | |
| 1189 | ✗ | } | |
| 1190 | |||
| 1191 | ✗ | static inline void yuvPlanartouyvy_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1192 | int width, int height, | ||
| 1193 | int lumStride, int chromStride, int dstStride, int vertLumPerChroma) | ||
| 1194 | { | ||
| 1195 | ✗ | const x86_reg chromWidth= width>>1; | |
| 1196 | ✗ | for (int y = 0; y < height; y++) { | |
| 1197 | //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway) | ||
| 1198 | ✗ | __asm__ volatile( | |
| 1199 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 1200 | ".p2align 4 \n\t" | ||
| 1201 | "1: \n\t" | ||
| 1202 | PREFETCH" 32(%1, %%"FF_REG_a", 2) \n\t" | ||
| 1203 | PREFETCH" 32(%2, %%"FF_REG_a") \n\t" | ||
| 1204 | PREFETCH" 32(%3, %%"FF_REG_a") \n\t" | ||
| 1205 | "movq (%2, %%"FF_REG_a"), %%mm0 \n\t" // U(0) | ||
| 1206 | "movq %%mm0, %%mm2 \n\t" // U(0) | ||
| 1207 | "movq (%3, %%"FF_REG_a"), %%mm1 \n\t" // V(0) | ||
| 1208 | "punpcklbw %%mm1, %%mm0 \n\t" // UVUV UVUV(0) | ||
| 1209 | "punpckhbw %%mm1, %%mm2 \n\t" // UVUV UVUV(8) | ||
| 1210 | |||
| 1211 | "movq (%1, %%"FF_REG_a",2), %%mm3 \n\t" // Y(0) | ||
| 1212 | "movq 8(%1, %%"FF_REG_a",2), %%mm5 \n\t" // Y(8) | ||
| 1213 | "movq %%mm0, %%mm4 \n\t" // Y(0) | ||
| 1214 | "movq %%mm2, %%mm6 \n\t" // Y(8) | ||
| 1215 | "punpcklbw %%mm3, %%mm0 \n\t" // YUYV YUYV(0) | ||
| 1216 | "punpckhbw %%mm3, %%mm4 \n\t" // YUYV YUYV(4) | ||
| 1217 | "punpcklbw %%mm5, %%mm2 \n\t" // YUYV YUYV(8) | ||
| 1218 | "punpckhbw %%mm5, %%mm6 \n\t" // YUYV YUYV(12) | ||
| 1219 | |||
| 1220 | MOVNTQ" %%mm0, (%0, %%"FF_REG_a", 4) \n\t" | ||
| 1221 | MOVNTQ" %%mm4, 8(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1222 | MOVNTQ" %%mm2, 16(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1223 | MOVNTQ" %%mm6, 24(%0, %%"FF_REG_a", 4) \n\t" | ||
| 1224 | |||
| 1225 | "add $8, %%"FF_REG_a" \n\t" | ||
| 1226 | "cmp %4, %%"FF_REG_a" \n\t" | ||
| 1227 | " jb 1b \n\t" | ||
| 1228 | ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth) | ||
| 1229 | : "%"FF_REG_a | ||
| 1230 | ); | ||
| 1231 | ✗ | if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) { | |
| 1232 | ✗ | usrc += chromStride; | |
| 1233 | ✗ | vsrc += chromStride; | |
| 1234 | } | ||
| 1235 | ✗ | ysrc += lumStride; | |
| 1236 | ✗ | dst += dstStride; | |
| 1237 | } | ||
| 1238 | ✗ | __asm__(EMMS" \n\t" | |
| 1239 | SFENCE" \n\t" | ||
| 1240 | :::"memory"); | ||
| 1241 | ✗ | } | |
| 1242 | |||
| 1243 | /** | ||
| 1244 | * Height should be a multiple of 2 and width should be a multiple of 16 | ||
| 1245 | * (If this is a problem for anyone then tell me, and I will fix it.) | ||
| 1246 | */ | ||
| 1247 | ✗ | static inline void yv12touyvy_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1248 | int width, int height, | ||
| 1249 | int lumStride, int chromStride, int dstStride) | ||
| 1250 | { | ||
| 1251 | //FIXME interpolate chroma | ||
| 1252 | ✗ | yuvPlanartouyvy_mmxext(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2); | |
| 1253 | ✗ | } | |
| 1254 | |||
| 1255 | /** | ||
| 1256 | * Width should be a multiple of 16. | ||
| 1257 | */ | ||
| 1258 | ✗ | static inline void yuv422ptouyvy_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1259 | int width, int height, | ||
| 1260 | int lumStride, int chromStride, int dstStride) | ||
| 1261 | { | ||
| 1262 | ✗ | yuvPlanartouyvy_mmxext(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1); | |
| 1263 | ✗ | } | |
| 1264 | |||
| 1265 | /** | ||
| 1266 | * Width should be a multiple of 16. | ||
| 1267 | */ | ||
| 1268 | ✗ | static inline void yuv422ptoyuy2_mmxext(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, | |
| 1269 | int width, int height, | ||
| 1270 | int lumStride, int chromStride, int dstStride) | ||
| 1271 | { | ||
| 1272 | ✗ | yuvPlanartoyuy2_mmxext(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1); | |
| 1273 | ✗ | } | |
| 1274 | |||
| 1275 | ✗ | static inline void planar2x_mmxext(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride) | |
| 1276 | { | ||
| 1277 | ✗ | dst[0]= src[0]; | |
| 1278 | |||
| 1279 | // first line | ||
| 1280 | ✗ | for (int x = 0; x < srcWidth - 1; x++) { | |
| 1281 | ✗ | dst[2*x+1]= (3*src[x] + src[x+1])>>2; | |
| 1282 | ✗ | dst[2*x+2]= ( src[x] + 3*src[x+1])>>2; | |
| 1283 | } | ||
| 1284 | ✗ | dst[2*srcWidth-1]= src[srcWidth-1]; | |
| 1285 | |||
| 1286 | ✗ | dst+= dstStride; | |
| 1287 | |||
| 1288 | ✗ | for (int y = 1; y < srcHeight; y++) { | |
| 1289 | ✗ | x86_reg mmxSize= srcWidth&~15; | |
| 1290 | |||
| 1291 | ✗ | if (mmxSize) { | |
| 1292 | ✗ | __asm__ volatile( | |
| 1293 | "mov %4, %%"FF_REG_a" \n\t" | ||
| 1294 | "movq "MANGLE(mmx_ff)", %%mm0 \n\t" | ||
| 1295 | "movq (%0, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 1296 | "movq %%mm4, %%mm2 \n\t" | ||
| 1297 | "psllq $8, %%mm4 \n\t" | ||
| 1298 | "pand %%mm0, %%mm2 \n\t" | ||
| 1299 | "por %%mm2, %%mm4 \n\t" | ||
| 1300 | "movq (%1, %%"FF_REG_a"), %%mm5 \n\t" | ||
| 1301 | "movq %%mm5, %%mm3 \n\t" | ||
| 1302 | "psllq $8, %%mm5 \n\t" | ||
| 1303 | "pand %%mm0, %%mm3 \n\t" | ||
| 1304 | "por %%mm3, %%mm5 \n\t" | ||
| 1305 | "1: \n\t" | ||
| 1306 | "movq (%0, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 1307 | "movq (%1, %%"FF_REG_a"), %%mm1 \n\t" | ||
| 1308 | "movq 1(%0, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 1309 | "movq 1(%1, %%"FF_REG_a"), %%mm3 \n\t" | ||
| 1310 | PAVGB" %%mm0, %%mm5 \n\t" | ||
| 1311 | PAVGB" %%mm0, %%mm3 \n\t" | ||
| 1312 | PAVGB" %%mm0, %%mm5 \n\t" | ||
| 1313 | PAVGB" %%mm0, %%mm3 \n\t" | ||
| 1314 | PAVGB" %%mm1, %%mm4 \n\t" | ||
| 1315 | PAVGB" %%mm1, %%mm2 \n\t" | ||
| 1316 | PAVGB" %%mm1, %%mm4 \n\t" | ||
| 1317 | PAVGB" %%mm1, %%mm2 \n\t" | ||
| 1318 | "movq %%mm5, %%mm7 \n\t" | ||
| 1319 | "movq %%mm4, %%mm6 \n\t" | ||
| 1320 | "punpcklbw %%mm3, %%mm5 \n\t" | ||
| 1321 | "punpckhbw %%mm3, %%mm7 \n\t" | ||
| 1322 | "punpcklbw %%mm2, %%mm4 \n\t" | ||
| 1323 | "punpckhbw %%mm2, %%mm6 \n\t" | ||
| 1324 | MOVNTQ" %%mm5, (%2, %%"FF_REG_a", 2) \n\t" | ||
| 1325 | MOVNTQ" %%mm7, 8(%2, %%"FF_REG_a", 2) \n\t" | ||
| 1326 | MOVNTQ" %%mm4, (%3, %%"FF_REG_a", 2) \n\t" | ||
| 1327 | MOVNTQ" %%mm6, 8(%3, %%"FF_REG_a", 2) \n\t" | ||
| 1328 | "add $8, %%"FF_REG_a" \n\t" | ||
| 1329 | "movq -1(%0, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 1330 | "movq -1(%1, %%"FF_REG_a"), %%mm5 \n\t" | ||
| 1331 | " js 1b \n\t" | ||
| 1332 | ✗ | :: "r" (src + mmxSize ), "r" (src + srcStride + mmxSize ), | |
| 1333 | ✗ | "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2), | |
| 1334 | ✗ | "g" (-mmxSize) | |
| 1335 | NAMED_CONSTRAINTS_ADD(mmx_ff) | ||
| 1336 | : "%"FF_REG_a | ||
| 1337 | ); | ||
| 1338 | } else { | ||
| 1339 | ✗ | mmxSize = 1; | |
| 1340 | ✗ | dst[0] = (src[0] * 3 + src[srcStride]) >> 2; | |
| 1341 | ✗ | dst[dstStride] = (src[0] + 3 * src[srcStride]) >> 2; | |
| 1342 | } | ||
| 1343 | |||
| 1344 | ✗ | for (int x = mmxSize - 1; x < srcWidth - 1; x++) { | |
| 1345 | ✗ | dst[2*x +1]= (3*src[x+0] + src[x+srcStride+1])>>2; | |
| 1346 | ✗ | dst[2*x+dstStride+2]= ( src[x+0] + 3*src[x+srcStride+1])>>2; | |
| 1347 | ✗ | dst[2*x+dstStride+1]= ( src[x+1] + 3*src[x+srcStride ])>>2; | |
| 1348 | ✗ | dst[2*x +2]= (3*src[x+1] + src[x+srcStride ])>>2; | |
| 1349 | } | ||
| 1350 | ✗ | dst[srcWidth*2 -1 ]= (3*src[srcWidth-1] + src[srcWidth-1 + srcStride])>>2; | |
| 1351 | ✗ | dst[srcWidth*2 -1 + dstStride]= ( src[srcWidth-1] + 3*src[srcWidth-1 + srcStride])>>2; | |
| 1352 | |||
| 1353 | ✗ | dst+=dstStride*2; | |
| 1354 | ✗ | src+=srcStride; | |
| 1355 | } | ||
| 1356 | |||
| 1357 | // last line | ||
| 1358 | ✗ | dst[0]= src[0]; | |
| 1359 | |||
| 1360 | ✗ | for (int x = 0; x < srcWidth - 1; x++) { | |
| 1361 | ✗ | dst[2*x+1]= (3*src[x] + src[x+1])>>2; | |
| 1362 | ✗ | dst[2*x+2]= ( src[x] + 3*src[x+1])>>2; | |
| 1363 | } | ||
| 1364 | ✗ | dst[2*srcWidth-1]= src[srcWidth-1]; | |
| 1365 | |||
| 1366 | ✗ | __asm__ volatile(EMMS" \n\t" | |
| 1367 | SFENCE" \n\t" | ||
| 1368 | :::"memory"); | ||
| 1369 | ✗ | } | |
| 1370 | |||
| 1371 | /** | ||
| 1372 | * Height should be a multiple of 2 and width should be a multiple of 2. | ||
| 1373 | * (If this is a problem for anyone then tell me, and I will fix it.) | ||
| 1374 | * Chrominance data is only taken from every second line, | ||
| 1375 | * others are ignored in the C version. | ||
| 1376 | * FIXME: Write HQ version. | ||
| 1377 | */ | ||
| 1378 | #if ARCH_X86_32 && HAVE_7REGS | ||
| 1379 | DECLARE_ASM_CONST(8, uint64_t, bgr2YOffset) = 0x1010101010101010ULL; | ||
| 1380 | DECLARE_ASM_CONST(8, uint64_t, bgr2UVOffset) = 0x8080808080808080ULL; | ||
| 1381 | DECLARE_ASM_CONST(8, uint64_t, w1111) = 0x0001000100010001ULL; | ||
| 1382 | |||
| 1383 | static inline void rgb24toyv12_mmxext(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, | ||
| 1384 | int width, int height, | ||
| 1385 | int lumStride, int chromStride, int srcStride, | ||
| 1386 | const int32_t *rgb2yuv) | ||
| 1387 | { | ||
| 1388 | #define BGR2Y_IDX "16*4+16*32" | ||
| 1389 | #define BGR2U_IDX "16*4+16*33" | ||
| 1390 | #define BGR2V_IDX "16*4+16*34" | ||
| 1391 | int y; | ||
| 1392 | const x86_reg chromWidth= width>>1; | ||
| 1393 | |||
| 1394 | if (height > 2) { | ||
| 1395 | ff_rgb24toyv12_c(src, ydst, udst, vdst, width, 2, lumStride, chromStride, srcStride, rgb2yuv); | ||
| 1396 | src += 2*srcStride; | ||
| 1397 | ydst += 2*lumStride; | ||
| 1398 | udst += chromStride; | ||
| 1399 | vdst += chromStride; | ||
| 1400 | height -= 2; | ||
| 1401 | } | ||
| 1402 | |||
| 1403 | for (y = 0; y < height - 2; y += 2) { | ||
| 1404 | for (int i = 0; i < 2; i++) { | ||
| 1405 | __asm__ volatile( | ||
| 1406 | "mov %2, %%"FF_REG_a"\n\t" | ||
| 1407 | "movq "BGR2Y_IDX"(%3), %%mm6 \n\t" | ||
| 1408 | "movq "MANGLE(w1111)", %%mm5 \n\t" | ||
| 1409 | "pxor %%mm7, %%mm7 \n\t" | ||
| 1410 | "lea (%%"FF_REG_a", %%"FF_REG_a", 2), %%"FF_REG_d" \n\t" | ||
| 1411 | ".p2align 4 \n\t" | ||
| 1412 | "1: \n\t" | ||
| 1413 | PREFETCH" 64(%0, %%"FF_REG_d") \n\t" | ||
| 1414 | "movd (%0, %%"FF_REG_d"), %%mm0 \n\t" | ||
| 1415 | "movd 3(%0, %%"FF_REG_d"), %%mm1 \n\t" | ||
| 1416 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 1417 | "punpcklbw %%mm7, %%mm1 \n\t" | ||
| 1418 | "movd 6(%0, %%"FF_REG_d"), %%mm2 \n\t" | ||
| 1419 | "movd 9(%0, %%"FF_REG_d"), %%mm3 \n\t" | ||
| 1420 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 1421 | "punpcklbw %%mm7, %%mm3 \n\t" | ||
| 1422 | "pmaddwd %%mm6, %%mm0 \n\t" | ||
| 1423 | "pmaddwd %%mm6, %%mm1 \n\t" | ||
| 1424 | "pmaddwd %%mm6, %%mm2 \n\t" | ||
| 1425 | "pmaddwd %%mm6, %%mm3 \n\t" | ||
| 1426 | "psrad $8, %%mm0 \n\t" | ||
| 1427 | "psrad $8, %%mm1 \n\t" | ||
| 1428 | "psrad $8, %%mm2 \n\t" | ||
| 1429 | "psrad $8, %%mm3 \n\t" | ||
| 1430 | "packssdw %%mm1, %%mm0 \n\t" | ||
| 1431 | "packssdw %%mm3, %%mm2 \n\t" | ||
| 1432 | "pmaddwd %%mm5, %%mm0 \n\t" | ||
| 1433 | "pmaddwd %%mm5, %%mm2 \n\t" | ||
| 1434 | "packssdw %%mm2, %%mm0 \n\t" | ||
| 1435 | "psraw $7, %%mm0 \n\t" | ||
| 1436 | |||
| 1437 | "movd 12(%0, %%"FF_REG_d"), %%mm4 \n\t" | ||
| 1438 | "movd 15(%0, %%"FF_REG_d"), %%mm1 \n\t" | ||
| 1439 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 1440 | "punpcklbw %%mm7, %%mm1 \n\t" | ||
| 1441 | "movd 18(%0, %%"FF_REG_d"), %%mm2 \n\t" | ||
| 1442 | "movd 21(%0, %%"FF_REG_d"), %%mm3 \n\t" | ||
| 1443 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 1444 | "punpcklbw %%mm7, %%mm3 \n\t" | ||
| 1445 | "pmaddwd %%mm6, %%mm4 \n\t" | ||
| 1446 | "pmaddwd %%mm6, %%mm1 \n\t" | ||
| 1447 | "pmaddwd %%mm6, %%mm2 \n\t" | ||
| 1448 | "pmaddwd %%mm6, %%mm3 \n\t" | ||
| 1449 | "psrad $8, %%mm4 \n\t" | ||
| 1450 | "psrad $8, %%mm1 \n\t" | ||
| 1451 | "psrad $8, %%mm2 \n\t" | ||
| 1452 | "psrad $8, %%mm3 \n\t" | ||
| 1453 | "packssdw %%mm1, %%mm4 \n\t" | ||
| 1454 | "packssdw %%mm3, %%mm2 \n\t" | ||
| 1455 | "pmaddwd %%mm5, %%mm4 \n\t" | ||
| 1456 | "pmaddwd %%mm5, %%mm2 \n\t" | ||
| 1457 | "add $24, %%"FF_REG_d"\n\t" | ||
| 1458 | "packssdw %%mm2, %%mm4 \n\t" | ||
| 1459 | "psraw $7, %%mm4 \n\t" | ||
| 1460 | |||
| 1461 | "packuswb %%mm4, %%mm0 \n\t" | ||
| 1462 | "paddusb "MANGLE(bgr2YOffset)", %%mm0 \n\t" | ||
| 1463 | |||
| 1464 | MOVNTQ" %%mm0, (%1, %%"FF_REG_a") \n\t" | ||
| 1465 | "add $8, %%"FF_REG_a" \n\t" | ||
| 1466 | " js 1b \n\t" | ||
| 1467 | : : "r" (src+width*3), "r" (ydst+width), "g" ((x86_reg)-width), "r"(rgb2yuv) | ||
| 1468 | NAMED_CONSTRAINTS_ADD(w1111,bgr2YOffset) | ||
| 1469 | : "%"FF_REG_a, "%"FF_REG_d | ||
| 1470 | ); | ||
| 1471 | ydst += lumStride; | ||
| 1472 | src += srcStride; | ||
| 1473 | } | ||
| 1474 | src -= srcStride*2; | ||
| 1475 | __asm__ volatile( | ||
| 1476 | "mov %4, %%"FF_REG_a"\n\t" | ||
| 1477 | "movq "MANGLE(w1111)", %%mm5 \n\t" | ||
| 1478 | "movq "BGR2U_IDX"(%5), %%mm6 \n\t" | ||
| 1479 | "pxor %%mm7, %%mm7 \n\t" | ||
| 1480 | "lea (%%"FF_REG_a", %%"FF_REG_a", 2), %%"FF_REG_d" \n\t" | ||
| 1481 | "add %%"FF_REG_d", %%"FF_REG_d"\n\t" | ||
| 1482 | ".p2align 4 \n\t" | ||
| 1483 | "1: \n\t" | ||
| 1484 | PREFETCH" 64(%0, %%"FF_REG_d") \n\t" | ||
| 1485 | PREFETCH" 64(%1, %%"FF_REG_d") \n\t" | ||
| 1486 | "movq (%0, %%"FF_REG_d"), %%mm0 \n\t" | ||
| 1487 | "movq (%1, %%"FF_REG_d"), %%mm1 \n\t" | ||
| 1488 | "movq 6(%0, %%"FF_REG_d"), %%mm2 \n\t" | ||
| 1489 | "movq 6(%1, %%"FF_REG_d"), %%mm3 \n\t" | ||
| 1490 | PAVGB" %%mm1, %%mm0 \n\t" | ||
| 1491 | PAVGB" %%mm3, %%mm2 \n\t" | ||
| 1492 | "movq %%mm0, %%mm1 \n\t" | ||
| 1493 | "movq %%mm2, %%mm3 \n\t" | ||
| 1494 | "psrlq $24, %%mm0 \n\t" | ||
| 1495 | "psrlq $24, %%mm2 \n\t" | ||
| 1496 | PAVGB" %%mm1, %%mm0 \n\t" | ||
| 1497 | PAVGB" %%mm3, %%mm2 \n\t" | ||
| 1498 | "punpcklbw %%mm7, %%mm0 \n\t" | ||
| 1499 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 1500 | "movq "BGR2V_IDX"(%5), %%mm1 \n\t" | ||
| 1501 | "movq "BGR2V_IDX"(%5), %%mm3 \n\t" | ||
| 1502 | |||
| 1503 | "pmaddwd %%mm0, %%mm1 \n\t" | ||
| 1504 | "pmaddwd %%mm2, %%mm3 \n\t" | ||
| 1505 | "pmaddwd %%mm6, %%mm0 \n\t" | ||
| 1506 | "pmaddwd %%mm6, %%mm2 \n\t" | ||
| 1507 | "psrad $8, %%mm0 \n\t" | ||
| 1508 | "psrad $8, %%mm1 \n\t" | ||
| 1509 | "psrad $8, %%mm2 \n\t" | ||
| 1510 | "psrad $8, %%mm3 \n\t" | ||
| 1511 | "packssdw %%mm2, %%mm0 \n\t" | ||
| 1512 | "packssdw %%mm3, %%mm1 \n\t" | ||
| 1513 | "pmaddwd %%mm5, %%mm0 \n\t" | ||
| 1514 | "pmaddwd %%mm5, %%mm1 \n\t" | ||
| 1515 | "packssdw %%mm1, %%mm0 \n\t" // V1 V0 U1 U0 | ||
| 1516 | "psraw $7, %%mm0 \n\t" | ||
| 1517 | |||
| 1518 | "movq 12(%0, %%"FF_REG_d"), %%mm4 \n\t" | ||
| 1519 | "movq 12(%1, %%"FF_REG_d"), %%mm1 \n\t" | ||
| 1520 | "movq 18(%0, %%"FF_REG_d"), %%mm2 \n\t" | ||
| 1521 | "movq 18(%1, %%"FF_REG_d"), %%mm3 \n\t" | ||
| 1522 | PAVGB" %%mm1, %%mm4 \n\t" | ||
| 1523 | PAVGB" %%mm3, %%mm2 \n\t" | ||
| 1524 | "movq %%mm4, %%mm1 \n\t" | ||
| 1525 | "movq %%mm2, %%mm3 \n\t" | ||
| 1526 | "psrlq $24, %%mm4 \n\t" | ||
| 1527 | "psrlq $24, %%mm2 \n\t" | ||
| 1528 | PAVGB" %%mm1, %%mm4 \n\t" | ||
| 1529 | PAVGB" %%mm3, %%mm2 \n\t" | ||
| 1530 | "punpcklbw %%mm7, %%mm4 \n\t" | ||
| 1531 | "punpcklbw %%mm7, %%mm2 \n\t" | ||
| 1532 | "movq "BGR2V_IDX"(%5), %%mm1 \n\t" | ||
| 1533 | "movq "BGR2V_IDX"(%5), %%mm3 \n\t" | ||
| 1534 | |||
| 1535 | "pmaddwd %%mm4, %%mm1 \n\t" | ||
| 1536 | "pmaddwd %%mm2, %%mm3 \n\t" | ||
| 1537 | "pmaddwd %%mm6, %%mm4 \n\t" | ||
| 1538 | "pmaddwd %%mm6, %%mm2 \n\t" | ||
| 1539 | "psrad $8, %%mm4 \n\t" | ||
| 1540 | "psrad $8, %%mm1 \n\t" | ||
| 1541 | "psrad $8, %%mm2 \n\t" | ||
| 1542 | "psrad $8, %%mm3 \n\t" | ||
| 1543 | "packssdw %%mm2, %%mm4 \n\t" | ||
| 1544 | "packssdw %%mm3, %%mm1 \n\t" | ||
| 1545 | "pmaddwd %%mm5, %%mm4 \n\t" | ||
| 1546 | "pmaddwd %%mm5, %%mm1 \n\t" | ||
| 1547 | "add $24, %%"FF_REG_d"\n\t" | ||
| 1548 | "packssdw %%mm1, %%mm4 \n\t" // V3 V2 U3 U2 | ||
| 1549 | "psraw $7, %%mm4 \n\t" | ||
| 1550 | |||
| 1551 | "movq %%mm0, %%mm1 \n\t" | ||
| 1552 | "punpckldq %%mm4, %%mm0 \n\t" | ||
| 1553 | "punpckhdq %%mm4, %%mm1 \n\t" | ||
| 1554 | "packsswb %%mm1, %%mm0 \n\t" | ||
| 1555 | "paddb "MANGLE(bgr2UVOffset)", %%mm0 \n\t" | ||
| 1556 | "movd %%mm0, (%2, %%"FF_REG_a") \n\t" | ||
| 1557 | "punpckhdq %%mm0, %%mm0 \n\t" | ||
| 1558 | "movd %%mm0, (%3, %%"FF_REG_a") \n\t" | ||
| 1559 | "add $4, %%"FF_REG_a" \n\t" | ||
| 1560 | " js 1b \n\t" | ||
| 1561 | : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth), "r"(rgb2yuv) | ||
| 1562 | NAMED_CONSTRAINTS_ADD(w1111,bgr2UVOffset) | ||
| 1563 | : "%"FF_REG_a, "%"FF_REG_d | ||
| 1564 | ); | ||
| 1565 | |||
| 1566 | udst += chromStride; | ||
| 1567 | vdst += chromStride; | ||
| 1568 | src += srcStride*2; | ||
| 1569 | } | ||
| 1570 | |||
| 1571 | __asm__ volatile(EMMS" \n\t" | ||
| 1572 | SFENCE" \n\t" | ||
| 1573 | :::"memory"); | ||
| 1574 | |||
| 1575 | ff_rgb24toyv12_c(src, ydst, udst, vdst, width, height-y, lumStride, chromStride, srcStride, rgb2yuv); | ||
| 1576 | } | ||
| 1577 | #endif /* HAVE_7REGS */ | ||
| 1578 | |||
| 1579 | ✗ | static void extract_even_mmxext(const uint8_t *src, uint8_t *dst, x86_reg count) | |
| 1580 | { | ||
| 1581 | ✗ | dst += count; | |
| 1582 | ✗ | src += 2*count; | |
| 1583 | ✗ | count= - count; | |
| 1584 | |||
| 1585 | ✗ | if(count <= -16) { | |
| 1586 | ✗ | count += 15; | |
| 1587 | ✗ | __asm__ volatile( | |
| 1588 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1589 | "psrlw $8, %%mm7 \n\t" | ||
| 1590 | "1: \n\t" | ||
| 1591 | "movq -30(%1, %0, 2), %%mm0 \n\t" | ||
| 1592 | "movq -22(%1, %0, 2), %%mm1 \n\t" | ||
| 1593 | "movq -14(%1, %0, 2), %%mm2 \n\t" | ||
| 1594 | "movq -6(%1, %0, 2), %%mm3 \n\t" | ||
| 1595 | "pand %%mm7, %%mm0 \n\t" | ||
| 1596 | "pand %%mm7, %%mm1 \n\t" | ||
| 1597 | "pand %%mm7, %%mm2 \n\t" | ||
| 1598 | "pand %%mm7, %%mm3 \n\t" | ||
| 1599 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1600 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1601 | MOVNTQ" %%mm0,-15(%2, %0) \n\t" | ||
| 1602 | MOVNTQ" %%mm2,- 7(%2, %0) \n\t" | ||
| 1603 | "add $16, %0 \n\t" | ||
| 1604 | " js 1b \n\t" | ||
| 1605 | : "+r"(count) | ||
| 1606 | : "r"(src), "r"(dst) | ||
| 1607 | ); | ||
| 1608 | ✗ | count -= 15; | |
| 1609 | } | ||
| 1610 | ✗ | while(count<0) { | |
| 1611 | ✗ | dst[count]= src[2*count]; | |
| 1612 | ✗ | count++; | |
| 1613 | } | ||
| 1614 | ✗ | } | |
| 1615 | |||
| 1616 | ✗ | static void extract_odd_mmxext(const uint8_t *src, uint8_t *dst, x86_reg count) | |
| 1617 | { | ||
| 1618 | ✗ | src ++; | |
| 1619 | ✗ | dst += count; | |
| 1620 | ✗ | src += 2*count; | |
| 1621 | ✗ | count= - count; | |
| 1622 | |||
| 1623 | ✗ | if(count < -16) { | |
| 1624 | ✗ | count += 16; | |
| 1625 | ✗ | __asm__ volatile( | |
| 1626 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1627 | "psrlw $8, %%mm7 \n\t" | ||
| 1628 | "1: \n\t" | ||
| 1629 | "movq -32(%1, %0, 2), %%mm0 \n\t" | ||
| 1630 | "movq -24(%1, %0, 2), %%mm1 \n\t" | ||
| 1631 | "movq -16(%1, %0, 2), %%mm2 \n\t" | ||
| 1632 | "movq -8(%1, %0, 2), %%mm3 \n\t" | ||
| 1633 | "pand %%mm7, %%mm0 \n\t" | ||
| 1634 | "pand %%mm7, %%mm1 \n\t" | ||
| 1635 | "pand %%mm7, %%mm2 \n\t" | ||
| 1636 | "pand %%mm7, %%mm3 \n\t" | ||
| 1637 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1638 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1639 | MOVNTQ" %%mm0,-16(%2, %0) \n\t" | ||
| 1640 | MOVNTQ" %%mm2,- 8(%2, %0) \n\t" | ||
| 1641 | "add $16, %0 \n\t" | ||
| 1642 | " js 1b \n\t" | ||
| 1643 | : "+r"(count) | ||
| 1644 | : "r"(src), "r"(dst) | ||
| 1645 | ); | ||
| 1646 | ✗ | count -= 16; | |
| 1647 | } | ||
| 1648 | ✗ | while(count<0) { | |
| 1649 | ✗ | dst[count]= src[2*count]; | |
| 1650 | ✗ | count++; | |
| 1651 | } | ||
| 1652 | ✗ | } | |
| 1653 | |||
| 1654 | #if ARCH_X86_32 | ||
| 1655 | static void extract_even2_mmxext(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count) | ||
| 1656 | { | ||
| 1657 | dst0+= count; | ||
| 1658 | dst1+= count; | ||
| 1659 | src += 4*count; | ||
| 1660 | count= - count; | ||
| 1661 | if(count <= -8) { | ||
| 1662 | count += 7; | ||
| 1663 | __asm__ volatile( | ||
| 1664 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1665 | "psrlw $8, %%mm7 \n\t" | ||
| 1666 | "1: \n\t" | ||
| 1667 | "movq -28(%1, %0, 4), %%mm0 \n\t" | ||
| 1668 | "movq -20(%1, %0, 4), %%mm1 \n\t" | ||
| 1669 | "movq -12(%1, %0, 4), %%mm2 \n\t" | ||
| 1670 | "movq -4(%1, %0, 4), %%mm3 \n\t" | ||
| 1671 | "pand %%mm7, %%mm0 \n\t" | ||
| 1672 | "pand %%mm7, %%mm1 \n\t" | ||
| 1673 | "pand %%mm7, %%mm2 \n\t" | ||
| 1674 | "pand %%mm7, %%mm3 \n\t" | ||
| 1675 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1676 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1677 | "movq %%mm0, %%mm1 \n\t" | ||
| 1678 | "movq %%mm2, %%mm3 \n\t" | ||
| 1679 | "psrlw $8, %%mm0 \n\t" | ||
| 1680 | "psrlw $8, %%mm2 \n\t" | ||
| 1681 | "pand %%mm7, %%mm1 \n\t" | ||
| 1682 | "pand %%mm7, %%mm3 \n\t" | ||
| 1683 | "packuswb %%mm2, %%mm0 \n\t" | ||
| 1684 | "packuswb %%mm3, %%mm1 \n\t" | ||
| 1685 | MOVNTQ" %%mm0,- 7(%3, %0) \n\t" | ||
| 1686 | MOVNTQ" %%mm1,- 7(%2, %0) \n\t" | ||
| 1687 | "add $8, %0 \n\t" | ||
| 1688 | " js 1b \n\t" | ||
| 1689 | : "+r"(count) | ||
| 1690 | : "r"(src), "r"(dst0), "r"(dst1) | ||
| 1691 | ); | ||
| 1692 | count -= 7; | ||
| 1693 | } | ||
| 1694 | while(count<0) { | ||
| 1695 | dst0[count]= src[4*count+0]; | ||
| 1696 | dst1[count]= src[4*count+2]; | ||
| 1697 | count++; | ||
| 1698 | } | ||
| 1699 | } | ||
| 1700 | #endif /* ARCH_X86_32 */ | ||
| 1701 | |||
| 1702 | ✗ | static void extract_even2avg_mmxext(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count) | |
| 1703 | { | ||
| 1704 | ✗ | dst0 += count; | |
| 1705 | ✗ | dst1 += count; | |
| 1706 | ✗ | src0 += 4*count; | |
| 1707 | ✗ | src1 += 4*count; | |
| 1708 | ✗ | count= - count; | |
| 1709 | #ifdef PAVGB | ||
| 1710 | ✗ | if(count <= -8) { | |
| 1711 | ✗ | count += 7; | |
| 1712 | ✗ | __asm__ volatile( | |
| 1713 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1714 | "psrlw $8, %%mm7 \n\t" | ||
| 1715 | "1: \n\t" | ||
| 1716 | "movq -28(%1, %0, 4), %%mm0 \n\t" | ||
| 1717 | "movq -20(%1, %0, 4), %%mm1 \n\t" | ||
| 1718 | "movq -12(%1, %0, 4), %%mm2 \n\t" | ||
| 1719 | "movq -4(%1, %0, 4), %%mm3 \n\t" | ||
| 1720 | PAVGB" -28(%2, %0, 4), %%mm0 \n\t" | ||
| 1721 | PAVGB" -20(%2, %0, 4), %%mm1 \n\t" | ||
| 1722 | PAVGB" -12(%2, %0, 4), %%mm2 \n\t" | ||
| 1723 | PAVGB" - 4(%2, %0, 4), %%mm3 \n\t" | ||
| 1724 | "pand %%mm7, %%mm0 \n\t" | ||
| 1725 | "pand %%mm7, %%mm1 \n\t" | ||
| 1726 | "pand %%mm7, %%mm2 \n\t" | ||
| 1727 | "pand %%mm7, %%mm3 \n\t" | ||
| 1728 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1729 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1730 | "movq %%mm0, %%mm1 \n\t" | ||
| 1731 | "movq %%mm2, %%mm3 \n\t" | ||
| 1732 | "psrlw $8, %%mm0 \n\t" | ||
| 1733 | "psrlw $8, %%mm2 \n\t" | ||
| 1734 | "pand %%mm7, %%mm1 \n\t" | ||
| 1735 | "pand %%mm7, %%mm3 \n\t" | ||
| 1736 | "packuswb %%mm2, %%mm0 \n\t" | ||
| 1737 | "packuswb %%mm3, %%mm1 \n\t" | ||
| 1738 | MOVNTQ" %%mm0,- 7(%4, %0) \n\t" | ||
| 1739 | MOVNTQ" %%mm1,- 7(%3, %0) \n\t" | ||
| 1740 | "add $8, %0 \n\t" | ||
| 1741 | " js 1b \n\t" | ||
| 1742 | : "+r"(count) | ||
| 1743 | : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1) | ||
| 1744 | ); | ||
| 1745 | ✗ | count -= 7; | |
| 1746 | } | ||
| 1747 | #endif | ||
| 1748 | ✗ | while(count<0) { | |
| 1749 | ✗ | dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1; | |
| 1750 | ✗ | dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1; | |
| 1751 | ✗ | count++; | |
| 1752 | } | ||
| 1753 | ✗ | } | |
| 1754 | |||
| 1755 | ✗ | static void extract_odd2_mmxext(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count) | |
| 1756 | { | ||
| 1757 | ✗ | dst0+= count; | |
| 1758 | ✗ | dst1+= count; | |
| 1759 | ✗ | src += 4*count; | |
| 1760 | ✗ | count= - count; | |
| 1761 | ✗ | if(count <= -8) { | |
| 1762 | ✗ | count += 7; | |
| 1763 | ✗ | __asm__ volatile( | |
| 1764 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1765 | "psrlw $8, %%mm7 \n\t" | ||
| 1766 | "1: \n\t" | ||
| 1767 | "movq -28(%1, %0, 4), %%mm0 \n\t" | ||
| 1768 | "movq -20(%1, %0, 4), %%mm1 \n\t" | ||
| 1769 | "movq -12(%1, %0, 4), %%mm2 \n\t" | ||
| 1770 | "movq -4(%1, %0, 4), %%mm3 \n\t" | ||
| 1771 | "psrlw $8, %%mm0 \n\t" | ||
| 1772 | "psrlw $8, %%mm1 \n\t" | ||
| 1773 | "psrlw $8, %%mm2 \n\t" | ||
| 1774 | "psrlw $8, %%mm3 \n\t" | ||
| 1775 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1776 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1777 | "movq %%mm0, %%mm1 \n\t" | ||
| 1778 | "movq %%mm2, %%mm3 \n\t" | ||
| 1779 | "psrlw $8, %%mm0 \n\t" | ||
| 1780 | "psrlw $8, %%mm2 \n\t" | ||
| 1781 | "pand %%mm7, %%mm1 \n\t" | ||
| 1782 | "pand %%mm7, %%mm3 \n\t" | ||
| 1783 | "packuswb %%mm2, %%mm0 \n\t" | ||
| 1784 | "packuswb %%mm3, %%mm1 \n\t" | ||
| 1785 | MOVNTQ" %%mm0,- 7(%3, %0) \n\t" | ||
| 1786 | MOVNTQ" %%mm1,- 7(%2, %0) \n\t" | ||
| 1787 | "add $8, %0 \n\t" | ||
| 1788 | " js 1b \n\t" | ||
| 1789 | : "+r"(count) | ||
| 1790 | : "r"(src), "r"(dst0), "r"(dst1) | ||
| 1791 | ); | ||
| 1792 | ✗ | count -= 7; | |
| 1793 | } | ||
| 1794 | ✗ | src++; | |
| 1795 | ✗ | while(count<0) { | |
| 1796 | ✗ | dst0[count]= src[4*count+0]; | |
| 1797 | ✗ | dst1[count]= src[4*count+2]; | |
| 1798 | ✗ | count++; | |
| 1799 | } | ||
| 1800 | ✗ | } | |
| 1801 | |||
| 1802 | ✗ | static void extract_odd2avg_mmxext(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count) | |
| 1803 | { | ||
| 1804 | ✗ | dst0 += count; | |
| 1805 | ✗ | dst1 += count; | |
| 1806 | ✗ | src0 += 4*count; | |
| 1807 | ✗ | src1 += 4*count; | |
| 1808 | ✗ | count= - count; | |
| 1809 | #ifdef PAVGB | ||
| 1810 | ✗ | if(count <= -8) { | |
| 1811 | ✗ | count += 7; | |
| 1812 | ✗ | __asm__ volatile( | |
| 1813 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 1814 | "psrlw $8, %%mm7 \n\t" | ||
| 1815 | "1: \n\t" | ||
| 1816 | "movq -28(%1, %0, 4), %%mm0 \n\t" | ||
| 1817 | "movq -20(%1, %0, 4), %%mm1 \n\t" | ||
| 1818 | "movq -12(%1, %0, 4), %%mm2 \n\t" | ||
| 1819 | "movq -4(%1, %0, 4), %%mm3 \n\t" | ||
| 1820 | PAVGB" -28(%2, %0, 4), %%mm0 \n\t" | ||
| 1821 | PAVGB" -20(%2, %0, 4), %%mm1 \n\t" | ||
| 1822 | PAVGB" -12(%2, %0, 4), %%mm2 \n\t" | ||
| 1823 | PAVGB" - 4(%2, %0, 4), %%mm3 \n\t" | ||
| 1824 | "psrlw $8, %%mm0 \n\t" | ||
| 1825 | "psrlw $8, %%mm1 \n\t" | ||
| 1826 | "psrlw $8, %%mm2 \n\t" | ||
| 1827 | "psrlw $8, %%mm3 \n\t" | ||
| 1828 | "packuswb %%mm1, %%mm0 \n\t" | ||
| 1829 | "packuswb %%mm3, %%mm2 \n\t" | ||
| 1830 | "movq %%mm0, %%mm1 \n\t" | ||
| 1831 | "movq %%mm2, %%mm3 \n\t" | ||
| 1832 | "psrlw $8, %%mm0 \n\t" | ||
| 1833 | "psrlw $8, %%mm2 \n\t" | ||
| 1834 | "pand %%mm7, %%mm1 \n\t" | ||
| 1835 | "pand %%mm7, %%mm3 \n\t" | ||
| 1836 | "packuswb %%mm2, %%mm0 \n\t" | ||
| 1837 | "packuswb %%mm3, %%mm1 \n\t" | ||
| 1838 | MOVNTQ" %%mm0,- 7(%4, %0) \n\t" | ||
| 1839 | MOVNTQ" %%mm1,- 7(%3, %0) \n\t" | ||
| 1840 | "add $8, %0 \n\t" | ||
| 1841 | " js 1b \n\t" | ||
| 1842 | : "+r"(count) | ||
| 1843 | : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1) | ||
| 1844 | ); | ||
| 1845 | ✗ | count -= 7; | |
| 1846 | } | ||
| 1847 | #endif | ||
| 1848 | ✗ | src0++; | |
| 1849 | ✗ | src1++; | |
| 1850 | ✗ | while(count<0) { | |
| 1851 | ✗ | dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1; | |
| 1852 | ✗ | dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1; | |
| 1853 | ✗ | count++; | |
| 1854 | } | ||
| 1855 | ✗ | } | |
| 1856 | |||
| 1857 | ✗ | static void yuyvtoyuv420_mmxext(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, | |
| 1858 | int width, int height, | ||
| 1859 | int lumStride, int chromStride, int srcStride) | ||
| 1860 | { | ||
| 1861 | ✗ | const int chromWidth = AV_CEIL_RSHIFT(width, 1); | |
| 1862 | |||
| 1863 | ✗ | for (int y = 0; y < height; y++) { | |
| 1864 | ✗ | extract_even_mmxext(src, ydst, width); | |
| 1865 | ✗ | if(y&1) { | |
| 1866 | ✗ | extract_odd2avg_mmxext(src-srcStride, src, udst, vdst, chromWidth); | |
| 1867 | ✗ | udst+= chromStride; | |
| 1868 | ✗ | vdst+= chromStride; | |
| 1869 | } | ||
| 1870 | |||
| 1871 | ✗ | src += srcStride; | |
| 1872 | ✗ | ydst+= lumStride; | |
| 1873 | } | ||
| 1874 | ✗ | __asm__( | |
| 1875 | EMMS" \n\t" | ||
| 1876 | SFENCE" \n\t" | ||
| 1877 | ::: "memory" | ||
| 1878 | ); | ||
| 1879 | ✗ | } | |
| 1880 | |||
| 1881 | ✗ | static void yuyvtoyuv422_mmxext(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, | |
| 1882 | int width, int height, | ||
| 1883 | int lumStride, int chromStride, int srcStride) | ||
| 1884 | { | ||
| 1885 | ✗ | const int chromWidth = AV_CEIL_RSHIFT(width, 1); | |
| 1886 | |||
| 1887 | ✗ | for (int y = 0; y < height; y++) { | |
| 1888 | ✗ | extract_even_mmxext(src, ydst, width); | |
| 1889 | ✗ | extract_odd2_mmxext(src, udst, vdst, chromWidth); | |
| 1890 | |||
| 1891 | ✗ | src += srcStride; | |
| 1892 | ✗ | ydst+= lumStride; | |
| 1893 | ✗ | udst+= chromStride; | |
| 1894 | ✗ | vdst+= chromStride; | |
| 1895 | } | ||
| 1896 | ✗ | __asm__( | |
| 1897 | EMMS" \n\t" | ||
| 1898 | SFENCE" \n\t" | ||
| 1899 | ::: "memory" | ||
| 1900 | ); | ||
| 1901 | ✗ | } | |
| 1902 | |||
| 1903 | ✗ | static void uyvytoyuv420_mmxext(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, | |
| 1904 | int width, int height, | ||
| 1905 | int lumStride, int chromStride, int srcStride) | ||
| 1906 | { | ||
| 1907 | ✗ | const int chromWidth = AV_CEIL_RSHIFT(width, 1); | |
| 1908 | |||
| 1909 | ✗ | for (int y = 0; y < height; y++) { | |
| 1910 | ✗ | extract_odd_mmxext(src, ydst, width); | |
| 1911 | ✗ | if(y&1) { | |
| 1912 | ✗ | extract_even2avg_mmxext(src-srcStride, src, udst, vdst, chromWidth); | |
| 1913 | ✗ | udst+= chromStride; | |
| 1914 | ✗ | vdst+= chromStride; | |
| 1915 | } | ||
| 1916 | |||
| 1917 | ✗ | src += srcStride; | |
| 1918 | ✗ | ydst+= lumStride; | |
| 1919 | } | ||
| 1920 | ✗ | __asm__( | |
| 1921 | EMMS" \n\t" | ||
| 1922 | SFENCE" \n\t" | ||
| 1923 | ::: "memory" | ||
| 1924 | ); | ||
| 1925 | ✗ | } | |
| 1926 | |||
| 1927 | #if ARCH_X86_32 | ||
| 1928 | static void uyvytoyuv422_mmxext(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src, | ||
| 1929 | int width, int height, | ||
| 1930 | int lumStride, int chromStride, int srcStride) | ||
| 1931 | { | ||
| 1932 | const int chromWidth = AV_CEIL_RSHIFT(width, 1); | ||
| 1933 | |||
| 1934 | for (int y = 0; y < height; y++) { | ||
| 1935 | extract_odd_mmxext(src, ydst, width); | ||
| 1936 | extract_even2_mmxext(src, udst, vdst, chromWidth); | ||
| 1937 | |||
| 1938 | src += srcStride; | ||
| 1939 | ydst+= lumStride; | ||
| 1940 | udst+= chromStride; | ||
| 1941 | vdst+= chromStride; | ||
| 1942 | } | ||
| 1943 | __asm__( | ||
| 1944 | EMMS" \n\t" | ||
| 1945 | SFENCE" \n\t" | ||
| 1946 | ::: "memory" | ||
| 1947 | ); | ||
| 1948 | } | ||
| 1949 | #endif /* ARCH_X86_32 */ | ||
| 1950 | |||
| 1951 | 18 | static av_cold void rgb2rgb_init_mmxext(void) | |
| 1952 | { | ||
| 1953 | 18 | rgb15to16 = rgb15to16_mmxext; | |
| 1954 | 18 | rgb15tobgr24 = rgb15tobgr24_mmxext; | |
| 1955 | 18 | rgb15to32 = rgb15to32_mmxext; | |
| 1956 | 18 | rgb16tobgr24 = rgb16tobgr24_mmxext; | |
| 1957 | 18 | rgb16to32 = rgb16to32_mmxext; | |
| 1958 | 18 | rgb16to15 = rgb16to15_mmxext; | |
| 1959 | 18 | rgb24tobgr16 = rgb24tobgr16_mmxext; | |
| 1960 | 18 | rgb24tobgr15 = rgb24tobgr15_mmxext; | |
| 1961 | 18 | rgb24tobgr32 = rgb24tobgr32_mmxext; | |
| 1962 | 18 | rgb32to16 = rgb32to16_mmxext; | |
| 1963 | 18 | rgb32to15 = rgb32to15_mmxext; | |
| 1964 | 18 | rgb32tobgr24 = rgb32tobgr24_mmxext; | |
| 1965 | 18 | rgb24to15 = rgb24to15_mmxext; | |
| 1966 | 18 | rgb24to16 = rgb24to16_mmxext; | |
| 1967 | 18 | rgb24tobgr24 = rgb24tobgr24_mmxext; | |
| 1968 | 18 | rgb32tobgr16 = rgb32tobgr16_mmxext; | |
| 1969 | 18 | rgb32tobgr15 = rgb32tobgr15_mmxext; | |
| 1970 | 18 | yv12toyuy2 = yv12toyuy2_mmxext; | |
| 1971 | 18 | yv12touyvy = yv12touyvy_mmxext; | |
| 1972 | 18 | yuv422ptoyuy2 = yuv422ptoyuy2_mmxext; | |
| 1973 | 18 | yuv422ptouyvy = yuv422ptouyvy_mmxext; | |
| 1974 | #if ARCH_X86_32 | ||
| 1975 | uyvytoyuv422 = uyvytoyuv422_mmxext; | ||
| 1976 | #endif | ||
| 1977 | 18 | yuyvtoyuv422 = yuyvtoyuv422_mmxext; | |
| 1978 | |||
| 1979 | 18 | planar2x = planar2x_mmxext; | |
| 1980 | #if ARCH_X86_32 && HAVE_7REGS | ||
| 1981 | ff_rgb24toyv12 = rgb24toyv12_mmxext; | ||
| 1982 | #endif /* ARCH_X86_32 && HAVE_7REGS */ | ||
| 1983 | |||
| 1984 | 18 | yuyvtoyuv420 = yuyvtoyuv420_mmxext; | |
| 1985 | 18 | uyvytoyuv420 = uyvytoyuv420_mmxext; | |
| 1986 | 18 | } | |
| 1987 | |||
| 1988 | //SSE2 versions | ||
| 1989 | 23 | static void interleave_bytes_sse2(const uint8_t *src1, const uint8_t *src2, uint8_t *dest, | |
| 1990 | int width, int height, int src1Stride, | ||
| 1991 | int src2Stride, int dstStride) | ||
| 1992 | { | ||
| 1993 |
2/2✓ Branch 0 taken 1664 times.
✓ Branch 1 taken 23 times.
|
1687 | for (int h = 0; h < height; h++) { |
| 1994 |
2/2✓ Branch 0 taken 650 times.
✓ Branch 1 taken 1014 times.
|
1664 | if (width >= 16) { |
| 1995 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 74 times.
|
650 | if (!((((intptr_t)src1) | ((intptr_t)src2) | ((intptr_t)dest))&15)) { |
| 1996 | 576 | __asm__( | |
| 1997 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 1998 | "1: \n\t" | ||
| 1999 | PREFETCH" 64(%1, %%"FF_REG_a") \n\t" | ||
| 2000 | PREFETCH" 64(%2, %%"FF_REG_a") \n\t" | ||
| 2001 | "movdqa (%1, %%"FF_REG_a"), %%xmm0 \n\t" | ||
| 2002 | "movdqa (%1, %%"FF_REG_a"), %%xmm1 \n\t" | ||
| 2003 | "movdqa (%2, %%"FF_REG_a"), %%xmm2 \n\t" | ||
| 2004 | "punpcklbw %%xmm2, %%xmm0 \n\t" | ||
| 2005 | "punpckhbw %%xmm2, %%xmm1 \n\t" | ||
| 2006 | "movntdq %%xmm0, (%0, %%"FF_REG_a", 2) \n\t" | ||
| 2007 | "movntdq %%xmm1, 16(%0, %%"FF_REG_a", 2) \n\t" | ||
| 2008 | "add $16, %%"FF_REG_a" \n\t" | ||
| 2009 | "cmp %3, %%"FF_REG_a" \n\t" | ||
| 2010 | " jb 1b \n\t" | ||
| 2011 | 576 | ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15) | |
| 2012 | : "memory", XMM_CLOBBERS("xmm0", "xmm1", "xmm2",) "%"FF_REG_a | ||
| 2013 | ); | ||
| 2014 | } else | ||
| 2015 | 74 | __asm__( | |
| 2016 | "xor %%"FF_REG_a", %%"FF_REG_a" \n\t" | ||
| 2017 | "1: \n\t" | ||
| 2018 | PREFETCH" 64(%1, %%"FF_REG_a") \n\t" | ||
| 2019 | PREFETCH" 64(%2, %%"FF_REG_a") \n\t" | ||
| 2020 | "movq (%1, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 2021 | "movq 8(%1, %%"FF_REG_a"), %%mm2 \n\t" | ||
| 2022 | "movq %%mm0, %%mm1 \n\t" | ||
| 2023 | "movq %%mm2, %%mm3 \n\t" | ||
| 2024 | "movq (%2, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 2025 | "movq 8(%2, %%"FF_REG_a"), %%mm5 \n\t" | ||
| 2026 | "punpcklbw %%mm4, %%mm0 \n\t" | ||
| 2027 | "punpckhbw %%mm4, %%mm1 \n\t" | ||
| 2028 | "punpcklbw %%mm5, %%mm2 \n\t" | ||
| 2029 | "punpckhbw %%mm5, %%mm3 \n\t" | ||
| 2030 | MOVNTQ" %%mm0, (%0, %%"FF_REG_a", 2) \n\t" | ||
| 2031 | MOVNTQ" %%mm1, 8(%0, %%"FF_REG_a", 2) \n\t" | ||
| 2032 | MOVNTQ" %%mm2, 16(%0, %%"FF_REG_a", 2) \n\t" | ||
| 2033 | MOVNTQ" %%mm3, 24(%0, %%"FF_REG_a", 2) \n\t" | ||
| 2034 | "add $16, %%"FF_REG_a" \n\t" | ||
| 2035 | "cmp %3, %%"FF_REG_a" \n\t" | ||
| 2036 | " jb 1b \n\t" | ||
| 2037 | 74 | ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15) | |
| 2038 | : "memory", "%"FF_REG_a | ||
| 2039 | ); | ||
| 2040 | |||
| 2041 | } | ||
| 2042 |
2/2✓ Branch 0 taken 8423 times.
✓ Branch 1 taken 1664 times.
|
10087 | for (int w = (width & (~15)); w < width; w++) { |
| 2043 | 8423 | dest[2*w+0] = src1[w]; | |
| 2044 | 8423 | dest[2*w+1] = src2[w]; | |
| 2045 | } | ||
| 2046 | 1664 | dest += dstStride; | |
| 2047 | 1664 | src1 += src1Stride; | |
| 2048 | 1664 | src2 += src2Stride; | |
| 2049 | } | ||
| 2050 | 23 | __asm__( | |
| 2051 | EMMS" \n\t" | ||
| 2052 | SFENCE" \n\t" | ||
| 2053 | ::: "memory" | ||
| 2054 | ); | ||
| 2055 | 23 | } | |
| 2056 | |||
| 2057 | /* | ||
| 2058 | RGB15->RGB16 original by Strepto/Astral | ||
| 2059 | ported to gcc & bugfixed : A'rpi | ||
| 2060 | MMXEXT, 3DNOW optimization by Nick Kurshev | ||
| 2061 | 32-bit C version, and and&add trick by Michael Niedermayer | ||
| 2062 | */ | ||
| 2063 | |||
| 2064 | #endif /* HAVE_INLINE_ASM */ | ||
| 2065 | |||
| 2066 | void ff_shuffle_bytes_2103_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2067 | void ff_shuffle_bytes_0321_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2068 | void ff_shuffle_bytes_1230_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2069 | void ff_shuffle_bytes_3012_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2070 | void ff_shuffle_bytes_3210_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2071 | void ff_shuffle_bytes_3102_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2072 | void ff_shuffle_bytes_2013_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2073 | void ff_shuffle_bytes_2130_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2074 | void ff_shuffle_bytes_1203_ssse3(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2075 | |||
| 2076 | #if ARCH_X86_64 | ||
| 2077 | void ff_shuffle_bytes_2103_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2078 | void ff_shuffle_bytes_0321_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2079 | void ff_shuffle_bytes_1230_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2080 | void ff_shuffle_bytes_3012_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2081 | void ff_shuffle_bytes_3210_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2082 | void ff_shuffle_bytes_3102_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2083 | void ff_shuffle_bytes_2013_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2084 | void ff_shuffle_bytes_2130_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2085 | void ff_shuffle_bytes_1203_avx2(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2086 | |||
| 2087 | void ff_shuffle_bytes_2103_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2088 | void ff_shuffle_bytes_0321_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2089 | void ff_shuffle_bytes_1230_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2090 | void ff_shuffle_bytes_3012_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2091 | void ff_shuffle_bytes_3210_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2092 | void ff_shuffle_bytes_3102_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2093 | void ff_shuffle_bytes_2013_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2094 | void ff_shuffle_bytes_2130_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2095 | void ff_shuffle_bytes_1203_avx512icl(const uint8_t *src, uint8_t *dst, int src_size); | ||
| 2096 | |||
| 2097 | void ff_uyvytoyuv422_sse2(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, | ||
| 2098 | const uint8_t *src, int width, int height, | ||
| 2099 | int lumStride, int chromStride, int srcStride); | ||
| 2100 | void ff_uyvytoyuv422_avx(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, | ||
| 2101 | const uint8_t *src, int width, int height, | ||
| 2102 | int lumStride, int chromStride, int srcStride); | ||
| 2103 | void ff_uyvytoyuv422_avx2(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, | ||
| 2104 | const uint8_t *src, int width, int height, | ||
| 2105 | int lumStride, int chromStride, int srcStride); | ||
| 2106 | void ff_uyvytoyuv422_avx512icl(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, | ||
| 2107 | const uint8_t *src, int width, int height, | ||
| 2108 | int lumStride, int chromStride, int srcStride); | ||
| 2109 | #endif | ||
| 2110 | |||
| 2111 | #define DEINTERLEAVE_BYTES(cpuext) \ | ||
| 2112 | void ff_nv12ToUV_ ## cpuext(uint8_t *dstU, uint8_t *dstV, \ | ||
| 2113 | const uint8_t *unused, \ | ||
| 2114 | const uint8_t *src1, \ | ||
| 2115 | const uint8_t *src2, \ | ||
| 2116 | int w, \ | ||
| 2117 | uint32_t *unused2, \ | ||
| 2118 | void *opq); \ | ||
| 2119 | static void deinterleave_bytes_ ## cpuext(const uint8_t *src, uint8_t *dst1, uint8_t *dst2, \ | ||
| 2120 | int width, int height, int srcStride, \ | ||
| 2121 | int dst1Stride, int dst2Stride) \ | ||
| 2122 | { \ | ||
| 2123 | for (int h = 0; h < height; h++) { \ | ||
| 2124 | if (width >= 16) \ | ||
| 2125 | ff_nv12ToUV_ ## cpuext(dst1, dst2, NULL, src, NULL, width - 15, NULL, NULL); \ | ||
| 2126 | for (int w = (width & (~15)); w < width; w++) { \ | ||
| 2127 | dst1[w] = src[2*w+0]; \ | ||
| 2128 | dst2[w] = src[2*w+1]; \ | ||
| 2129 | } \ | ||
| 2130 | src += srcStride; \ | ||
| 2131 | dst1 += dst1Stride; \ | ||
| 2132 | dst2 += dst2Stride; \ | ||
| 2133 | } \ | ||
| 2134 | } | ||
| 2135 | |||
| 2136 | #if HAVE_SSE2_EXTERNAL | ||
| 2137 |
6/6✓ Branch 0 taken 148 times.
✓ Branch 1 taken 1859 times.
✓ Branch 3 taken 13473 times.
✓ Branch 4 taken 2007 times.
✓ Branch 5 taken 2007 times.
✓ Branch 6 taken 34 times.
|
15514 | DEINTERLEAVE_BYTES(sse2) |
| 2138 | #endif | ||
| 2139 | #if HAVE_AVX_EXTERNAL | ||
| 2140 |
6/6✓ Branch 0 taken 325 times.
✓ Branch 1 taken 973 times.
✓ Branch 3 taken 8096 times.
✓ Branch 4 taken 1298 times.
✓ Branch 5 taken 1298 times.
✓ Branch 6 taken 19 times.
|
9413 | DEINTERLEAVE_BYTES(avx) |
| 2141 | #endif | ||
| 2142 | |||
| 2143 | 5019 | av_cold void rgb2rgb_init_x86(void) | |
| 2144 | { | ||
| 2145 | 5019 | int cpu_flags = av_get_cpu_flags(); | |
| 2146 | |||
| 2147 | #if HAVE_INLINE_ASM | ||
| 2148 |
2/2✓ Branch 0 taken 18 times.
✓ Branch 1 taken 5001 times.
|
5019 | if (INLINE_MMXEXT(cpu_flags)) |
| 2149 | 18 | rgb2rgb_init_mmxext(); | |
| 2150 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 5003 times.
|
5019 | if (INLINE_SSE2(cpu_flags)) |
| 2151 | 16 | interleaveBytes = interleave_bytes_sse2; | |
| 2152 | #endif /* HAVE_INLINE_ASM */ | ||
| 2153 | |||
| 2154 | #if HAVE_SSE2_EXTERNAL | ||
| 2155 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 5003 times.
|
5019 | if (EXTERNAL_SSE2(cpu_flags)) { |
| 2156 | #if ARCH_X86_64 | ||
| 2157 | 16 | uyvytoyuv422 = ff_uyvytoyuv422_sse2; | |
| 2158 | #endif | ||
| 2159 | 16 | deinterleaveBytes = deinterleave_bytes_sse2; | |
| 2160 | } | ||
| 2161 | #endif | ||
| 2162 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 5005 times.
|
5019 | if (EXTERNAL_SSSE3(cpu_flags)) { |
| 2163 | 14 | shuffle_bytes_0321 = ff_shuffle_bytes_0321_ssse3; | |
| 2164 | 14 | shuffle_bytes_2103 = ff_shuffle_bytes_2103_ssse3; | |
| 2165 | 14 | shuffle_bytes_1230 = ff_shuffle_bytes_1230_ssse3; | |
| 2166 | 14 | shuffle_bytes_3012 = ff_shuffle_bytes_3012_ssse3; | |
| 2167 | 14 | shuffle_bytes_3210 = ff_shuffle_bytes_3210_ssse3; | |
| 2168 | 14 | shuffle_bytes_3102 = ff_shuffle_bytes_3102_ssse3; | |
| 2169 | 14 | shuffle_bytes_2013 = ff_shuffle_bytes_2013_ssse3; | |
| 2170 | 14 | shuffle_bytes_2130 = ff_shuffle_bytes_2130_ssse3; | |
| 2171 | 14 | shuffle_bytes_1203 = ff_shuffle_bytes_1203_ssse3; | |
| 2172 | } | ||
| 2173 | #if HAVE_AVX_EXTERNAL | ||
| 2174 |
2/2✓ Branch 0 taken 9 times.
✓ Branch 1 taken 5010 times.
|
5019 | if (EXTERNAL_AVX(cpu_flags)) { |
| 2175 | 9 | deinterleaveBytes = deinterleave_bytes_avx; | |
| 2176 | #if ARCH_X86_64 | ||
| 2177 | 9 | uyvytoyuv422 = ff_uyvytoyuv422_avx; | |
| 2178 | } | ||
| 2179 |
3/4✓ Branch 0 taken 7 times.
✓ Branch 1 taken 5012 times.
✓ Branch 2 taken 7 times.
✗ Branch 3 not taken.
|
5019 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
| 2180 | 7 | shuffle_bytes_0321 = ff_shuffle_bytes_0321_avx2; | |
| 2181 | 7 | shuffle_bytes_2103 = ff_shuffle_bytes_2103_avx2; | |
| 2182 | 7 | shuffle_bytes_1230 = ff_shuffle_bytes_1230_avx2; | |
| 2183 | 7 | shuffle_bytes_3012 = ff_shuffle_bytes_3012_avx2; | |
| 2184 | 7 | shuffle_bytes_3210 = ff_shuffle_bytes_3210_avx2; | |
| 2185 | 7 | shuffle_bytes_3102 = ff_shuffle_bytes_3102_avx2; | |
| 2186 | 7 | shuffle_bytes_2013 = ff_shuffle_bytes_2013_avx2; | |
| 2187 | 7 | shuffle_bytes_2130 = ff_shuffle_bytes_2130_avx2; | |
| 2188 | 7 | shuffle_bytes_1203 = ff_shuffle_bytes_1203_avx2; | |
| 2189 | } | ||
| 2190 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5019 times.
|
5019 | if (EXTERNAL_AVX512ICL(cpu_flags)) { |
| 2191 | ✗ | shuffle_bytes_0321 = ff_shuffle_bytes_0321_avx512icl; | |
| 2192 | ✗ | shuffle_bytes_2103 = ff_shuffle_bytes_2103_avx512icl; | |
| 2193 | ✗ | shuffle_bytes_1230 = ff_shuffle_bytes_1230_avx512icl; | |
| 2194 | ✗ | shuffle_bytes_3012 = ff_shuffle_bytes_3012_avx512icl; | |
| 2195 | ✗ | shuffle_bytes_3210 = ff_shuffle_bytes_3210_avx512icl; | |
| 2196 | ✗ | shuffle_bytes_3102 = ff_shuffle_bytes_3102_avx512icl; | |
| 2197 | ✗ | shuffle_bytes_2013 = ff_shuffle_bytes_2013_avx512icl; | |
| 2198 | ✗ | shuffle_bytes_2130 = ff_shuffle_bytes_2130_avx512icl; | |
| 2199 | ✗ | shuffle_bytes_1203 = ff_shuffle_bytes_1203_avx512icl; | |
| 2200 | } | ||
| 2201 |
3/4✓ Branch 0 taken 7 times.
✓ Branch 1 taken 5012 times.
✓ Branch 2 taken 7 times.
✗ Branch 3 not taken.
|
5019 | if (EXTERNAL_AVX2_FAST(cpu_flags)) { |
| 2202 | 7 | uyvytoyuv422 = ff_uyvytoyuv422_avx2; | |
| 2203 | } | ||
| 2204 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5019 times.
|
5019 | if (EXTERNAL_AVX512ICL(cpu_flags)) { |
| 2205 | ✗ | uyvytoyuv422 = ff_uyvytoyuv422_avx512icl; | |
| 2206 | #endif | ||
| 2207 | } | ||
| 2208 | #endif | ||
| 2209 | 5019 | } | |
| 2210 |