mpegvideo_mmx.c 27.5 KB
Newer Older
Nick Kurshev's avatar
Nick Kurshev committed
1 2
/*
 * The simplest mpeg encoder (well, it was the simplest!)
3
 * Copyright (c) 2000,2001 Fabrice Bellard.
Nick Kurshev's avatar
Nick Kurshev committed
4
 *
5 6 7
 * Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
 * h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
 *
8 9 10
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
Nick Kurshev's avatar
Nick Kurshev committed
14
 *
15
 * FFmpeg is distributed in the hope that it will be useful,
Nick Kurshev's avatar
Nick Kurshev committed
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 18
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
Nick Kurshev's avatar
Nick Kurshev committed
19
 *
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with FFmpeg; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Nick Kurshev's avatar
Nick Kurshev committed
23 24
 */

25 26 27 28
#include "libavutil/x86_cpu.h"
#include "libavcodec/avcodec.h"
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
29
#include "dsputil_mmx.h"
30

Måns Rullgård's avatar
Måns Rullgård committed
31
extern uint16_t inv_zigzag_direct16[64];
32

33

34
static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
35 36
                                  DCTELEM *block, int n, int qscale)
{
37
    x86_reg level, qmul, qadd, nCoeffs;
38 39

    qmul = qscale << 1;
40

41
    assert(s->block_last_index[n]>=0 || s->h263_aic);
42

43 44 45 46 47 48 49 50 51
    if (!s->h263_aic) {
        if (n < 4)
            level = block[0] * s->y_dc_scale;
        else
            level = block[0] * s->c_dc_scale;
        qadd = (qscale - 1) | 1;
    }else{
        qadd = 0;
        level= block[0];
52
    }
53 54 55 56 57
    if(s->ac_pred)
        nCoeffs=63;
    else
        nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
//printf("%d %d  ", qmul, qadd);
58
__asm__ volatile(
59 60 61 62 63 64 65 66 67
                "movd %1, %%mm6                 \n\t" //qmul
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "movd %2, %%mm5                 \n\t" //qadd
                "pxor %%mm7, %%mm7              \n\t"
                "packssdw %%mm5, %%mm5          \n\t"
                "packssdw %%mm5, %%mm5          \n\t"
                "psubw %%mm5, %%mm7             \n\t"
                "pxor %%mm4, %%mm4              \n\t"
68
                ASMALIGN(4)
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 99 100 101
                "1:                             \n\t"
                "movq (%0, %3), %%mm0           \n\t"
                "movq 8(%0, %3), %%mm1          \n\t"

                "pmullw %%mm6, %%mm0            \n\t"
                "pmullw %%mm6, %%mm1            \n\t"

                "movq (%0, %3), %%mm2           \n\t"
                "movq 8(%0, %3), %%mm3          \n\t"

                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0

                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"

                "paddw %%mm7, %%mm0             \n\t"
                "paddw %%mm7, %%mm1             \n\t"

                "pxor %%mm0, %%mm2              \n\t"
                "pxor %%mm1, %%mm3              \n\t"

                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0

                "pandn %%mm2, %%mm0             \n\t"
                "pandn %%mm3, %%mm1             \n\t"

                "movq %%mm0, (%0, %3)           \n\t"
                "movq %%mm1, 8(%0, %3)          \n\t"

                "add $16, %3                    \n\t"
                "jng 1b                         \n\t"
102
                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
103 104
                : "memory"
        );
105 106 107 108 109 110 111
        block[0]= level;
}


static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
                                  DCTELEM *block, int n, int qscale)
{
112
    x86_reg qmul, qadd, nCoeffs;
113 114 115 116 117

    qmul = qscale << 1;
    qadd = (qscale - 1) | 1;

    assert(s->block_last_index[n]>=0 || s->h263_aic);
118

119
    nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
120
//printf("%d %d  ", qmul, qadd);
121
__asm__ volatile(
122 123 124 125 126 127 128 129 130
                "movd %1, %%mm6                 \n\t" //qmul
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "movd %2, %%mm5                 \n\t" //qadd
                "pxor %%mm7, %%mm7              \n\t"
                "packssdw %%mm5, %%mm5          \n\t"
                "packssdw %%mm5, %%mm5          \n\t"
                "psubw %%mm5, %%mm7             \n\t"
                "pxor %%mm4, %%mm4              \n\t"
131
                ASMALIGN(4)
132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
                "1:                             \n\t"
                "movq (%0, %3), %%mm0           \n\t"
                "movq 8(%0, %3), %%mm1          \n\t"

                "pmullw %%mm6, %%mm0            \n\t"
                "pmullw %%mm6, %%mm1            \n\t"

                "movq (%0, %3), %%mm2           \n\t"
                "movq 8(%0, %3), %%mm3          \n\t"

                "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0

                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"

                "paddw %%mm7, %%mm0             \n\t"
                "paddw %%mm7, %%mm1             \n\t"

                "pxor %%mm0, %%mm2              \n\t"
                "pxor %%mm1, %%mm3              \n\t"

                "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0

                "pandn %%mm2, %%mm0             \n\t"
                "pandn %%mm3, %%mm1             \n\t"

                "movq %%mm0, (%0, %3)           \n\t"
                "movq %%mm1, 8(%0, %3)          \n\t"

                "add $16, %3                    \n\t"
                "jng 1b                         \n\t"
165
                ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
166 167
                : "memory"
        );
168 169 170
}


Nick Kurshev's avatar
Nick Kurshev committed
171 172 173 174 175 176 177 178 179
/*
  NK:
  Note: looking at PARANOID:
  "enable all paranoid tests for rounding, overflows, etc..."

#ifdef PARANOID
                if (level < -2048 || level > 2047)
                    fprintf(stderr, "unquant error %d %d\n", i, level);
#endif
Vitor Sessak's avatar
Vitor Sessak committed
180
  We can suppose that result of two multiplications can't be greater than 0xFFFF
Nick Kurshev's avatar
Nick Kurshev committed
181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
  i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
  a complex multiplication.
=====================================================
 Full formula for multiplication of 2 integer numbers
 which are represent as high:low words:
 input: value1 = high1:low1
        value2 = high2:low2
 output: value3 = value1*value2
 value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
 this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
 but this algorithm will compute only 0x66cb0ce4
 this limited by 16-bit size of operands
 ---------------------------------
 tlow1 = high1*low2
 tlow2 = high2*low1
 tlow1 = tlow1 + tlow2
 high3:low3 = low1*low2
 high3 += tlow1
*/
200
static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
201
                                     DCTELEM *block, int n, int qscale)
Nick Kurshev's avatar
Nick Kurshev committed
202
{
203
    x86_reg nCoeffs;
204
    const uint16_t *quant_matrix;
205
    int block0;
206 207 208 209

    assert(s->block_last_index[n]>=0);

    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
210

211
    if (n < 4)
212 213 214 215 216
        block0 = block[0] * s->y_dc_scale;
    else
        block0 = block[0] * s->c_dc_scale;
    /* XXX: only mpeg1 */
    quant_matrix = s->intra_matrix;
217
__asm__ volatile(
218 219 220 221 222 223
                "pcmpeqw %%mm7, %%mm7           \n\t"
                "psrlw $15, %%mm7               \n\t"
                "movd %2, %%mm6                 \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "mov %3, %%"REG_a"              \n\t"
224
                ASMALIGN(4)
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
                "1:                             \n\t"
                "movq (%0, %%"REG_a"), %%mm0    \n\t"
                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
                "movq (%1, %%"REG_a"), %%mm4    \n\t"
                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
                "pxor %%mm2, %%mm2              \n\t"
                "pxor %%mm3, %%mm3              \n\t"
                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
                "pxor %%mm4, %%mm4              \n\t"
                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
                "psraw $3, %%mm0                \n\t"
                "psraw $3, %%mm1                \n\t"
                "psubw %%mm7, %%mm0             \n\t"
                "psubw %%mm7, %%mm1             \n\t"
                "por %%mm7, %%mm0               \n\t"
                "por %%mm7, %%mm1               \n\t"
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t"
                "psubw %%mm3, %%mm1             \n\t"
                "pandn %%mm0, %%mm4             \n\t"
                "pandn %%mm1, %%mm5             \n\t"
                "movq %%mm4, (%0, %%"REG_a")    \n\t"
                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"

                "add $16, %%"REG_a"             \n\t"
                "js 1b                          \n\t"
263
                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
264 265
                : "%"REG_a, "memory"
        );
266 267 268 269 270 271
    block[0]= block0;
}

static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
                                     DCTELEM *block, int n, int qscale)
{
272
    x86_reg nCoeffs;
273 274 275 276 277
    const uint16_t *quant_matrix;

    assert(s->block_last_index[n]>=0);

    nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
278

279
        quant_matrix = s->inter_matrix;
280
__asm__ volatile(
281 282 283 284 285 286
                "pcmpeqw %%mm7, %%mm7           \n\t"
                "psrlw $15, %%mm7               \n\t"
                "movd %2, %%mm6                 \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "mov %3, %%"REG_a"              \n\t"
287
                ASMALIGN(4)
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
                "1:                             \n\t"
                "movq (%0, %%"REG_a"), %%mm0    \n\t"
                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
                "movq (%1, %%"REG_a"), %%mm4    \n\t"
                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
                "pxor %%mm2, %%mm2              \n\t"
                "pxor %%mm3, %%mm3              \n\t"
                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
                "paddw %%mm7, %%mm0             \n\t" // abs(block[i])*2 + 1
                "paddw %%mm7, %%mm1             \n\t" // abs(block[i])*2 + 1
                "pmullw %%mm4, %%mm0            \n\t" // (abs(block[i])*2 + 1)*q
                "pmullw %%mm5, %%mm1            \n\t" // (abs(block[i])*2 + 1)*q
                "pxor %%mm4, %%mm4              \n\t"
                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
                "psraw $4, %%mm0                \n\t"
                "psraw $4, %%mm1                \n\t"
                "psubw %%mm7, %%mm0             \n\t"
                "psubw %%mm7, %%mm1             \n\t"
                "por %%mm7, %%mm0               \n\t"
                "por %%mm7, %%mm1               \n\t"
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t"
                "psubw %%mm3, %%mm1             \n\t"
                "pandn %%mm0, %%mm4             \n\t"
                "pandn %%mm1, %%mm5             \n\t"
                "movq %%mm4, (%0, %%"REG_a")    \n\t"
                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"

                "add $16, %%"REG_a"             \n\t"
                "js 1b                          \n\t"
330
                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
331 332
                : "%"REG_a, "memory"
        );
333 334
}

335
static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
336 337
                                     DCTELEM *block, int n, int qscale)
{
338
    x86_reg nCoeffs;
339
    const uint16_t *quant_matrix;
340
    int block0;
341

342 343 344 345
    assert(s->block_last_index[n]>=0);

    if(s->alternate_scan) nCoeffs= 63; //FIXME
    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
346

347
    if (n < 4)
348 349 350 351
        block0 = block[0] * s->y_dc_scale;
    else
        block0 = block[0] * s->c_dc_scale;
    quant_matrix = s->intra_matrix;
352
__asm__ volatile(
353 354 355 356 357 358
                "pcmpeqw %%mm7, %%mm7           \n\t"
                "psrlw $15, %%mm7               \n\t"
                "movd %2, %%mm6                 \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "mov %3, %%"REG_a"              \n\t"
359
                ASMALIGN(4)
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
                "1:                             \n\t"
                "movq (%0, %%"REG_a"), %%mm0    \n\t"
                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
                "movq (%1, %%"REG_a"), %%mm4    \n\t"
                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
                "pxor %%mm2, %%mm2              \n\t"
                "pxor %%mm3, %%mm3              \n\t"
                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
                "pxor %%mm4, %%mm4              \n\t"
                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
                "psraw $3, %%mm0                \n\t"
                "psraw $3, %%mm1                \n\t"
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t"
                "psubw %%mm3, %%mm1             \n\t"
                "pandn %%mm0, %%mm4             \n\t"
                "pandn %%mm1, %%mm5             \n\t"
                "movq %%mm4, (%0, %%"REG_a")    \n\t"
                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"

                "add $16, %%"REG_a"             \n\t"
                "jng 1b                         \n\t"
394
                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
395 396
                : "%"REG_a, "memory"
        );
397
    block[0]= block0;
Diego Biurrun's avatar
Diego Biurrun committed
398
        //Note, we do not do mismatch control for intra as errors cannot accumulate
399 400 401 402 403
}

static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
                                     DCTELEM *block, int n, int qscale)
{
404
    x86_reg nCoeffs;
405
    const uint16_t *quant_matrix;
406

407 408 409 410
    assert(s->block_last_index[n]>=0);

    if(s->alternate_scan) nCoeffs= 63; //FIXME
    else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
411

412
        quant_matrix = s->inter_matrix;
413
__asm__ volatile(
414 415 416 417 418 419
                "pcmpeqw %%mm7, %%mm7           \n\t"
                "psrlq $48, %%mm7               \n\t"
                "movd %2, %%mm6                 \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "packssdw %%mm6, %%mm6          \n\t"
                "mov %3, %%"REG_a"              \n\t"
420
                ASMALIGN(4)
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472
                "1:                             \n\t"
                "movq (%0, %%"REG_a"), %%mm0    \n\t"
                "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
                "movq (%1, %%"REG_a"), %%mm4    \n\t"
                "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
                "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
                "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
                "pxor %%mm2, %%mm2              \n\t"
                "pxor %%mm3, %%mm3              \n\t"
                "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
                "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
                "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
                "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
                "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
                "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*2*q
                "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*2*q
                "paddw %%mm4, %%mm0             \n\t" // (abs(block[i])*2 + 1)*q
                "paddw %%mm5, %%mm1             \n\t" // (abs(block[i])*2 + 1)*q
                "pxor %%mm4, %%mm4              \n\t"
                "pxor %%mm5, %%mm5              \n\t" // FIXME slow
                "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
                "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
                "psrlw $4, %%mm0                \n\t"
                "psrlw $4, %%mm1                \n\t"
                "pxor %%mm2, %%mm0              \n\t"
                "pxor %%mm3, %%mm1              \n\t"
                "psubw %%mm2, %%mm0             \n\t"
                "psubw %%mm3, %%mm1             \n\t"
                "pandn %%mm0, %%mm4             \n\t"
                "pandn %%mm1, %%mm5             \n\t"
                "pxor %%mm4, %%mm7              \n\t"
                "pxor %%mm5, %%mm7              \n\t"
                "movq %%mm4, (%0, %%"REG_a")    \n\t"
                "movq %%mm5, 8(%0, %%"REG_a")   \n\t"

                "add $16, %%"REG_a"             \n\t"
                "jng 1b                         \n\t"
                "movd 124(%0, %3), %%mm0        \n\t"
                "movq %%mm7, %%mm6              \n\t"
                "psrlq $32, %%mm7               \n\t"
                "pxor %%mm6, %%mm7              \n\t"
                "movq %%mm7, %%mm6              \n\t"
                "psrlq $16, %%mm7               \n\t"
                "pxor %%mm6, %%mm7              \n\t"
                "pslld $31, %%mm7               \n\t"
                "psrlq $15, %%mm7               \n\t"
                "pxor %%mm7, %%mm0              \n\t"
                "movd %%mm0, 124(%0, %3)        \n\t"

473
                ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs)
474 475
                : "%"REG_a, "memory"
        );
Nick Kurshev's avatar
Nick Kurshev committed
476 477
}

Michael Niedermayer's avatar
Michael Niedermayer committed
478 479 480 481 482 483 484
static void  denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
    const int intra= s->mb_intra;
    int *sum= s->dct_error_sum[intra];
    uint16_t *offset= s->dct_offset[intra];

    s->dct_count[intra]++;

485
    __asm__ volatile(
486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
        "pxor %%mm7, %%mm7                      \n\t"
        "1:                                     \n\t"
        "pxor %%mm0, %%mm0                      \n\t"
        "pxor %%mm1, %%mm1                      \n\t"
        "movq (%0), %%mm2                       \n\t"
        "movq 8(%0), %%mm3                      \n\t"
        "pcmpgtw %%mm2, %%mm0                   \n\t"
        "pcmpgtw %%mm3, %%mm1                   \n\t"
        "pxor %%mm0, %%mm2                      \n\t"
        "pxor %%mm1, %%mm3                      \n\t"
        "psubw %%mm0, %%mm2                     \n\t"
        "psubw %%mm1, %%mm3                     \n\t"
        "movq %%mm2, %%mm4                      \n\t"
        "movq %%mm3, %%mm5                      \n\t"
        "psubusw (%2), %%mm2                    \n\t"
        "psubusw 8(%2), %%mm3                   \n\t"
        "pxor %%mm0, %%mm2                      \n\t"
        "pxor %%mm1, %%mm3                      \n\t"
        "psubw %%mm0, %%mm2                     \n\t"
        "psubw %%mm1, %%mm3                     \n\t"
        "movq %%mm2, (%0)                       \n\t"
        "movq %%mm3, 8(%0)                      \n\t"
        "movq %%mm4, %%mm2                      \n\t"
        "movq %%mm5, %%mm3                      \n\t"
        "punpcklwd %%mm7, %%mm4                 \n\t"
        "punpckhwd %%mm7, %%mm2                 \n\t"
        "punpcklwd %%mm7, %%mm5                 \n\t"
        "punpckhwd %%mm7, %%mm3                 \n\t"
        "paddd (%1), %%mm4                      \n\t"
        "paddd 8(%1), %%mm2                     \n\t"
        "paddd 16(%1), %%mm5                    \n\t"
        "paddd 24(%1), %%mm3                    \n\t"
        "movq %%mm4, (%1)                       \n\t"
        "movq %%mm2, 8(%1)                      \n\t"
        "movq %%mm5, 16(%1)                     \n\t"
        "movq %%mm3, 24(%1)                     \n\t"
        "add $16, %0                            \n\t"
        "add $32, %1                            \n\t"
        "add $16, %2                            \n\t"
        "cmp %3, %0                             \n\t"
            " jb 1b                             \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
527 528 529 530 531
        : "+r" (block), "+r" (sum), "+r" (offset)
        : "r"(block+64)
    );
}

532 533 534 535 536 537 538
static void  denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
    const int intra= s->mb_intra;
    int *sum= s->dct_error_sum[intra];
    uint16_t *offset= s->dct_offset[intra];

    s->dct_count[intra]++;

539
    __asm__ volatile(
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
        "pxor %%xmm7, %%xmm7                    \n\t"
        "1:                                     \n\t"
        "pxor %%xmm0, %%xmm0                    \n\t"
        "pxor %%xmm1, %%xmm1                    \n\t"
        "movdqa (%0), %%xmm2                    \n\t"
        "movdqa 16(%0), %%xmm3                  \n\t"
        "pcmpgtw %%xmm2, %%xmm0                 \n\t"
        "pcmpgtw %%xmm3, %%xmm1                 \n\t"
        "pxor %%xmm0, %%xmm2                    \n\t"
        "pxor %%xmm1, %%xmm3                    \n\t"
        "psubw %%xmm0, %%xmm2                   \n\t"
        "psubw %%xmm1, %%xmm3                   \n\t"
        "movdqa %%xmm2, %%xmm4                  \n\t"
        "movdqa %%xmm3, %%xmm5                  \n\t"
        "psubusw (%2), %%xmm2                   \n\t"
        "psubusw 16(%2), %%xmm3                 \n\t"
        "pxor %%xmm0, %%xmm2                    \n\t"
        "pxor %%xmm1, %%xmm3                    \n\t"
        "psubw %%xmm0, %%xmm2                   \n\t"
        "psubw %%xmm1, %%xmm3                   \n\t"
        "movdqa %%xmm2, (%0)                    \n\t"
        "movdqa %%xmm3, 16(%0)                  \n\t"
        "movdqa %%xmm4, %%xmm6                  \n\t"
        "movdqa %%xmm5, %%xmm0                  \n\t"
        "punpcklwd %%xmm7, %%xmm4               \n\t"
        "punpckhwd %%xmm7, %%xmm6               \n\t"
        "punpcklwd %%xmm7, %%xmm5               \n\t"
        "punpckhwd %%xmm7, %%xmm0               \n\t"
        "paddd (%1), %%xmm4                     \n\t"
        "paddd 16(%1), %%xmm6                   \n\t"
        "paddd 32(%1), %%xmm5                   \n\t"
        "paddd 48(%1), %%xmm0                   \n\t"
        "movdqa %%xmm4, (%1)                    \n\t"
        "movdqa %%xmm6, 16(%1)                  \n\t"
        "movdqa %%xmm5, 32(%1)                  \n\t"
        "movdqa %%xmm0, 48(%1)                  \n\t"
        "add $32, %0                            \n\t"
        "add $64, %1                            \n\t"
        "add $32, %2                            \n\t"
        "cmp %3, %0                             \n\t"
            " jb 1b                             \n\t"
581 582 583 584 585
        : "+r" (block), "+r" (sum), "+r" (offset)
        : "r"(block+64)
    );
}

586 587 588 589 590 591
#ifdef HAVE_SSSE3
#define HAVE_SSSE3_BAK
#endif
#undef HAVE_SSSE3

#undef HAVE_SSE2
592 593
#undef HAVE_MMX2
#define RENAME(a) a ## _MMX
594
#define RENAMEl(a) a ## _mmx
595 596 597 598
#include "mpegvideo_mmx_template.c"

#define HAVE_MMX2
#undef RENAME
599
#undef RENAMEl
600
#define RENAME(a) a ## _MMX2
601
#define RENAMEl(a) a ## _mmx2
602
#include "mpegvideo_mmx_template.c"
603

604
#define HAVE_SSE2
605 606 607 608 609 610
#undef RENAME
#undef RENAMEl
#define RENAME(a) a ## _SSE2
#define RENAMEl(a) a ## _sse2
#include "mpegvideo_mmx_template.c"

611 612 613 614 615 616 617 618 619
#ifdef HAVE_SSSE3_BAK
#define HAVE_SSSE3
#undef RENAME
#undef RENAMEl
#define RENAME(a) a ## _SSSE3
#define RENAMEl(a) a ## _sse2
#include "mpegvideo_mmx_template.c"
#endif

620
void MPV_common_init_mmx(MpegEncContext *s)
Nick Kurshev's avatar
Nick Kurshev committed
621
{
622
    if (mm_flags & FF_MM_MMX) {
623
        const int dct_algo = s->avctx->dct_algo;
624

625 626 627 628
        s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
        s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
        s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
        s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
629 630
        if(!(s->flags & CODEC_FLAG_BITEXACT))
            s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
631
        s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
632

633
        if (mm_flags & FF_MM_SSE2) {
634 635 636 637
            s->denoise_dct= denoise_dct_sse2;
        } else {
                s->denoise_dct= denoise_dct_mmx;
        }
638

639
        if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
640
#ifdef HAVE_SSSE3
641
            if(mm_flags & FF_MM_SSSE3){
642 643 644
                s->dct_quantize= dct_quantize_SSSE3;
            } else
#endif
645
            if(mm_flags & FF_MM_SSE2){
646
                s->dct_quantize= dct_quantize_SSE2;
647
            } else if(mm_flags & FF_MM_MMXEXT){
648 649 650 651
                s->dct_quantize= dct_quantize_MMX2;
            } else {
                s->dct_quantize= dct_quantize_MMX;
            }
652
        }
653
    }
Nick Kurshev's avatar
Nick Kurshev committed
654
}