fdct_mmx.c 15.6 KB
Newer Older
1 2
/*
 * MMX optimized forward DCT
3
 * The gcc porting is Copyright (c) 2001 Fabrice Bellard.
4
 * cleanup/optimizations are Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 * SSE2 optimization is Copyright (c) 2004 Denes Balatoni.
6 7
 *
 * from  fdctam32.c - AP922 MMX(3D-Now) forward-DCT
8
 *
9 10
 *  Intel Application Note AP-922 - fast, precise implementation of DCT
 *        http://developer.intel.com/vtune/cbts/appnotes.htm
11 12 13 14
 *
 * Also of inspiration:
 * a page about fdct at http://www.geocities.com/ssavekar/dct.htm
 * Skal's fdct at http://skal.planet-d.net/coding/dct.html
15
 *
16 17 18
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
19 20
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
21
 * version 2.1 of the License, or (at your option) any later version.
22
 *
23
 * FFmpeg is distributed in the hope that it will be useful,
24 25 26 27 28
 * 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
29
 * License along with FFmpeg; if not, write to the Free Software
30
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
31
 */
32
#include "common.h"
33
#include "../dsputil.h"
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
#include "mmx.h"

#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))

//////////////////////////////////////////////////////////////////////
//
// constants for the forward DCT
// -----------------------------
//
// Be sure to check that your compiler is aligning all constants to QWORD
// (8-byte) memory boundaries!  Otherwise the unaligned memory access will
// severely stall MMX execution.
//
//////////////////////////////////////////////////////////////////////

49 50 51 52 53
#define BITS_FRW_ACC   3 //; 2 or 3 for accuracy
#define SHIFT_FRW_COL  BITS_FRW_ACC
#define SHIFT_FRW_ROW  (BITS_FRW_ACC + 17 - 3)
#define RND_FRW_ROW    (1 << (SHIFT_FRW_ROW-1))
//#define RND_FRW_COL    (1 << (SHIFT_FRW_COL-1))
54 55

//concatenated table, for forward DCT transformation
Michael Niedermayer's avatar
Michael Niedermayer committed
56
static const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = {
57 58 59
    13036,  13036,  13036,  13036,        // tg * (2<<16) + 0.5
    27146,  27146,  27146,  27146,        // tg * (2<<16) + 0.5
   -21746, -21746, -21746, -21746,        // tg * (2<<16) + 0.5
60 61
};

Michael Niedermayer's avatar
Michael Niedermayer committed
62
static const int16_t ocos_4_16[4] ATTR_ALIGN(8) = {
63
    23170, 23170, 23170, 23170,           //cos * (2<<15) + 0.5
64 65
};

66
static const int64_t fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL;
67

68
static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };
69

70
static struct
71
{
72
 const int32_t fdct_r_row_sse2[4] ATTR_ALIGN(16);
73 74 75 76 77
} fdct_r_row_sse2 ATTR_ALIGN(16)=
{{
 RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
}};
//static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};
78

Michael Niedermayer's avatar
Michael Niedermayer committed
79
static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
  16384,   16384,   22725,   19266,
  16384,   16384,   12873,    4520,
  21407,    8867,   19266,   -4520,
  -8867,  -21407,  -22725,  -12873,
  16384,  -16384,   12873,  -22725,
 -16384,   16384,    4520,   19266,
   8867,  -21407,    4520,  -12873,
  21407,   -8867,   19266,  -22725,

  22725,   22725,   31521,   26722,
  22725,   22725,   17855,    6270,
  29692,   12299,   26722,   -6270,
 -12299,  -29692,  -31521,  -17855,
  22725,  -22725,   17855,  -31521,
 -22725,   22725,    6270,   26722,
  12299,  -29692,    6270,  -17855,
  29692,  -12299,   26722,  -31521,

  21407,   21407,   29692,   25172,
  21407,   21407,   16819,    5906,
  27969,   11585,   25172,   -5906,
 -11585,  -27969,  -29692,  -16819,
  21407,  -21407,   16819,  -29692,
 -21407,   21407,    5906,   25172,
  11585,  -27969,    5906,  -16819,
  27969,  -11585,   25172,  -29692,

  19266,   19266,   26722,   22654,
  19266,   19266,   15137,    5315,
  25172,   10426,   22654,   -5315,
 -10426,  -25172,  -26722,  -15137,
  19266,  -19266,   15137,  -26722,
 -19266,   19266,    5315,   22654,
  10426,  -25172,    5315,  -15137,
  25172,  -10426,   22654,  -26722,

  16384,   16384,   22725,   19266,
  16384,   16384,   12873,    4520,
  21407,    8867,   19266,   -4520,
  -8867,  -21407,  -22725,  -12873,
  16384,  -16384,   12873,  -22725,
 -16384,   16384,    4520,   19266,
   8867,  -21407,    4520,  -12873,
  21407,   -8867,   19266,  -22725,

  19266,   19266,   26722,   22654,
  19266,   19266,   15137,    5315,
  25172,   10426,   22654,   -5315,
 -10426,  -25172,  -26722,  -15137,
  19266,  -19266,   15137,  -26722,
 -19266,   19266,    5315,   22654,
  10426,  -25172,    5315,  -15137,
  25172,  -10426,   22654,  -26722,

  21407,   21407,   29692,   25172,
  21407,   21407,   16819,    5906,
  27969,   11585,   25172,   -5906,
 -11585,  -27969,  -29692,  -16819,
  21407,  -21407,   16819,  -29692,
 -21407,   21407,    5906,   25172,
  11585,  -27969,    5906,  -16819,
  27969,  -11585,   25172,  -29692,

  22725,   22725,   31521,   26722,
  22725,   22725,   17855,    6270,
  29692,   12299,   26722,   -6270,
 -12299,  -29692,  -31521,  -17855,
  22725,  -22725,   17855,  -31521,
 -22725,   22725,    6270,   26722,
  12299,  -29692,    6270,  -17855,
  29692,  -12299,   26722,  -31521,
151 152
};

153
static struct
154 155 156 157
{
 const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
} tab_frw_01234567_sse2 ATTR_ALIGN(16) =
{{
158
//static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = {  // forward_dct coeff table
159 160 161
#define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
                   C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
                  -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
162 163
                   C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1,
// c1..c7 * cos(pi/4) * 2^15
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
#define C1 22725
#define C2 21407
#define C3 19266
#define C4 16384
#define C5 12873
#define C6 8867
#define C7 4520
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 31521
#define C2 29692
#define C3 26722
#define C4 22725
#define C5 17855
#define C6 12299
#define C7 6270
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 29692
#define C2 27969
#define C3 25172
#define C4 21407
#define C5 16819
#define C6 11585
#define C7 5906
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 26722
#define C2 25172
#define C3 22654
#define C4 19266
#define C5 15137
#define C6 10426
#define C7 5315
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 22725
#define C2 21407
#define C3 19266
#define C4 16384
#define C5 12873
#define C6 8867
#define C7 4520
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 26722
#define C2 25172
#define C3 22654
#define C4 19266
#define C5 15137
#define C6 10426
#define C7 5315
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 29692
#define C2 27969
#define C3 25172
#define C4 21407
#define C5 16819
#define C6 11585
#define C7 5906
TABLE_SSE2

#undef C1
#undef C2
#undef C3
#undef C4
#undef C5
#undef C6
#undef C7
#define C1 31521
#define C2 29692
#define C3 26722
#define C4 22725
#define C5 17855
#define C6 12299
#define C7 6270
TABLE_SSE2
284 285
}};

286

287
static av_always_inline void fdct_col(const int16_t *in, int16_t *out, int offset)
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365
{
    movq_m2r(*(in + offset + 1 * 8), mm0);
    movq_m2r(*(in + offset + 6 * 8), mm1);
    movq_r2r(mm0, mm2);
    movq_m2r(*(in + offset + 2 * 8), mm3);
    paddsw_r2r(mm1, mm0);
    movq_m2r(*(in + offset + 5 * 8), mm4);
    psllw_i2r(SHIFT_FRW_COL, mm0);
    movq_m2r(*(in + offset + 0 * 8), mm5);
    paddsw_r2r(mm3, mm4);
    paddsw_m2r(*(in + offset + 7 * 8), mm5);
    psllw_i2r(SHIFT_FRW_COL, mm4);
    movq_r2r(mm0, mm6);
    psubsw_r2r(mm1, mm2);
    movq_m2r(*(fdct_tg_all_16 + 4), mm1);
    psubsw_r2r(mm4, mm0);
    movq_m2r(*(in + offset + 3 * 8), mm7);
    pmulhw_r2r(mm0, mm1);
    paddsw_m2r(*(in + offset + 4 * 8), mm7);
    psllw_i2r(SHIFT_FRW_COL, mm5);
    paddsw_r2r(mm4, mm6);
    psllw_i2r(SHIFT_FRW_COL, mm7);
    movq_r2r(mm5, mm4);
    psubsw_r2r(mm7, mm5);
    paddsw_r2r(mm5, mm1);
    paddsw_r2r(mm7, mm4);
    por_m2r(fdct_one_corr, mm1);
    psllw_i2r(SHIFT_FRW_COL + 1, mm2);
    pmulhw_m2r(*(fdct_tg_all_16 + 4), mm5);
    movq_r2r(mm4, mm7);
    psubsw_m2r(*(in + offset + 5 * 8), mm3);
    psubsw_r2r(mm6, mm4);
    movq_r2m(mm1, *(out + offset + 2 * 8));
    paddsw_r2r(mm6, mm7);
    movq_m2r(*(in + offset + 3 * 8), mm1);
    psllw_i2r(SHIFT_FRW_COL + 1, mm3);
    psubsw_m2r(*(in + offset + 4 * 8), mm1);
    movq_r2r(mm2, mm6);
    movq_r2m(mm4, *(out + offset + 4 * 8));
    paddsw_r2r(mm3, mm2);
    pmulhw_m2r(*ocos_4_16, mm2);
    psubsw_r2r(mm3, mm6);
    pmulhw_m2r(*ocos_4_16, mm6);
    psubsw_r2r(mm0, mm5);
    por_m2r(fdct_one_corr, mm5);
    psllw_i2r(SHIFT_FRW_COL, mm1);
    por_m2r(fdct_one_corr, mm2);
    movq_r2r(mm1, mm4);
    movq_m2r(*(in + offset + 0 * 8), mm3);
    paddsw_r2r(mm6, mm1);
    psubsw_m2r(*(in + offset + 7 * 8), mm3);
    psubsw_r2r(mm6, mm4);
    movq_m2r(*(fdct_tg_all_16 + 0), mm0);
    psllw_i2r(SHIFT_FRW_COL, mm3);
    movq_m2r(*(fdct_tg_all_16 + 8), mm6);
    pmulhw_r2r(mm1, mm0);
    movq_r2m(mm7, *(out + offset + 0 * 8));
    pmulhw_r2r(mm4, mm6);
    movq_r2m(mm5, *(out + offset + 6 * 8));
    movq_r2r(mm3, mm7);
    movq_m2r(*(fdct_tg_all_16 + 8), mm5);
    psubsw_r2r(mm2, mm7);
    paddsw_r2r(mm2, mm3);
    pmulhw_r2r(mm7, mm5);
    paddsw_r2r(mm3, mm0);
    paddsw_r2r(mm4, mm6);
    pmulhw_m2r(*(fdct_tg_all_16 + 0), mm3);
    por_m2r(fdct_one_corr, mm0);
    paddsw_r2r(mm7, mm5);
    psubsw_r2r(mm6, mm7);
    movq_r2m(mm0, *(out + offset + 1 * 8));
    paddsw_r2r(mm4, mm5);
    movq_r2m(mm7, *(out + offset + 3 * 8));
    psubsw_r2r(mm1, mm3);
    movq_r2m(mm5, *(out + offset + 5 * 8));
    movq_r2m(mm3, *(out + offset + 7 * 8));
}

366

367
static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
368 369
{
    asm volatile(
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
#define FDCT_ROW_SSE2_H1(i,t)                    \
        "movq      " #i "(%0), %%xmm2      \n\t" \
        "movq      " #i "+8(%0), %%xmm0    \n\t" \
        "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
        "movdqa    " #t "+48(%1), %%xmm7   \n\t" \
        "movdqa    " #t "(%1), %%xmm4      \n\t" \
        "movdqa    " #t "+16(%1), %%xmm5   \n\t"

#define FDCT_ROW_SSE2_H2(i,t)                    \
        "movq      " #i "(%0), %%xmm2      \n\t" \
        "movq      " #i "+8(%0), %%xmm0    \n\t" \
        "movdqa    " #t "+32(%1), %%xmm3   \n\t" \
        "movdqa    " #t "+48(%1), %%xmm7   \n\t"

#define FDCT_ROW_SSE2(i)                      \
        "movq      %%xmm2, %%xmm1       \n\t" \
        "pshuflw   $27, %%xmm0, %%xmm0  \n\t" \
        "paddsw    %%xmm0, %%xmm1       \n\t" \
        "psubsw    %%xmm0, %%xmm2       \n\t" \
        "punpckldq %%xmm2, %%xmm1       \n\t" \
        "pshufd    $78, %%xmm1, %%xmm2  \n\t" \
        "pmaddwd   %%xmm2, %%xmm3       \n\t" \
        "pmaddwd   %%xmm1, %%xmm7       \n\t" \
        "pmaddwd   %%xmm5, %%xmm2       \n\t" \
        "pmaddwd   %%xmm4, %%xmm1       \n\t" \
        "paddd     %%xmm7, %%xmm3       \n\t" \
        "paddd     %%xmm2, %%xmm1       \n\t" \
        "paddd     %%xmm6, %%xmm3       \n\t" \
        "paddd     %%xmm6, %%xmm1       \n\t" \
        "psrad     %3, %%xmm3           \n\t" \
        "psrad     %3, %%xmm1           \n\t" \
        "packssdw  %%xmm3, %%xmm1       \n\t" \
        "movdqa    %%xmm1, " #i "(%4)   \n\t"

404
        "movdqa    (%2), %%xmm6         \n\t"
405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423
        FDCT_ROW_SSE2_H1(0,0)
        FDCT_ROW_SSE2(0)
        FDCT_ROW_SSE2_H2(64,0)
        FDCT_ROW_SSE2(64)

        FDCT_ROW_SSE2_H1(16,64)
        FDCT_ROW_SSE2(16)
        FDCT_ROW_SSE2_H2(112,64)
        FDCT_ROW_SSE2(112)

        FDCT_ROW_SSE2_H1(32,128)
        FDCT_ROW_SSE2(32)
        FDCT_ROW_SSE2_H2(96,128)
        FDCT_ROW_SSE2(96)

        FDCT_ROW_SSE2_H1(48,192)
        FDCT_ROW_SSE2(48)
        FDCT_ROW_SSE2_H2(80,192)
        FDCT_ROW_SSE2(80)
424 425
        :
        : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out)
426 427 428
    );
}

429
static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
430
{
431 432
    pshufw_m2r(*(in + 4), mm5, 0x1B);
    movq_m2r(*(in + 0), mm0);
433
    movq_r2r(mm0, mm1);
Michael Niedermayer's avatar
Michael Niedermayer committed
434 435
    paddsw_r2r(mm5, mm0);
    psubsw_r2r(mm5, mm1);
436 437 438 439 440 441 442 443
    movq_r2r(mm0, mm2);
    punpckldq_r2r(mm1, mm0);
    punpckhdq_r2r(mm1, mm2);
    movq_m2r(*(table + 0), mm1);
    movq_m2r(*(table + 4), mm3);
    movq_m2r(*(table + 8), mm4);
    movq_m2r(*(table + 12), mm5);
    movq_m2r(*(table + 16), mm6);
Michael Niedermayer's avatar
Michael Niedermayer committed
444
    movq_m2r(*(table + 20), mm7);
445 446
    pmaddwd_r2r(mm0, mm1);
    pmaddwd_r2r(mm2, mm3);
447
    pmaddwd_r2r(mm0, mm4);
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
    pmaddwd_r2r(mm2, mm5);
    pmaddwd_r2r(mm0, mm6);
    pmaddwd_r2r(mm2, mm7);
    pmaddwd_m2r(*(table + 24), mm0);
    pmaddwd_m2r(*(table + 28), mm2);
    paddd_r2r(mm1, mm3);
    paddd_r2r(mm4, mm5);
    paddd_r2r(mm6, mm7);
    paddd_r2r(mm0, mm2);
    movq_m2r(*fdct_r_row, mm0);
    paddd_r2r(mm0, mm3);
    paddd_r2r(mm0, mm5);
    paddd_r2r(mm0, mm7);
    paddd_r2r(mm0, mm2);
    psrad_i2r(SHIFT_FRW_ROW, mm3);
Michael Niedermayer's avatar
Michael Niedermayer committed
463
    psrad_i2r(SHIFT_FRW_ROW, mm5);
464 465 466 467 468 469
    psrad_i2r(SHIFT_FRW_ROW, mm7);
    psrad_i2r(SHIFT_FRW_ROW, mm2);
    packssdw_r2r(mm5, mm3);
    packssdw_r2r(mm2, mm7);
    movq_r2m(mm3, *(out + 0));
    movq_r2m(mm7, *(out + 4));
Michael Niedermayer's avatar
Michael Niedermayer committed
470 471
}

472
static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
473
{
474
//FIXME reorder (i dont have a old mmx only cpu here to benchmark ...)
475 476 477 478
    movd_m2r(*(in + 6), mm1);
    punpcklwd_m2r(*(in + 4), mm1);
    movq_r2r(mm1, mm2);
    psrlq_i2r(0x20, mm1);
479
    movq_m2r(*(in + 0), mm0);
480
    punpcklwd_r2r(mm2, mm1);
481
    movq_r2r(mm0, mm5);
482 483
    paddsw_r2r(mm1, mm0);
    psubsw_r2r(mm1, mm5);
484 485 486 487 488 489 490 491
    movq_r2r(mm0, mm2);
    punpckldq_r2r(mm5, mm0);
    punpckhdq_r2r(mm5, mm2);
    movq_m2r(*(table + 0), mm1);
    movq_m2r(*(table + 4), mm3);
    movq_m2r(*(table + 8), mm4);
    movq_m2r(*(table + 12), mm5);
    movq_m2r(*(table + 16), mm6);
492
    movq_m2r(*(table + 20), mm7);
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509
    pmaddwd_r2r(mm0, mm1);
    pmaddwd_r2r(mm2, mm3);
    pmaddwd_r2r(mm0, mm4);
    pmaddwd_r2r(mm2, mm5);
    pmaddwd_r2r(mm0, mm6);
    pmaddwd_r2r(mm2, mm7);
    pmaddwd_m2r(*(table + 24), mm0);
    pmaddwd_m2r(*(table + 28), mm2);
    paddd_r2r(mm1, mm3);
    paddd_r2r(mm4, mm5);
    paddd_r2r(mm6, mm7);
    paddd_r2r(mm0, mm2);
    movq_m2r(*fdct_r_row, mm0);
    paddd_r2r(mm0, mm3);
    paddd_r2r(mm0, mm5);
    paddd_r2r(mm0, mm7);
    paddd_r2r(mm0, mm2);
510 511
    psrad_i2r(SHIFT_FRW_ROW, mm3);
    psrad_i2r(SHIFT_FRW_ROW, mm5);
512 513 514 515
    psrad_i2r(SHIFT_FRW_ROW, mm7);
    psrad_i2r(SHIFT_FRW_ROW, mm2);
    packssdw_r2r(mm5, mm3);
    packssdw_r2r(mm2, mm7);
516
    movq_r2m(mm3, *(out + 0));
517
    movq_r2m(mm7, *(out + 4));
518 519
}

520
void ff_fdct_mmx(int16_t *block)
521
{
Michael Niedermayer's avatar
Michael Niedermayer committed
522
    int64_t align_tmp[16] ATTR_ALIGN(8);
Michael Niedermayer's avatar
Michael Niedermayer committed
523 524
    int16_t * block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
525 526 527 528 529 530
    int i;

    fdct_col(block, block1, 0);
    fdct_col(block, block1, 4);

    for(i=8;i>0;i--) {
Michael Niedermayer's avatar
Michael Niedermayer committed
531
        fdct_row_mmx(block1, block, table);
532 533
        block1 += 8;
        table += 32;
Michael Niedermayer's avatar
Michael Niedermayer committed
534
        block += 8;
535 536 537 538 539 540
    }
}

void ff_fdct_mmx2(int16_t *block)
{
    int64_t align_tmp[16] ATTR_ALIGN(8);
Michael Niedermayer's avatar
Michael Niedermayer committed
541 542
    int16_t *block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
543 544 545 546 547 548
    int i;

    fdct_col(block, block1, 0);
    fdct_col(block, block1, 4);

    for(i=8;i>0;i--) {
Michael Niedermayer's avatar
Michael Niedermayer committed
549
        fdct_row_mmx2(block1, block, table);
550 551
        block1 += 8;
        table += 32;
Michael Niedermayer's avatar
Michael Niedermayer committed
552
        block += 8;
553 554
    }
}
555

556
void ff_fdct_sse2(int16_t *block)
557
{
558
    int64_t align_tmp[16] ATTR_ALIGN(16);
Michael Niedermayer's avatar
Michael Niedermayer committed
559
    int16_t * const block1= (int16_t*)align_tmp;
560 561 562 563 564 565 566

    fdct_col(block, block1, 0);
    fdct_col(block, block1, 4);

    fdct_row_sse2(block1, block);
}