dsputilenc_mmx.c 29.6 KB
Newer Older
1 2
/*
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard
4 5
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
 *
6 7
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
 *
8
 * This file is part of Libav.
9
 *
10
 * Libav is free software; you can redistribute it and/or
11 12 13 14
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
15
 * Libav is distributed in the hope that it will be useful,
16 17 18 19 20
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with Libav; if not, write to the Free Software
22 23 24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

25
#include "libavutil/attributes.h"
26
#include "libavutil/cpu.h"
27
#include "libavutil/x86/asm.h"
28
#include "libavutil/x86/cpu.h"
29
#include "libavcodec/dct.h"
30 31
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
32
#include "libavcodec/mathops.h"
33
#include "dsputil_x86.h"
34

Diego Biurrun's avatar
Diego Biurrun committed
35 36 37
void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size);
void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size);
void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2, int stride);
38 39
int ff_pix_sum16_mmx(uint8_t * pix, int line_size);
int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
40

41 42
#if HAVE_INLINE_ASM

43 44
static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;
45
  __asm__ volatile (
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98
      "movl %4,%%ecx\n"
      "shr $1,%%ecx\n"
      "pxor %%mm0,%%mm0\n"      /* mm0 = 0 */
      "pxor %%mm7,%%mm7\n"      /* mm7 holds the sum */
      "1:\n"
      "movq (%0),%%mm1\n"       /* mm1 = pix1[0][0-7] */
      "movq (%1),%%mm2\n"       /* mm2 = pix2[0][0-7] */
      "movq (%0,%3),%%mm3\n"    /* mm3 = pix1[1][0-7] */
      "movq (%1,%3),%%mm4\n"    /* mm4 = pix2[1][0-7] */

      /* todo: mm1-mm2, mm3-mm4 */
      /* algo: subtract mm1 from mm2 with saturation and vice versa */
      /*       OR the results to get absolute difference */
      "movq %%mm1,%%mm5\n"
      "movq %%mm3,%%mm6\n"
      "psubusb %%mm2,%%mm1\n"
      "psubusb %%mm4,%%mm3\n"
      "psubusb %%mm5,%%mm2\n"
      "psubusb %%mm6,%%mm4\n"

      "por %%mm1,%%mm2\n"
      "por %%mm3,%%mm4\n"

      /* now convert to 16-bit vectors so we can square them */
      "movq %%mm2,%%mm1\n"
      "movq %%mm4,%%mm3\n"

      "punpckhbw %%mm0,%%mm2\n"
      "punpckhbw %%mm0,%%mm4\n"
      "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
      "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */

      "pmaddwd %%mm2,%%mm2\n"
      "pmaddwd %%mm4,%%mm4\n"
      "pmaddwd %%mm1,%%mm1\n"
      "pmaddwd %%mm3,%%mm3\n"

      "lea (%0,%3,2), %0\n"     /* pix1 += 2*line_size */
      "lea (%1,%3,2), %1\n"     /* pix2 += 2*line_size */

      "paddd %%mm2,%%mm1\n"
      "paddd %%mm4,%%mm3\n"
      "paddd %%mm1,%%mm7\n"
      "paddd %%mm3,%%mm7\n"

      "decl %%ecx\n"
      "jnz 1b\n"

      "movq %%mm7,%%mm1\n"
      "psrlq $32, %%mm7\n"      /* shift hi dword to lo */
      "paddd %%mm7,%%mm1\n"
      "movd %%mm1,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
99
      : "r" ((x86_reg)line_size) , "m" (h)
100 101 102 103 104 105
      : "%ecx");
    return tmp;
}

static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;
106
  __asm__ volatile (
107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158
      "movl %4,%%ecx\n"
      "pxor %%mm0,%%mm0\n"      /* mm0 = 0 */
      "pxor %%mm7,%%mm7\n"      /* mm7 holds the sum */
      "1:\n"
      "movq (%0),%%mm1\n"       /* mm1 = pix1[0-7] */
      "movq (%1),%%mm2\n"       /* mm2 = pix2[0-7] */
      "movq 8(%0),%%mm3\n"      /* mm3 = pix1[8-15] */
      "movq 8(%1),%%mm4\n"      /* mm4 = pix2[8-15] */

      /* todo: mm1-mm2, mm3-mm4 */
      /* algo: subtract mm1 from mm2 with saturation and vice versa */
      /*       OR the results to get absolute difference */
      "movq %%mm1,%%mm5\n"
      "movq %%mm3,%%mm6\n"
      "psubusb %%mm2,%%mm1\n"
      "psubusb %%mm4,%%mm3\n"
      "psubusb %%mm5,%%mm2\n"
      "psubusb %%mm6,%%mm4\n"

      "por %%mm1,%%mm2\n"
      "por %%mm3,%%mm4\n"

      /* now convert to 16-bit vectors so we can square them */
      "movq %%mm2,%%mm1\n"
      "movq %%mm4,%%mm3\n"

      "punpckhbw %%mm0,%%mm2\n"
      "punpckhbw %%mm0,%%mm4\n"
      "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
      "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */

      "pmaddwd %%mm2,%%mm2\n"
      "pmaddwd %%mm4,%%mm4\n"
      "pmaddwd %%mm1,%%mm1\n"
      "pmaddwd %%mm3,%%mm3\n"

      "add %3,%0\n"
      "add %3,%1\n"

      "paddd %%mm2,%%mm1\n"
      "paddd %%mm4,%%mm3\n"
      "paddd %%mm1,%%mm7\n"
      "paddd %%mm3,%%mm7\n"

      "decl %%ecx\n"
      "jnz 1b\n"

      "movq %%mm7,%%mm1\n"
      "psrlq $32, %%mm7\n"      /* shift hi dword to lo */
      "paddd %%mm7,%%mm1\n"
      "movd %%mm1,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
159
      : "r" ((x86_reg)line_size) , "m" (h)
160 161 162 163 164 165
      : "%ecx");
    return tmp;
}

static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
    int tmp;
166
  __asm__ volatile (
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
      "movl %3,%%ecx\n"
      "pxor %%mm7,%%mm7\n"
      "pxor %%mm6,%%mm6\n"

      "movq (%0),%%mm0\n"
      "movq %%mm0, %%mm1\n"
      "psllq $8, %%mm0\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm0\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq %%mm4, %%mm1\n"
      "psllq $8, %%mm4\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm4\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "1:\n"

      "movq (%0),%%mm0\n"
      "movq %%mm0, %%mm1\n"
      "psllq $8, %%mm0\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm0\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"
      "psubw %%mm0, %%mm4\n"
      "psubw %%mm2, %%mm5\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm4, %%mm3\n\t"
      "pcmpgtw %%mm5, %%mm1\n\t"
      "pxor %%mm3, %%mm4\n"
      "pxor %%mm1, %%mm5\n"
      "psubw %%mm3, %%mm4\n"
      "psubw %%mm1, %%mm5\n"
      "paddw %%mm4, %%mm5\n"
      "paddw %%mm5, %%mm6\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq %%mm4, %%mm1\n"
      "psllq $8, %%mm4\n"
      "psrlq $8, %%mm1\n"
      "psrlq $8, %%mm4\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "subl $2, %%ecx\n"
      " jnz 1b\n"

      "movq %%mm6, %%mm0\n"
      "punpcklwd %%mm7,%%mm0\n"
      "punpckhwd %%mm7,%%mm6\n"
      "paddd %%mm0, %%mm6\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddd %%mm6,%%mm0\n"
      "movd %%mm0,%1\n"
      : "+r" (pix1), "=r"(tmp)
284
      : "r" ((x86_reg)line_size) , "g" (h-2)
285 286 287 288 289 290 291
      : "%ecx");
      return tmp;
}

static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
    int tmp;
    uint8_t * pix= pix1;
292
  __asm__ volatile (
293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
      "movl %3,%%ecx\n"
      "pxor %%mm7,%%mm7\n"
      "pxor %%mm6,%%mm6\n"

      "movq (%0),%%mm0\n"
      "movq 1(%0),%%mm1\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq 1(%0),%%mm1\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "1:\n"

      "movq (%0),%%mm0\n"
      "movq 1(%0),%%mm1\n"
      "movq %%mm0, %%mm2\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm0\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm2\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm0\n"
      "psubw %%mm3, %%mm2\n"
      "psubw %%mm0, %%mm4\n"
      "psubw %%mm2, %%mm5\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm4, %%mm3\n\t"
      "pcmpgtw %%mm5, %%mm1\n\t"
      "pxor %%mm3, %%mm4\n"
      "pxor %%mm1, %%mm5\n"
      "psubw %%mm3, %%mm4\n"
      "psubw %%mm1, %%mm5\n"
      "paddw %%mm4, %%mm5\n"
      "paddw %%mm5, %%mm6\n"

      "add %2,%0\n"

      "movq (%0),%%mm4\n"
      "movq 1(%0),%%mm1\n"
      "movq %%mm4, %%mm5\n"
      "movq %%mm1, %%mm3\n"
      "punpcklbw %%mm7,%%mm4\n"
      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "subl $2, %%ecx\n"
      " jnz 1b\n"

      "movq %%mm6, %%mm0\n"
      "punpcklwd %%mm7,%%mm0\n"
      "punpckhwd %%mm7,%%mm6\n"
      "paddd %%mm0, %%mm6\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddd %%mm6,%%mm0\n"
      "movd %%mm0,%1\n"
      : "+r" (pix1), "=r"(tmp)
398
      : "r" ((x86_reg)line_size) , "g" (h-2)
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453
      : "%ecx");
      return tmp + hf_noise8_mmx(pix+8, line_size, h);
}

static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    MpegEncContext *c = p;
    int score1, score2;

    if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
    else  score1 = sse16_mmx(c, pix1, pix2, line_size, h);
    score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

    if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
    else  return score1 + FFABS(score2)*8;
}

static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    MpegEncContext *c = p;
    int score1= sse8_mmx(c, pix1, pix2, line_size, h);
    int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

    if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
    else  return score1 + FFABS(score2)*8;
}

static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
    int tmp;

    assert( (((int)pix) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0), %%mm2\n"\
      "movq 8(%0), %%mm3\n"\
      "add %2,%0\n"\
      "movq %%mm2, " #out0 "\n"\
      "movq %%mm3, " #out1 "\n"\
      "psubusb " #in0 ", %%mm2\n"\
      "psubusb " #in1 ", %%mm3\n"\
      "psubusb " #out0 ", " #in0 "\n"\
      "psubusb " #out1 ", " #in1 "\n"\
      "por %%mm2, " #in0 "\n"\
      "por %%mm3, " #in1 "\n"\
      "movq " #in0 ", %%mm2\n"\
      "movq " #in1 ", %%mm3\n"\
      "punpcklbw %%mm7, " #in0 "\n"\
      "punpcklbw %%mm7, " #in1 "\n"\
      "punpckhbw %%mm7, %%mm2\n"\
      "punpckhbw %%mm7, %%mm3\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw %%mm3, %%mm2\n"\
      "paddw %%mm2, " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"


454
  __asm__ volatile (
455 456 457 458 459 460
      "movl %3,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pxor %%mm7,%%mm7\n"
      "movq (%0),%%mm0\n"
      "movq 8(%0),%%mm1\n"
      "add %2,%0\n"
461
      "jmp 2f\n"
462 463 464
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)
465
      "2:\n"
466 467 468 469 470 471 472 473 474 475 476 477 478
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddw %%mm6,%%mm0\n"
      "movq %%mm0,%%mm6\n"
      "psrlq $16, %%mm0\n"
      "paddw %%mm6,%%mm0\n"
      "movd %%mm0,%1\n"
      : "+r" (pix), "=r"(tmp)
479
      : "r" ((x86_reg)line_size) , "m" (h)
480 481 482 483 484
      : "%ecx");
    return tmp & 0xFFFF;
}
#undef SUM

485 486 487
static int vsad_intra16_mmxext(void *v, uint8_t *pix, uint8_t *dummy,
                               int line_size, int h)
{
488 489 490 491 492 493 494 495 496 497 498 499 500 501
    int tmp;

    assert( (((int)pix) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0), " #out0 "\n"\
      "movq 8(%0), " #out1 "\n"\
      "add %2,%0\n"\
      "psadbw " #out0 ", " #in0 "\n"\
      "psadbw " #out1 ", " #in1 "\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"

502
  __asm__ volatile (
503 504 505 506 507 508
      "movl %3,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pxor %%mm7,%%mm7\n"
      "movq (%0),%%mm0\n"
      "movq 8(%0),%%mm1\n"
      "add %2,%0\n"
509
      "jmp 2f\n"
510 511 512
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)
513
      "2:\n"
514 515 516 517 518 519 520
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movd %%mm6,%1\n"
      : "+r" (pix), "=r"(tmp)
521
      : "r" ((x86_reg)line_size) , "m" (h)
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
      : "%ecx");
    return tmp;
}
#undef SUM

static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;

    assert( (((int)pix1) & 7) == 0);
    assert( (((int)pix2) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0),%%mm2\n"\
      "movq (%1)," #out0 "\n"\
      "movq 8(%0),%%mm3\n"\
      "movq 8(%1)," #out1 "\n"\
      "add %3,%0\n"\
      "add %3,%1\n"\
      "psubb " #out0 ", %%mm2\n"\
      "psubb " #out1 ", %%mm3\n"\
      "pxor %%mm7, %%mm2\n"\
      "pxor %%mm7, %%mm3\n"\
      "movq %%mm2, " #out0 "\n"\
      "movq %%mm3, " #out1 "\n"\
      "psubusb " #in0 ", %%mm2\n"\
      "psubusb " #in1 ", %%mm3\n"\
      "psubusb " #out0 ", " #in0 "\n"\
      "psubusb " #out1 ", " #in1 "\n"\
      "por %%mm2, " #in0 "\n"\
      "por %%mm3, " #in1 "\n"\
      "movq " #in0 ", %%mm2\n"\
      "movq " #in1 ", %%mm3\n"\
      "punpcklbw %%mm7, " #in0 "\n"\
      "punpcklbw %%mm7, " #in1 "\n"\
      "punpckhbw %%mm7, %%mm2\n"\
      "punpckhbw %%mm7, %%mm3\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw %%mm3, %%mm2\n"\
      "paddw %%mm2, " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"


565
  __asm__ volatile (
566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
      "movl %4,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pcmpeqw %%mm7,%%mm7\n"
      "psllw $15, %%mm7\n"
      "packsswb %%mm7, %%mm7\n"
      "movq (%0),%%mm0\n"
      "movq (%1),%%mm2\n"
      "movq 8(%0),%%mm1\n"
      "movq 8(%1),%%mm3\n"
      "add %3,%0\n"
      "add %3,%1\n"
      "psubb %%mm2, %%mm0\n"
      "psubb %%mm3, %%mm1\n"
      "pxor %%mm7, %%mm0\n"
      "pxor %%mm7, %%mm1\n"
581
      "jmp 2f\n"
582 583 584
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)
585
      "2:\n"
586 587 588 589 590 591 592 593 594 595 596 597 598
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddw %%mm6,%%mm0\n"
      "movq %%mm0,%%mm6\n"
      "psrlq $16, %%mm0\n"
      "paddw %%mm6,%%mm0\n"
      "movd %%mm0,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
599
      : "r" ((x86_reg)line_size) , "m" (h)
600 601 602 603 604
      : "%ecx");
    return tmp & 0x7FFF;
}
#undef SUM

605 606 607
static int vsad16_mmxext(void *v, uint8_t *pix1, uint8_t *pix2,
                         int line_size, int h)
{
608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
    int tmp;

    assert( (((int)pix1) & 7) == 0);
    assert( (((int)pix2) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0)," #out0 "\n"\
      "movq (%1),%%mm2\n"\
      "movq 8(%0)," #out1 "\n"\
      "movq 8(%1),%%mm3\n"\
      "add %3,%0\n"\
      "add %3,%1\n"\
      "psubb %%mm2, " #out0 "\n"\
      "psubb %%mm3, " #out1 "\n"\
      "pxor %%mm7, " #out0 "\n"\
      "pxor %%mm7, " #out1 "\n"\
      "psadbw " #out0 ", " #in0 "\n"\
      "psadbw " #out1 ", " #in1 "\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"

630
  __asm__ volatile (
631 632 633 634 635 636 637 638 639 640 641 642 643 644 645
      "movl %4,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pcmpeqw %%mm7,%%mm7\n"
      "psllw $15, %%mm7\n"
      "packsswb %%mm7, %%mm7\n"
      "movq (%0),%%mm0\n"
      "movq (%1),%%mm2\n"
      "movq 8(%0),%%mm1\n"
      "movq 8(%1),%%mm3\n"
      "add %3,%0\n"
      "add %3,%1\n"
      "psubb %%mm2, %%mm0\n"
      "psubb %%mm3, %%mm1\n"
      "pxor %%mm7, %%mm0\n"
      "pxor %%mm7, %%mm1\n"
646
      "jmp 2f\n"
647 648 649
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)
650
      "2:\n"
651 652 653 654 655 656 657
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movd %%mm6,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
658
      : "r" ((x86_reg)line_size) , "m" (h)
659 660 661 662 663 664
      : "%ecx");
    return tmp;
}
#undef SUM

static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
665
    x86_reg i=0;
666
    __asm__ volatile(
667 668 669 670 671 672 673 674 675 676 677 678 679
        "1:                             \n\t"
        "movq  (%2, %0), %%mm0          \n\t"
        "movq  (%1, %0), %%mm1          \n\t"
        "psubb %%mm0, %%mm1             \n\t"
        "movq %%mm1, (%3, %0)           \n\t"
        "movq 8(%2, %0), %%mm0          \n\t"
        "movq 8(%1, %0), %%mm1          \n\t"
        "psubb %%mm0, %%mm1             \n\t"
        "movq %%mm1, 8(%3, %0)          \n\t"
        "add $16, %0                    \n\t"
        "cmp %4, %0                     \n\t"
        " jb 1b                         \n\t"
        : "+r" (i)
680
        : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w-15)
681 682 683 684 685
    );
    for(; i<w; i++)
        dst[i+0] = src1[i+0]-src2[i+0];
}

686 687 688 689
static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,
                                              const uint8_t *src2, int w,
                                              int *left, int *left_top)
{
690
    x86_reg i=0;
691 692
    uint8_t l, lt;

693
    __asm__ volatile(
694 695
        "movq  (%1, %0), %%mm0          \n\t" // LT
        "psllq $8, %%mm0                \n\t"
696 697 698 699 700 701 702 703 704 705 706 707 708 709 710
        "1:                             \n\t"
        "movq  (%1, %0), %%mm1          \n\t" // T
        "movq  -1(%2, %0), %%mm2        \n\t" // L
        "movq  (%2, %0), %%mm3          \n\t" // X
        "movq %%mm2, %%mm4              \n\t" // L
        "psubb %%mm0, %%mm2             \n\t"
        "paddb %%mm1, %%mm2             \n\t" // L + T - LT
        "movq %%mm4, %%mm5              \n\t" // L
        "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)
        "pminub %%mm5, %%mm1            \n\t" // min(T, L)
        "pminub %%mm2, %%mm4            \n\t"
        "pmaxub %%mm1, %%mm4            \n\t"
        "psubb %%mm4, %%mm3             \n\t" // dst - pred
        "movq %%mm3, (%3, %0)           \n\t"
        "add $8, %0                     \n\t"
711
        "movq -1(%1, %0), %%mm0         \n\t" // LT
712 713 714
        "cmp %4, %0                     \n\t"
        " jb 1b                         \n\t"
        : "+r" (i)
715
        : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w)
716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
    );

    l= *left;
    lt= *left_top;

    dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

    *left_top= src1[w-1];
    *left    = src2[w-1];
}

#define MMABS_MMX(a,z)\
    "pxor " #z ", " #z "              \n\t"\
    "pcmpgtw " #a ", " #z "           \n\t"\
    "pxor " #z ", " #a "              \n\t"\
    "psubw " #z ", " #a "             \n\t"

733
#define MMABS_MMXEXT(a, z)                 \
734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
    "pxor " #z ", " #z "              \n\t"\
    "psubw " #a ", " #z "             \n\t"\
    "pmaxsw " #z ", " #a "            \n\t"

#define MMABS_SSSE3(a,z)\
    "pabsw " #a ", " #a "             \n\t"

#define MMABS_SUM(a,z, sum)\
    MMABS(a,z)\
    "paddusw " #a ", " #sum "         \n\t"

/* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
 * about 100k on extreme inputs. But that's very unlikely to occur in natural video,
 * and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
#define HSUM_MMX(a, t, dst)\
    "movq "#a", "#t"                  \n\t"\
    "psrlq $32, "#a"                  \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "movq "#a", "#t"                  \n\t"\
    "psrlq $16, "#a"                  \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "movd "#a", "#dst"                \n\t"\

757
#define HSUM_MMXEXT(a, t, dst)             \
758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801
    "pshufw $0x0E, "#a", "#t"         \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "pshufw $0x01, "#a", "#t"         \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "movd "#a", "#dst"                \n\t"\

#define HSUM_SSE2(a, t, dst)\
    "movhlps "#a", "#t"               \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "pshuflw $0x0E, "#a", "#t"        \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "pshuflw $0x01, "#a", "#t"        \n\t"\
    "paddusw "#t", "#a"               \n\t"\
    "movd "#a", "#dst"                \n\t"\

#define DCT_SAD4(m,mm,o)\
    "mov"#m" "#o"+ 0(%1), "#mm"2      \n\t"\
    "mov"#m" "#o"+16(%1), "#mm"3      \n\t"\
    "mov"#m" "#o"+32(%1), "#mm"4      \n\t"\
    "mov"#m" "#o"+48(%1), "#mm"5      \n\t"\
    MMABS_SUM(mm##2, mm##6, mm##0)\
    MMABS_SUM(mm##3, mm##7, mm##1)\
    MMABS_SUM(mm##4, mm##6, mm##0)\
    MMABS_SUM(mm##5, mm##7, mm##1)\

#define DCT_SAD_MMX\
    "pxor %%mm0, %%mm0                \n\t"\
    "pxor %%mm1, %%mm1                \n\t"\
    DCT_SAD4(q, %%mm, 0)\
    DCT_SAD4(q, %%mm, 8)\
    DCT_SAD4(q, %%mm, 64)\
    DCT_SAD4(q, %%mm, 72)\
    "paddusw %%mm1, %%mm0             \n\t"\
    HSUM(%%mm0, %%mm1, %0)

#define DCT_SAD_SSE2\
    "pxor %%xmm0, %%xmm0              \n\t"\
    "pxor %%xmm1, %%xmm1              \n\t"\
    DCT_SAD4(dqa, %%xmm, 0)\
    DCT_SAD4(dqa, %%xmm, 64)\
    "paddusw %%xmm1, %%xmm0           \n\t"\
    HSUM(%%xmm0, %%xmm1, %0)

#define DCT_SAD_FUNC(cpu) \
Diego Biurrun's avatar
Diego Biurrun committed
802
static int sum_abs_dctelem_##cpu(int16_t *block){\
803
    int sum;\
804
    __asm__ volatile(\
805 806 807 808 809 810 811 812 813 814 815 816 817 818
        DCT_SAD\
        :"=r"(sum)\
        :"r"(block)\
    );\
    return sum&0xFFFF;\
}

#define DCT_SAD       DCT_SAD_MMX
#define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
#define MMABS(a,z)    MMABS_MMX(a,z)
DCT_SAD_FUNC(mmx)
#undef MMABS
#undef HSUM

819 820
#define HSUM(a,t,dst) HSUM_MMXEXT(a,t,dst)
#define MMABS(a,z)    MMABS_MMXEXT(a,z)
821
DCT_SAD_FUNC(mmxext)
822 823 824 825 826 827 828 829
#undef HSUM
#undef DCT_SAD

#define DCT_SAD       DCT_SAD_SSE2
#define HSUM(a,t,dst) HSUM_SSE2(a,t,dst)
DCT_SAD_FUNC(sse2)
#undef MMABS

830
#if HAVE_SSSE3_INLINE
831 832 833 834 835 836 837 838 839
#define MMABS(a,z)    MMABS_SSSE3(a,z)
DCT_SAD_FUNC(ssse3)
#undef MMABS
#endif
#undef HSUM
#undef DCT_SAD

static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
    int sum;
840
    x86_reg i=size;
841
    __asm__ volatile(
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888
        "pxor %%mm4, %%mm4 \n"
        "1: \n"
        "sub $8, %0 \n"
        "movq (%2,%0), %%mm2 \n"
        "movq (%3,%0,2), %%mm0 \n"
        "movq 8(%3,%0,2), %%mm1 \n"
        "punpckhbw %%mm2, %%mm3 \n"
        "punpcklbw %%mm2, %%mm2 \n"
        "psraw $8, %%mm3 \n"
        "psraw $8, %%mm2 \n"
        "psubw %%mm3, %%mm1 \n"
        "psubw %%mm2, %%mm0 \n"
        "pmaddwd %%mm1, %%mm1 \n"
        "pmaddwd %%mm0, %%mm0 \n"
        "paddd %%mm1, %%mm4 \n"
        "paddd %%mm0, %%mm4 \n"
        "jg 1b \n"
        "movq %%mm4, %%mm3 \n"
        "psrlq $32, %%mm3 \n"
        "paddd %%mm3, %%mm4 \n"
        "movd %%mm4, %1 \n"
        :"+r"(i), "=r"(sum)
        :"r"(pix1), "r"(pix2)
    );
    return sum;
}

#define PHADDD(a, t)\
    "movq "#a", "#t"                  \n\t"\
    "psrlq $32, "#a"                  \n\t"\
    "paddd "#t", "#a"                 \n\t"
/*
   pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
   pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
   pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
 */
#define PMULHRW(x, y, s, o)\
    "pmulhw " #s ", "#x "            \n\t"\
    "pmulhw " #s ", "#y "            \n\t"\
    "paddw " #o ", "#x "             \n\t"\
    "paddw " #o ", "#y "             \n\t"\
    "psraw $1, "#x "                 \n\t"\
    "psraw $1, "#y "                 \n\t"
#define DEF(x) x ## _mmx
#define SET_RND MOVQ_WONE
#define SCALE_OFFSET 1

889
#include "dsputil_qns_template.c"
890 891 892 893 894 895 896 897 898 899 900 901 902

#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW

#define DEF(x) x ## _3dnow
#define SET_RND(x)
#define SCALE_OFFSET 0
#define PMULHRW(x, y, s, o)\
    "pmulhrw " #s ", "#x "           \n\t"\
    "pmulhrw " #s ", "#y "           \n\t"

903
#include "dsputil_qns_template.c"
904 905 906 907 908 909

#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW

910
#if HAVE_SSSE3_INLINE
911 912 913 914 915 916 917 918 919 920 921
#undef PHADDD
#define DEF(x) x ## _ssse3
#define SET_RND(x)
#define SCALE_OFFSET -1
#define PHADDD(a, t)\
    "pshufw $0x0E, "#a", "#t"         \n\t"\
    "paddd "#t", "#a"                 \n\t" /* faster than phaddd on core2 */
#define PMULHRW(x, y, s, o)\
    "pmulhrsw " #s ", "#x "          \n\t"\
    "pmulhrsw " #s ", "#y "          \n\t"

922
#include "dsputil_qns_template.c"
923 924 925 926 927 928

#undef DEF
#undef SET_RND
#undef SCALE_OFFSET
#undef PMULHRW
#undef PHADDD
929
#endif /* HAVE_SSSE3_INLINE */
930

931 932 933 934 935 936 937 938 939 940 941
#endif /* HAVE_INLINE_ASM */

int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);

#define hadamard_func(cpu) \
int ff_hadamard8_diff_##cpu  (void *s, uint8_t *src1, uint8_t *src2, \
                              int stride, int h); \
int ff_hadamard8_diff16_##cpu(void *s, uint8_t *src1, uint8_t *src2, \
                              int stride, int h);

hadamard_func(mmx)
942
hadamard_func(mmxext)
943 944
hadamard_func(sse2)
hadamard_func(ssse3)
945

946
av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
947
{
948
    int cpu_flags = av_get_cpu_flags();
949
    const int dct_algo = avctx->dct_algo;
950

951
#if HAVE_YASM
952 953
    int bit_depth = avctx->bits_per_raw_sample;

954
    if (EXTERNAL_MMX(cpu_flags)) {
955 956 957 958 959 960 961
        if (bit_depth <= 8)
            c->get_pixels = ff_get_pixels_mmx;
        c->diff_pixels = ff_diff_pixels_mmx;
        c->pix_sum = ff_pix_sum16_mmx;

        c->pix_norm1 = ff_pix_norm1_mmx;
    }
962
    if (EXTERNAL_SSE2(cpu_flags))
963 964 965 966 967
        if (bit_depth <= 8)
            c->get_pixels = ff_get_pixels_sse2;
#endif /* HAVE_YASM */

#if HAVE_INLINE_ASM
968
    if (INLINE_MMX(cpu_flags)) {
969
        if (avctx->bits_per_raw_sample <= 8 &&
970 971
            (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
            c->fdct = ff_fdct_mmx;
972 973 974 975

        c->diff_bytes= diff_bytes_mmx;
        c->sum_abs_dctelem= sum_abs_dctelem_mmx;

976 977
        c->sse[0] = sse16_mmx;
        c->sse[1] = sse8_mmx;
978 979 980 981 982 983 984 985 986 987 988 989 990 991
        c->vsad[4]= vsad_intra16_mmx;

        c->nsse[0] = nsse16_mmx;
        c->nsse[1] = nsse8_mmx;
        if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
            c->vsad[0] = vsad16_mmx;
        }

        if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
            c->try_8x8basis= try_8x8basis_mmx;
        }
        c->add_8x8basis= add_8x8basis_mmx;

        c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
992
    }
993

994 995 996 997 998 999 1000
    if (INLINE_AMD3DNOW(cpu_flags)) {
        if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
            c->try_8x8basis = try_8x8basis_3dnow;
        }
        c->add_8x8basis = add_8x8basis_3dnow;
    }

1001
    if (INLINE_MMXEXT(cpu_flags)) {
1002 1003 1004 1005
        if (avctx->bits_per_raw_sample <= 8 &&
            (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
            c->fdct = ff_fdct_mmxext;

1006 1007
        c->sum_abs_dctelem = sum_abs_dctelem_mmxext;
        c->vsad[4]         = vsad_intra16_mmxext;
1008

1009 1010
        if (!(avctx->flags & CODEC_FLAG_BITEXACT)){
            c->vsad[0] = vsad16_mmxext;
1011 1012
        }

1013 1014 1015 1016
        c->sub_hfyu_median_prediction = sub_hfyu_median_prediction_mmxext;
    }

    if (INLINE_SSE2(cpu_flags)) {
1017 1018 1019 1020
        if (avctx->bits_per_raw_sample <= 8 &&
            (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
            c->fdct = ff_fdct_sse2;

1021 1022
        c->sum_abs_dctelem= sum_abs_dctelem_sse2;
    }
1023

1024
#if HAVE_SSSE3_INLINE
1025 1026 1027
    if (INLINE_SSSE3(cpu_flags)) {
        if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
            c->try_8x8basis = try_8x8basis_ssse3;
1028
        }
1029 1030 1031
        c->add_8x8basis    = add_8x8basis_ssse3;
        c->sum_abs_dctelem = sum_abs_dctelem_ssse3;
    }
1032
#endif
1033 1034
#endif /* HAVE_INLINE_ASM */

1035
    if (EXTERNAL_MMX(cpu_flags)) {
1036 1037
        c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;
        c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;
1038
    }
1039

1040 1041 1042 1043
    if (EXTERNAL_MMXEXT(cpu_flags)) {
        c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;
        c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;
    }
1044

1045 1046
    if (EXTERNAL_SSE2(cpu_flags)) {
        c->sse[0] = ff_sse16_sse2;
1047 1048

#if HAVE_ALIGNED_STACK
1049 1050
        c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
        c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
1051
#endif
1052
    }
1053

1054 1055 1056
    if (EXTERNAL_SSSE3(cpu_flags) && HAVE_ALIGNED_STACK) {
        c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
        c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
1057
    }
1058

1059
    ff_dsputil_init_pix_mmx(c, avctx);
1060
}