rgb2rgb_template.c 98 KB
Newer Older
1
/*
2 3 4 5 6 7 8
 * software RGB to RGB converter
 * pluralize by software PAL8 to RGB converter
 *              software YUV to YUV converter
 *              software YUV to RGB converter
 * Written by Nick Kurshev.
 * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at)
 * lot of big-endian byte order fixes by Alex Beregszaszi
9
 *
10 11 12
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or modify
13 14 15 16
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
17
 * FFmpeg is distributed in the hope that it will be useful,
18 19 20 21 22
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
23
 * along with FFmpeg; if not, write to the Free Software
24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
 *
26
 * The C code (not assembly, MMX, ...) of this file can be used
Diego Biurrun's avatar
Diego Biurrun committed
27
 * under the LGPL license.
28 29
 */

Arpi's avatar
Arpi committed
30 31
#include <stddef.h>

Michael Niedermayer's avatar
Michael Niedermayer committed
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
#undef PREFETCH
#undef MOVNTQ
#undef EMMS
#undef SFENCE
#undef MMREG_SIZE
#undef PREFETCHW
#undef PAVGB

#ifdef HAVE_SSE2
#define MMREG_SIZE 16
#else
#define MMREG_SIZE 8
#endif

#ifdef HAVE_3DNOW
#define PREFETCH  "prefetch"
#define PREFETCHW "prefetchw"
49
#define PAVGB     "pavgusb"
50
#elif defined (HAVE_MMX2)
Michael Niedermayer's avatar
Michael Niedermayer committed
51 52
#define PREFETCH "prefetchnta"
#define PREFETCHW "prefetcht0"
53
#define PAVGB     "pavgb"
Michael Niedermayer's avatar
Michael Niedermayer committed
54
#else
55 56
#define PREFETCH  " # nop"
#define PREFETCHW " # nop"
57
#endif
Michael Niedermayer's avatar
Michael Niedermayer committed
58 59

#ifdef HAVE_3DNOW
Diego Biurrun's avatar
Diego Biurrun committed
60
/* On K6 femms is faster than emms. On K7 femms is directly mapped on emms. */
Michael Niedermayer's avatar
Michael Niedermayer committed
61 62 63
#define EMMS     "femms"
#else
#define EMMS     "emms"
64
#endif
Nick Kurshev's avatar
sfence  
Nick Kurshev committed
65

Michael Niedermayer's avatar
Michael Niedermayer committed
66 67 68 69 70
#ifdef HAVE_MMX2
#define MOVNTQ "movntq"
#define SFENCE "sfence"
#else
#define MOVNTQ "movq"
71
#define SFENCE " # nop"
72
#endif
Michael Niedermayer's avatar
Michael Niedermayer committed
73

74
static inline void RENAME(rgb24tobgr32)(const uint8_t *src, uint8_t *dst, long src_size)
Nick Kurshev's avatar
Nick Kurshev committed
75
{
76 77 78 79 80 81 82 83
    uint8_t *dest = dst;
    const uint8_t *s = src;
    const uint8_t *end;
    #ifdef HAVE_MMX
        const uint8_t *mm_end;
    #endif
    end = s + src_size;
    #ifdef HAVE_MMX
84
        asm volatile(PREFETCH"    %0"::"m"(*s):"memory");
85
        mm_end = end - 23;
86
        asm volatile("movq        %0, %%mm7"::"m"(mask32):"memory");
87 88
        while (s < mm_end)
        {
89
            asm volatile(
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
            PREFETCH"    32%1           \n\t"
            "movd          %1, %%mm0    \n\t"
            "punpckldq    3%1, %%mm0    \n\t"
            "movd         6%1, %%mm1    \n\t"
            "punpckldq    9%1, %%mm1    \n\t"
            "movd        12%1, %%mm2    \n\t"
            "punpckldq   15%1, %%mm2    \n\t"
            "movd        18%1, %%mm3    \n\t"
            "punpckldq   21%1, %%mm3    \n\t"
            "pand       %%mm7, %%mm0    \n\t"
            "pand       %%mm7, %%mm1    \n\t"
            "pand       %%mm7, %%mm2    \n\t"
            "pand       %%mm7, %%mm3    \n\t"
            MOVNTQ"     %%mm0,   %0     \n\t"
            MOVNTQ"     %%mm1,  8%0     \n\t"
            MOVNTQ"     %%mm2, 16%0     \n\t"
            MOVNTQ"     %%mm3, 24%0"
            :"=m"(*dest)
            :"m"(*s)
            :"memory");
            dest += 32;
            s += 24;
        }
113 114
        asm volatile(SFENCE:::"memory");
        asm volatile(EMMS:::"memory");
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
    #endif
    while (s < end)
    {
    #ifdef WORDS_BIGENDIAN
        /* RGB24 (= R,G,B) -> RGB32 (= A,B,G,R) */
        *dest++ = 0;
        *dest++ = s[2];
        *dest++ = s[1];
        *dest++ = s[0];
        s+=3;
    #else
        *dest++ = *s++;
        *dest++ = *s++;
        *dest++ = *s++;
        *dest++ = 0;
    #endif
    }
Nick Kurshev's avatar
Nick Kurshev committed
132
}
Nick Kurshev's avatar
Nick Kurshev committed
133

134
static inline void RENAME(rgb32tobgr24)(const uint8_t *src, uint8_t *dst, long src_size)
Nick Kurshev's avatar
Nick Kurshev committed
135
{
136 137 138
    uint8_t *dest = dst;
    const uint8_t *s = src;
    const uint8_t *end;
139
#ifdef HAVE_MMX
140
    const uint8_t *mm_end;
141
#endif
142
    end = s + src_size;
143
#ifdef HAVE_MMX
144
    asm volatile(PREFETCH"    %0"::"m"(*s):"memory");
145 146 147
    mm_end = end - 31;
    while (s < mm_end)
    {
148
        asm volatile(
149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198
        PREFETCH"    32%1           \n\t"
        "movq          %1, %%mm0    \n\t"
        "movq         8%1, %%mm1    \n\t"
        "movq        16%1, %%mm4    \n\t"
        "movq        24%1, %%mm5    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm1, %%mm3    \n\t"
        "movq       %%mm4, %%mm6    \n\t"
        "movq       %%mm5, %%mm7    \n\t"
        "psrlq         $8, %%mm2    \n\t"
        "psrlq         $8, %%mm3    \n\t"
        "psrlq         $8, %%mm6    \n\t"
        "psrlq         $8, %%mm7    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm1    \n\t"
        "pand          %2, %%mm4    \n\t"
        "pand          %2, %%mm5    \n\t"
        "pand          %3, %%mm2    \n\t"
        "pand          %3, %%mm3    \n\t"
        "pand          %3, %%mm6    \n\t"
        "pand          %3, %%mm7    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm3, %%mm1    \n\t"
        "por        %%mm6, %%mm4    \n\t"
        "por        %%mm7, %%mm5    \n\t"

        "movq       %%mm1, %%mm2    \n\t"
        "movq       %%mm4, %%mm3    \n\t"
        "psllq        $48, %%mm2    \n\t"
        "psllq        $32, %%mm3    \n\t"
        "pand          %4, %%mm2    \n\t"
        "pand          %5, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psrlq        $16, %%mm1    \n\t"
        "psrlq        $32, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm3, %%mm1    \n\t"
        "pand          %6, %%mm5    \n\t"
        "por        %%mm5, %%mm4    \n\t"

        MOVNTQ"     %%mm0,   %0     \n\t"
        MOVNTQ"     %%mm1,  8%0     \n\t"
        MOVNTQ"     %%mm4, 16%0"
        :"=m"(*dest)
        :"m"(*s),"m"(mask24l),
         "m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
        :"memory");
        dest += 24;
        s += 32;
    }
199 200
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
201 202 203
#endif
    while (s < end)
    {
Alex Beregszaszi's avatar
Alex Beregszaszi committed
204
#ifdef WORDS_BIGENDIAN
205 206 207 208 209 210
        /* RGB32 (= A,B,G,R) -> RGB24 (= R,G,B) */
        s++;
        dest[2] = *s++;
        dest[1] = *s++;
        dest[0] = *s++;
        dest += 3;
Alex Beregszaszi's avatar
Alex Beregszaszi committed
211
#else
212 213 214 215
        *dest++ = *s++;
        *dest++ = *s++;
        *dest++ = *s++;
        s++;
Alex Beregszaszi's avatar
Alex Beregszaszi committed
216
#endif
217
    }
Nick Kurshev's avatar
Nick Kurshev committed
218
}
Nick Kurshev's avatar
Nick Kurshev committed
219

220
/*
221 222
 original by Strepto/Astral
 ported to gcc & bugfixed: A'rpi
Nick Kurshev's avatar
Nick Kurshev committed
223
 MMX2, 3DNOW optimization by Nick Kurshev
224
 32-bit C version, and and&add trick by Michael Niedermayer
225
*/
Benoit Fouet's avatar
Benoit Fouet committed
226
static inline void RENAME(rgb15to16)(const uint8_t *src, uint8_t *dst, long src_size)
Nick Kurshev's avatar
Nick Kurshev committed
227
{
228 229 230 231 232
    register const uint8_t* s=src;
    register uint8_t* d=dst;
    register const uint8_t *end;
    const uint8_t *mm_end;
    end = s + src_size;
Nick Kurshev's avatar
Nick Kurshev committed
233
#ifdef HAVE_MMX
234 235
    asm volatile(PREFETCH"    %0"::"m"(*s));
    asm volatile("movq        %0, %%mm4"::"m"(mask15s));
236 237 238
    mm_end = end - 15;
    while (s<mm_end)
    {
239
        asm volatile(
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
        PREFETCH"  32%1         \n\t"
        "movq        %1, %%mm0  \n\t"
        "movq       8%1, %%mm2  \n\t"
        "movq     %%mm0, %%mm1  \n\t"
        "movq     %%mm2, %%mm3  \n\t"
        "pand     %%mm4, %%mm0  \n\t"
        "pand     %%mm4, %%mm2  \n\t"
        "paddw    %%mm1, %%mm0  \n\t"
        "paddw    %%mm3, %%mm2  \n\t"
        MOVNTQ"   %%mm0,  %0    \n\t"
        MOVNTQ"   %%mm2, 8%0"
        :"=m"(*d)
        :"m"(*s)
        );
        d+=16;
        s+=16;
    }
257 258
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
Nick Kurshev's avatar
Nick Kurshev committed
259
#endif
260
    mm_end = end - 3;
261
    while (s < mm_end)
Arpi's avatar
Arpi committed
262
    {
263
        register unsigned x= *((const uint32_t *)s);
264 265 266
        *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
        d+=4;
        s+=4;
Arpi's avatar
Arpi committed
267
    }
268
    if (s < end)
Arpi's avatar
Arpi committed
269
    {
270
        register unsigned short x= *((const uint16_t *)s);
271
        *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
Arpi's avatar
Arpi committed
272
    }
Nick Kurshev's avatar
Nick Kurshev committed
273
}
274

Benoit Fouet's avatar
Benoit Fouet committed
275
static inline void RENAME(rgb16to15)(const uint8_t *src, uint8_t *dst, long src_size)
276
{
277 278 279 280 281
    register const uint8_t* s=src;
    register uint8_t* d=dst;
    register const uint8_t *end;
    const uint8_t *mm_end;
    end = s + src_size;
282
#ifdef HAVE_MMX
283 284 285
    asm volatile(PREFETCH"    %0"::"m"(*s));
    asm volatile("movq        %0, %%mm7"::"m"(mask15rg));
    asm volatile("movq        %0, %%mm6"::"m"(mask15b));
286 287 288
    mm_end = end - 15;
    while (s<mm_end)
    {
289
        asm volatile(
290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310
        PREFETCH"  32%1         \n\t"
        "movq        %1, %%mm0  \n\t"
        "movq       8%1, %%mm2  \n\t"
        "movq     %%mm0, %%mm1  \n\t"
        "movq     %%mm2, %%mm3  \n\t"
        "psrlq       $1, %%mm0  \n\t"
        "psrlq       $1, %%mm2  \n\t"
        "pand     %%mm7, %%mm0  \n\t"
        "pand     %%mm7, %%mm2  \n\t"
        "pand     %%mm6, %%mm1  \n\t"
        "pand     %%mm6, %%mm3  \n\t"
        "por      %%mm1, %%mm0  \n\t"
        "por      %%mm3, %%mm2  \n\t"
        MOVNTQ"   %%mm0,  %0    \n\t"
        MOVNTQ"   %%mm2, 8%0"
        :"=m"(*d)
        :"m"(*s)
        );
        d+=16;
        s+=16;
    }
311 312
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
313
#endif
314
    mm_end = end - 3;
315
    while (s < mm_end)
316
    {
317
        register uint32_t x= *((const uint32_t*)s);
318 319 320
        *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
        s+=4;
        d+=4;
321
    }
322
    if (s < end)
323
    {
324
        register uint16_t x= *((const uint16_t*)s);
325 326 327
        *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
        s+=2;
        d+=2;
328 329 330
    }
}

331
static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, long src_size)
332
{
333 334
    const uint8_t *s = src;
    const uint8_t *end;
Arpi's avatar
Arpi committed
335
#ifdef HAVE_MMX
336
    const uint8_t *mm_end;
Arpi's avatar
Arpi committed
337
#endif
338 339
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
Arpi's avatar
Arpi committed
340
#ifdef HAVE_MMX
341
    mm_end = end - 15;
Diego Biurrun's avatar
Diego Biurrun committed
342
#if 1 //is faster only if multiplies are reasonably fast (FIXME figure out on which CPUs this is faster, on Athlon it is slightly faster)
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
    asm volatile(
    "movq           %3, %%mm5   \n\t"
    "movq           %4, %%mm6   \n\t"
    "movq           %5, %%mm7   \n\t"
    "jmp 2f                     \n\t"
    ASMALIGN(4)
    "1:                         \n\t"
    PREFETCH"   32(%1)          \n\t"
    "movd         (%1), %%mm0   \n\t"
    "movd        4(%1), %%mm3   \n\t"
    "punpckldq   8(%1), %%mm0   \n\t"
    "punpckldq  12(%1), %%mm3   \n\t"
    "movq        %%mm0, %%mm1   \n\t"
    "movq        %%mm3, %%mm4   \n\t"
    "pand        %%mm6, %%mm0   \n\t"
    "pand        %%mm6, %%mm3   \n\t"
    "pmaddwd     %%mm7, %%mm0   \n\t"
    "pmaddwd     %%mm7, %%mm3   \n\t"
    "pand        %%mm5, %%mm1   \n\t"
    "pand        %%mm5, %%mm4   \n\t"
    "por         %%mm1, %%mm0   \n\t"
    "por         %%mm4, %%mm3   \n\t"
    "psrld          $5, %%mm0   \n\t"
    "pslld         $11, %%mm3   \n\t"
    "por         %%mm3, %%mm0   \n\t"
    MOVNTQ"      %%mm0, (%0)    \n\t"
    "add           $16,  %1     \n\t"
    "add            $8,  %0     \n\t"
    "2:                         \n\t"
    "cmp            %2,  %1     \n\t"
    " jb            1b          \n\t"
    : "+r" (d), "+r"(s)
    : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216)
    );
Michael Niedermayer's avatar
Michael Niedermayer committed
377
#else
378 379
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
380 381 382 383 384
        "movq    %0, %%mm7    \n\t"
        "movq    %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    while (s < mm_end)
    {
385
        asm volatile(
386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         4%1, %%mm3    \n\t"
        "punpckldq    8%1, %%mm0    \n\t"
        "punpckldq   12%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psrlq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm3    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm3    \n\t"
        "psrlq         $5, %%mm1    \n\t"
        "psrlq         $5, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq         $8, %%mm2    \n\t"
        "psrlq         $8, %%mm5    \n\t"
        "pand       %%mm7, %%mm2    \n\t"
        "pand       %%mm7, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
        d += 4;
        s += 16;
    }
#endif
419 420
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
421 422 423
#endif
    while (s < end)
    {
424
        register int rgb = *(const uint32_t*)s; s += 4;
425 426
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);
    }
427 428
}

429
static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, long src_size)
430
{
431 432
    const uint8_t *s = src;
    const uint8_t *end;
433
#ifdef HAVE_MMX
434
    const uint8_t *mm_end;
435
#endif
436 437
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
438
#ifdef HAVE_MMX
439 440
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
441 442 443 444 445 446
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
447
        asm volatile(
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         4%1, %%mm3    \n\t"
        "punpckldq    8%1, %%mm0    \n\t"
        "punpckldq   12%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psllq         $8, %%mm0    \n\t"
        "psllq         $8, %%mm3    \n\t"
        "pand       %%mm7, %%mm0    \n\t"
        "pand       %%mm7, %%mm3    \n\t"
        "psrlq         $5, %%mm1    \n\t"
        "psrlq         $5, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq        $19, %%mm2    \n\t"
        "psrlq        $19, %%mm5    \n\t"
        "pand          %2, %%mm2    \n\t"
        "pand          %2, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
        d += 4;
        s += 16;
    }
480 481
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
482 483 484
#endif
    while (s < end)
    {
485
        register int rgb = *(const uint32_t*)s; s += 4;
486 487
        *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);
    }
488 489
}

490
static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, long src_size)
491
{
492 493
    const uint8_t *s = src;
    const uint8_t *end;
Arpi's avatar
Arpi committed
494
#ifdef HAVE_MMX
495
    const uint8_t *mm_end;
Arpi's avatar
Arpi committed
496
#endif
497 498
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
Arpi's avatar
Arpi committed
499
#ifdef HAVE_MMX
500
    mm_end = end - 15;
Diego Biurrun's avatar
Diego Biurrun committed
501
#if 1 //is faster only if multiplies are reasonably fast (FIXME figure out on which CPUs this is faster, on Athlon it is slightly faster)
502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535
    asm volatile(
    "movq           %3, %%mm5   \n\t"
    "movq           %4, %%mm6   \n\t"
    "movq           %5, %%mm7   \n\t"
    "jmp            2f          \n\t"
    ASMALIGN(4)
    "1:                         \n\t"
    PREFETCH"   32(%1)          \n\t"
    "movd         (%1), %%mm0   \n\t"
    "movd        4(%1), %%mm3   \n\t"
    "punpckldq   8(%1), %%mm0   \n\t"
    "punpckldq  12(%1), %%mm3   \n\t"
    "movq        %%mm0, %%mm1   \n\t"
    "movq        %%mm3, %%mm4   \n\t"
    "pand        %%mm6, %%mm0   \n\t"
    "pand        %%mm6, %%mm3   \n\t"
    "pmaddwd     %%mm7, %%mm0   \n\t"
    "pmaddwd     %%mm7, %%mm3   \n\t"
    "pand        %%mm5, %%mm1   \n\t"
    "pand        %%mm5, %%mm4   \n\t"
    "por         %%mm1, %%mm0   \n\t"
    "por         %%mm4, %%mm3   \n\t"
    "psrld          $6, %%mm0   \n\t"
    "pslld         $10, %%mm3   \n\t"
    "por         %%mm3, %%mm0   \n\t"
    MOVNTQ"      %%mm0, (%0)    \n\t"
    "add           $16,  %1     \n\t"
    "add            $8,  %0     \n\t"
    "2:                         \n\t"
    "cmp            %2,  %1     \n\t"
    " jb            1b          \n\t"
    : "+r" (d), "+r"(s)
    : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215)
    );
Michael Niedermayer's avatar
Michael Niedermayer committed
536
#else
537 538
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
539 540 541 542 543
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    while (s < mm_end)
    {
544
        asm volatile(
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         4%1, %%mm3    \n\t"
        "punpckldq    8%1, %%mm0    \n\t"
        "punpckldq   12%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psrlq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm3    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm3    \n\t"
        "psrlq         $6, %%mm1    \n\t"
        "psrlq         $6, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq         $9, %%mm2    \n\t"
        "psrlq         $9, %%mm5    \n\t"
        "pand       %%mm7, %%mm2    \n\t"
        "pand       %%mm7, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
        d += 4;
        s += 16;
    }
#endif
578 579
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
580 581 582
#endif
    while (s < end)
    {
583
        register int rgb = *(const uint32_t*)s; s += 4;
584 585
        *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);
    }
586 587
}

588
static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, long src_size)
589
{
590 591
    const uint8_t *s = src;
    const uint8_t *end;
592
#ifdef HAVE_MMX
593
    const uint8_t *mm_end;
594
#endif
595 596
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
597
#ifdef HAVE_MMX
598 599
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
600 601 602 603 604 605
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
606
        asm volatile(
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         4%1, %%mm3    \n\t"
        "punpckldq    8%1, %%mm0    \n\t"
        "punpckldq   12%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psllq         $7, %%mm0    \n\t"
        "psllq         $7, %%mm3    \n\t"
        "pand       %%mm7, %%mm0    \n\t"
        "pand       %%mm7, %%mm3    \n\t"
        "psrlq         $6, %%mm1    \n\t"
        "psrlq         $6, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq        $19, %%mm2    \n\t"
        "psrlq        $19, %%mm5    \n\t"
        "pand          %2, %%mm2    \n\t"
        "pand          %2, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
        d += 4;
        s += 16;
    }
639 640
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
641 642 643
#endif
    while (s < end)
    {
644
        register int rgb = *(const uint32_t*)s; s += 4;
645 646
        *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);
    }
647 648
}

649
static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, long src_size)
650
{
651 652
    const uint8_t *s = src;
    const uint8_t *end;
Arpi's avatar
Arpi committed
653
#ifdef HAVE_MMX
654
    const uint8_t *mm_end;
Arpi's avatar
Arpi committed
655
#endif
656 657
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
Arpi's avatar
Arpi committed
658
#ifdef HAVE_MMX
659 660
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
661 662 663 664 665 666
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 11;
    while (s < mm_end)
    {
667
        asm volatile(
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         3%1, %%mm3    \n\t"
        "punpckldq    6%1, %%mm0    \n\t"
        "punpckldq    9%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psrlq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm3    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm3    \n\t"
        "psrlq         $5, %%mm1    \n\t"
        "psrlq         $5, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq         $8, %%mm2    \n\t"
        "psrlq         $8, %%mm5    \n\t"
        "pand       %%mm7, %%mm2    \n\t"
        "pand       %%mm7, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
        d += 4;
        s += 12;
    }
700 701
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
702 703 704 705 706 707 708 709
#endif
    while (s < end)
    {
        const int b = *s++;
        const int g = *s++;
        const int r = *s++;
        *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
    }
710 711
}

712
static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, long src_size)
713
{
714 715
    const uint8_t *s = src;
    const uint8_t *end;
716
#ifdef HAVE_MMX
717
    const uint8_t *mm_end;
718
#endif
719 720
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
721
#ifdef HAVE_MMX
722 723
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
724 725 726 727 728 729
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_16mask),"m"(green_16mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
730
        asm volatile(
731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         3%1, %%mm3    \n\t"
        "punpckldq    6%1, %%mm0    \n\t"
        "punpckldq    9%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psllq         $8, %%mm0    \n\t"
        "psllq         $8, %%mm3    \n\t"
        "pand       %%mm7, %%mm0    \n\t"
        "pand       %%mm7, %%mm3    \n\t"
        "psrlq         $5, %%mm1    \n\t"
        "psrlq         $5, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq        $19, %%mm2    \n\t"
        "psrlq        $19, %%mm5    \n\t"
        "pand          %2, %%mm2    \n\t"
        "pand          %2, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
        d += 4;
        s += 12;
    }
763 764
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
765 766 767 768 769 770 771 772
#endif
    while (s < end)
    {
        const int r = *s++;
        const int g = *s++;
        const int b = *s++;
        *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
    }
773 774
}

775
static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, long src_size)
776
{
777 778
    const uint8_t *s = src;
    const uint8_t *end;
Arpi's avatar
Arpi committed
779
#ifdef HAVE_MMX
780
    const uint8_t *mm_end;
Arpi's avatar
Arpi committed
781
#endif
782 783
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
Arpi's avatar
Arpi committed
784
#ifdef HAVE_MMX
785 786
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
787 788 789 790 791 792
        "movq          %0, %%mm7    \n\t"
        "movq          %1, %%mm6    \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 11;
    while (s < mm_end)
    {
793
        asm volatile(
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
        PREFETCH"    32%1           \n\t"
        "movd          %1, %%mm0    \n\t"
        "movd         3%1, %%mm3    \n\t"
        "punpckldq    6%1, %%mm0    \n\t"
        "punpckldq    9%1, %%mm3    \n\t"
        "movq       %%mm0, %%mm1    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm3, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "psrlq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm3    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm3    \n\t"
        "psrlq         $6, %%mm1    \n\t"
        "psrlq         $6, %%mm4    \n\t"
        "pand       %%mm6, %%mm1    \n\t"
        "pand       %%mm6, %%mm4    \n\t"
        "psrlq         $9, %%mm2    \n\t"
        "psrlq         $9, %%mm5    \n\t"
        "pand       %%mm7, %%mm2    \n\t"
        "pand       %%mm7, %%mm5    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        "psllq        $16, %%mm3    \n\t"
        "por        %%mm3, %%mm0    \n\t"
        MOVNTQ"     %%mm0, %0       \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
        d += 4;
        s += 12;
    }
826 827
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
828 829 830 831 832 833 834 835
#endif
    while (s < end)
    {
        const int b = *s++;
        const int g = *s++;
        const int r = *s++;
        *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
    }
Arpi's avatar
Arpi committed
836 837
}

838
static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, long src_size)
839
{
840 841
    const uint8_t *s = src;
    const uint8_t *end;
842
#ifdef HAVE_MMX
843
    const uint8_t *mm_end;
844
#endif
845 846
    uint16_t *d = (uint16_t *)dst;
    end = s + src_size;
847
#ifdef HAVE_MMX
848 849
    asm volatile(PREFETCH"    %0"::"m"(*src):"memory");
    asm volatile(
850 851 852 853 854 855
        "movq         %0, %%mm7     \n\t"
        "movq         %1, %%mm6     \n\t"
        ::"m"(red_15mask),"m"(green_15mask));
    mm_end = end - 15;
    while (s < mm_end)
    {
856
        asm volatile(
857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888
        PREFETCH"   32%1            \n\t"
        "movd         %1, %%mm0     \n\t"
        "movd        3%1, %%mm3     \n\t"
        "punpckldq   6%1, %%mm0     \n\t"
        "punpckldq   9%1, %%mm3     \n\t"
        "movq      %%mm0, %%mm1     \n\t"
        "movq      %%mm0, %%mm2     \n\t"
        "movq      %%mm3, %%mm4     \n\t"
        "movq      %%mm3, %%mm5     \n\t"
        "psllq        $7, %%mm0     \n\t"
        "psllq        $7, %%mm3     \n\t"
        "pand      %%mm7, %%mm0     \n\t"
        "pand      %%mm7, %%mm3     \n\t"
        "psrlq        $6, %%mm1     \n\t"
        "psrlq        $6, %%mm4     \n\t"
        "pand      %%mm6, %%mm1     \n\t"
        "pand      %%mm6, %%mm4     \n\t"
        "psrlq       $19, %%mm2     \n\t"
        "psrlq       $19, %%mm5     \n\t"
        "pand         %2, %%mm2     \n\t"
        "pand         %2, %%mm5     \n\t"
        "por       %%mm1, %%mm0     \n\t"
        "por       %%mm4, %%mm3     \n\t"
        "por       %%mm2, %%mm0     \n\t"
        "por       %%mm5, %%mm3     \n\t"
        "psllq       $16, %%mm3     \n\t"
        "por       %%mm3, %%mm0     \n\t"
        MOVNTQ"    %%mm0, %0        \n\t"
        :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
        d += 4;
        s += 12;
    }
889 890
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
891 892 893 894 895 896 897 898
#endif
    while (s < end)
    {
        const int r = *s++;
        const int g = *s++;
        const int b = *s++;
        *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
    }
899 900
}

Arpi's avatar
Arpi committed
901
/*
Diego Biurrun's avatar
Diego Biurrun committed
902 903 904 905
  I use less accurate approximation here by simply left-shifting the input
  value and filling the low order bits with zeroes. This method improves PNG
  compression but this scheme cannot reproduce white exactly, since it does
  not generate an all-ones maximum value; the net effect is to darken the
Arpi's avatar
Arpi committed
906 907 908 909 910 911 912 913 914 915 916 917
  image slightly.

  The better method should be "left bit replication":

   4 3 2 1 0
   ---------
   1 1 0 1 1

   7 6 5 4 3  2 1 0
   ----------------
   1 1 0 1 1  1 1 0
   |=======|  |===|
918
       |      leftmost bits repeated to fill open bits
Arpi's avatar
Arpi committed
919
       |
920
   original bits
Arpi's avatar
Arpi committed
921
*/
922
static inline void RENAME(rgb15tobgr24)(const uint8_t *src, uint8_t *dst, long src_size)
Arpi's avatar
Arpi committed
923
{
924
    const uint16_t *end;
Arpi's avatar
Arpi committed
925
#ifdef HAVE_MMX
926
    const uint16_t *mm_end;
Arpi's avatar
Arpi committed
927
#endif
Baptiste Coudurier's avatar
Baptiste Coudurier committed
928
    uint8_t *d = dst;
929
    const uint16_t *s = (const uint16_t*)src;
930
    end = s + src_size/2;
Arpi's avatar
Arpi committed
931
#ifdef HAVE_MMX
932
    asm volatile(PREFETCH"    %0"::"m"(*s):"memory");
933 934 935
    mm_end = end - 7;
    while (s < mm_end)
    {
936
        asm volatile(
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
        PREFETCH"    32%1           \n\t"
        "movq          %1, %%mm0    \n\t"
        "movq          %1, %%mm1    \n\t"
        "movq          %1, %%mm2    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %3, %%mm1    \n\t"
        "pand          %4, %%mm2    \n\t"
        "psllq         $3, %%mm0    \n\t"
        "psrlq         $2, %%mm1    \n\t"
        "psrlq         $7, %%mm2    \n\t"
        "movq       %%mm0, %%mm3    \n\t"
        "movq       %%mm1, %%mm4    \n\t"
        "movq       %%mm2, %%mm5    \n\t"
        "punpcklwd     %5, %%mm0    \n\t"
        "punpcklwd     %5, %%mm1    \n\t"
        "punpcklwd     %5, %%mm2    \n\t"
        "punpckhwd     %5, %%mm3    \n\t"
        "punpckhwd     %5, %%mm4    \n\t"
        "punpckhwd     %5, %%mm5    \n\t"
        "psllq         $8, %%mm1    \n\t"
        "psllq        $16, %%mm2    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psllq         $8, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm5, %%mm3    \n\t"

        "movq       %%mm0, %%mm6    \n\t"
        "movq       %%mm3, %%mm7    \n\t"

        "movq         8%1, %%mm0    \n\t"
        "movq         8%1, %%mm1    \n\t"
        "movq         8%1, %%mm2    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %3, %%mm1    \n\t"
        "pand          %4, %%mm2    \n\t"
        "psllq         $3, %%mm0    \n\t"
        "psrlq         $2, %%mm1    \n\t"
        "psrlq         $7, %%mm2    \n\t"
        "movq       %%mm0, %%mm3    \n\t"
        "movq       %%mm1, %%mm4    \n\t"
        "movq       %%mm2, %%mm5    \n\t"
        "punpcklwd     %5, %%mm0    \n\t"
        "punpcklwd     %5, %%mm1    \n\t"
        "punpcklwd     %5, %%mm2    \n\t"
        "punpckhwd     %5, %%mm3    \n\t"
        "punpckhwd     %5, %%mm4    \n\t"
        "punpckhwd     %5, %%mm5    \n\t"
        "psllq         $8, %%mm1    \n\t"
        "psllq        $16, %%mm2    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psllq         $8, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm5, %%mm3    \n\t"

        :"=m"(*d)
        :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
        :"memory");
998
        /* borrowed 32 to 24 */
999
        asm volatile(
1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
        "movq       %%mm0, %%mm4    \n\t"
        "movq       %%mm3, %%mm5    \n\t"
        "movq       %%mm6, %%mm0    \n\t"
        "movq       %%mm7, %%mm1    \n\t"

        "movq       %%mm4, %%mm6    \n\t"
        "movq       %%mm5, %%mm7    \n\t"
        "movq       %%mm0, %%mm2    \n\t"
        "movq       %%mm1, %%mm3    \n\t"

        "psrlq         $8, %%mm2    \n\t"
        "psrlq         $8, %%mm3    \n\t"
        "psrlq         $8, %%mm6    \n\t"
        "psrlq         $8, %%mm7    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %2, %%mm1    \n\t"
        "pand          %2, %%mm4    \n\t"
        "pand          %2, %%mm5    \n\t"
        "pand          %3, %%mm2    \n\t"
        "pand          %3, %%mm3    \n\t"
        "pand          %3, %%mm6    \n\t"
        "pand          %3, %%mm7    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "por        %%mm3, %%mm1    \n\t"
        "por        %%mm6, %%mm4    \n\t"
        "por        %%mm7, %%mm5    \n\t"

        "movq       %%mm1, %%mm2    \n\t"
        "movq       %%mm4, %%mm3    \n\t"
        "psllq        $48, %%mm2    \n\t"
        "psllq        $32, %%mm3    \n\t"
        "pand          %4, %%mm2    \n\t"
        "pand          %5, %%mm3    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psrlq        $16, %%mm1    \n\t"
        "psrlq        $32, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm3, %%mm1    \n\t"
        "pand          %6, %%mm5    \n\t"
        "por        %%mm5, %%mm4    \n\t"

        MOVNTQ"     %%mm0,   %0     \n\t"
        MOVNTQ"     %%mm1,  8%0     \n\t"
        MOVNTQ"     %%mm4, 16%0"

        :"=m"(*d)
        :"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
        :"memory");
        d += 24;
        s += 8;
    }
1051 1052
    asm volatile(SFENCE:::"memory");
    asm volatile(EMMS:::"memory");
1053 1054 1055 1056 1057 1058 1059 1060 1061
#endif
    while (s < end)
    {
        register uint16_t bgr;
        bgr = *s++;
        *d++ = (bgr&0x1F)<<3;
        *d++ = (bgr&0x3E0)>>2;
        *d++ = (bgr&0x7C00)>>7;
    }
Arpi's avatar
Arpi committed
1062 1063
}

1064
static inline void RENAME(rgb16tobgr24)(const uint8_t *src, uint8_t *dst, long src_size)
Arpi's avatar
Arpi committed
1065
{
1066
    const uint16_t *end;
Arpi's avatar
Arpi committed
1067
#ifdef HAVE_MMX
1068
    const uint16_t *mm_end;
Arpi's avatar
Arpi committed
1069
#endif
1070 1071 1072
    uint8_t *d = (uint8_t *)dst;
    const uint16_t *s = (const uint16_t *)src;
    end = s + src_size/2;
Arpi's avatar
Arpi committed
1073
#ifdef HAVE_MMX
1074
    asm volatile(PREFETCH"    %0"::"m"(*s):"memory");
1075 1076 1077
    mm_end = end - 7;
    while (s < mm_end)
    {
1078
        asm volatile(
1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
        PREFETCH"    32%1           \n\t"
        "movq          %1, %%mm0    \n\t"
        "movq          %1, %%mm1    \n\t"
        "movq          %1, %%mm2    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %3, %%mm1    \n\t"
        "pand          %4, %%mm2    \n\t"
        "psllq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm1    \n\t"
        "psrlq         $8, %%mm2    \n\t"
        "movq       %%mm0, %%mm3    \n\t"
        "movq       %%mm1, %%mm4    \n\t"
        "movq       %%mm2, %%mm5    \n\t"
        "punpcklwd     %5, %%mm0    \n\t"
        "punpcklwd     %5, %%mm1    \n\t"
        "punpcklwd     %5, %%mm2    \n\t"
        "punpckhwd     %5, %%mm3    \n\t"
        "punpckhwd     %5, %%mm4    \n\t"
        "punpckhwd     %5, %%mm5    \n\t"
        "psllq         $8, %%mm1    \n\t"
        "psllq        $16, %%mm2    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psllq         $8, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm5, %%mm3    \n\t"

        "movq       %%mm0, %%mm6    \n\t"
        "movq       %%mm3, %%mm7    \n\t"

        "movq         8%1, %%mm0    \n\t"
        "movq         8%1, %%mm1    \n\t"
        "movq         8%1, %%mm2    \n\t"
        "pand          %2, %%mm0    \n\t"
        "pand          %3, %%mm1    \n\t"
        "pand          %4, %%mm2    \n\t"
        "psllq         $3, %%mm0    \n\t"
        "psrlq         $3, %%mm1    \n\t"
        "psrlq         $8, %%mm2    \n\t"
        "movq       %%mm0, %%mm3    \n\t"
        "movq       %%mm1, %%mm4    \n\t"
        "movq       %%mm2, %%mm5    \n\t"
        "punpcklwd     %5, %%mm0    \n\t"
        "punpcklwd     %5, %%mm1    \n\t"
        "punpcklwd     %5, %%mm2    \n\t"
        "punpckhwd     %5, %%mm3    \n\t"
        "punpckhwd     %5, %%mm4    \n\t"
        "punpckhwd     %5, %%mm5    \n\t"
        "psllq         $8, %%mm1    \n\t"
        "psllq        $16, %%mm2    \n\t"
        "por        %%mm1, %%mm0    \n\t"
        "por        %%mm2, %%mm0    \n\t"
        "psllq         $8, %%mm4    \n\t"
        "psllq        $16, %%mm5    \n\t"
        "por        %%mm4, %%mm3    \n\t"
        "por        %%mm5, %%mm3    \n\t"
        :"=m"(*d)
        :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)
        :"memory");
1139
        /* borrowed 32 to 24 */
1140
        asm volatile(
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158