dsputil_mlib.c 11.2 KB
Newer Older
Fabrice Bellard's avatar
Fabrice Bellard committed
1 2
/*
 * Sun mediaLib optimized DSP utils
3
 * Copyright (c) 2001 Fabrice Bellard
Fabrice Bellard's avatar
Fabrice Bellard committed
4
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
Fabrice Bellard's avatar
Fabrice Bellard committed
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
Fabrice Bellard's avatar
Fabrice Bellard committed
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
Fabrice Bellard's avatar
Fabrice Bellard committed
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Fabrice Bellard's avatar
Fabrice Bellard committed
20 21
 */

22 23
#include "libavcodec/dsputil.h"
#include "libavcodec/mpegvideo.h"
Fabrice Bellard's avatar
Fabrice Bellard committed
24 25 26 27

#include <mlib_types.h>
#include <mlib_status.h>
#include <mlib_sys.h>
28
#include <mlib_algebra.h>
Fabrice Bellard's avatar
Fabrice Bellard committed
29 30
#include <mlib_video.h>

31
/* misc */
Fabrice Bellard's avatar
Fabrice Bellard committed
32

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
static void get_pixels_mlib(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
{
  int i;

  for (i=0;i<8;i++) {
    mlib_VectorConvert_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)pixels, 8);

    pixels += line_size;
    block += 8;
  }
}

static void diff_pixels_mlib(DCTELEM *restrict block, const uint8_t *s1, const uint8_t *s2, int line_size)
{
  int i;

  for (i=0;i<8;i++) {
    mlib_VectorSub_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)s1, (mlib_u8 *)s2, 8);

    s1 += line_size;
    s2 += line_size;
    block += 8;
  }
}

static void add_pixels_clamped_mlib(const DCTELEM *block, uint8_t *pixels, int line_size)
{
    mlib_VideoAddBlock_U8_S16(pixels, (mlib_s16 *)block, line_size);
}

/* put block, width 16 pixel, height 8/16 */
64 65

static void put_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
66
                               int stride, int height)
67
{
68 69 70 71 72 73 74 75 76 77 78 79
  switch (height) {
    case 8:
      mlib_VideoCopyRef_U8_U8_16x8(dest, (uint8_t *)ref, stride);
    break;

    case 16:
      mlib_VideoCopyRef_U8_U8_16x16(dest, (uint8_t *)ref, stride);
    break;

    default:
      assert(0);
  }
80 81 82
}

static void put_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
83
                                  int stride, int height)
84
{
85 86 87 88 89 90 91 92 93 94 95 96
  switch (height) {
    case 8:
      mlib_VideoInterpX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
97 98 99
}

static void put_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
100
                                  int stride, int height)
101
{
102 103 104 105 106 107 108 109 110 111 112 113
  switch (height) {
    case 8:
      mlib_VideoInterpY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
114 115 116
}

static void put_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
117
                                  int stride, int height)
118
{
119 120 121 122 123 124 125 126 127 128 129 130
  switch (height) {
    case 8:
      mlib_VideoInterpXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
131 132
}

133
/* put block, width 8 pixel, height 4/8/16 */
134 135

static void put_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
136
                               int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
137
{
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
  switch (height) {
    case 4:
      mlib_VideoCopyRef_U8_U8_8x4(dest, (uint8_t *)ref, stride);
    break;

    case 8:
      mlib_VideoCopyRef_U8_U8_8x8(dest, (uint8_t *)ref, stride);
    break;

    case 16:
      mlib_VideoCopyRef_U8_U8_8x16(dest, (uint8_t *)ref, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
154 155
}

156
static void put_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
157
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
158
{
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174
  switch (height) {
    case 4:
      mlib_VideoInterpX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
175 176
}

177
static void put_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
178
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
179
{
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
  switch (height) {
    case 4:
      mlib_VideoInterpY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
196 197
}

198
static void put_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
199
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
200
{
201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
  switch (height) {
    case 4:
      mlib_VideoInterpXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
217 218
}

219
/* average block, width 16 pixel, height 8/16 */
220 221

static void avg_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
222
                               int stride, int height)
223
{
224 225 226 227 228 229 230 231 232 233 234 235
  switch (height) {
    case 8:
      mlib_VideoCopyRefAve_U8_U8_16x8(dest, (uint8_t *)ref, stride);
    break;

    case 16:
      mlib_VideoCopyRefAve_U8_U8_16x16(dest, (uint8_t *)ref, stride);
    break;

    default:
      assert(0);
  }
236 237 238
}

static void avg_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
239
                                  int stride, int height)
240
{
241 242 243 244 245 246 247 248 249 250 251 252
  switch (height) {
    case 8:
      mlib_VideoInterpAveX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
253 254 255
}

static void avg_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
256
                                  int stride, int height)
257
{
258 259 260 261 262 263 264 265 266 267 268 269
  switch (height) {
    case 8:
      mlib_VideoInterpAveY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
270 271
}

272
static void avg_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
273
                                  int stride, int height)
274
{
275 276 277 278 279 280 281 282 283 284 285 286
  switch (height) {
    case 8:
      mlib_VideoInterpAveXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
287 288
}

289
/* average block, width 8 pixel, height 4/8/16 */
290 291

static void avg_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
292
                               int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
293
{
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
  switch (height) {
    case 4:
      mlib_VideoCopyRefAve_U8_U8_8x4(dest, (uint8_t *)ref, stride);
    break;

    case 8:
      mlib_VideoCopyRefAve_U8_U8_8x8(dest, (uint8_t *)ref, stride);
    break;

    case 16:
      mlib_VideoCopyRefAve_U8_U8_8x16(dest, (uint8_t *)ref, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
310 311
}

312
static void avg_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
313
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
314
{
315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
  switch (height) {
    case 4:
      mlib_VideoInterpAveX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpAveX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
331 332
}

333
static void avg_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
334
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
335
{
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
  switch (height) {
    case 4:
      mlib_VideoInterpAveY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpAveY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
352 353
}

354
static void avg_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
355
                                  int stride, int height)
Fabrice Bellard's avatar
Fabrice Bellard committed
356
{
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372
  switch (height) {
    case 4:
      mlib_VideoInterpAveXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
    break;

    case 8:
      mlib_VideoInterpAveXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
    break;

    case 16:
      mlib_VideoInterpAveXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
    break;

    default:
      assert(0);
  }
Fabrice Bellard's avatar
Fabrice Bellard committed
373 374
}

375
/* swap byte order of a buffer */
Fabrice Bellard's avatar
Fabrice Bellard committed
376

377
static void bswap_buf_mlib(uint32_t *dst, const uint32_t *src, int w)
Fabrice Bellard's avatar
Fabrice Bellard committed
378
{
379
  mlib_VectorReverseByteOrder_U32_U32(dst, src, w);
Fabrice Bellard's avatar
Fabrice Bellard committed
380 381
}

382
/* transformations */
383

384
static void ff_idct_put_mlib(uint8_t *dest, int line_size, DCTELEM *data)
Fabrice Bellard's avatar
Fabrice Bellard committed
385
{
386
    int i;
387
    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
388

Fabrice Bellard's avatar
Fabrice Bellard committed
389
    mlib_VideoIDCT8x8_S16_S16 (data, data);
390

391 392 393 394 395 396 397 398 399 400 401 402 403
    for(i=0;i<8;i++) {
        dest[0] = cm[data[0]];
        dest[1] = cm[data[1]];
        dest[2] = cm[data[2]];
        dest[3] = cm[data[3]];
        dest[4] = cm[data[4]];
        dest[5] = cm[data[5]];
        dest[6] = cm[data[6]];
        dest[7] = cm[data[7]];

        dest += line_size;
        data += 8;
    }
404
}
405

406
static void ff_idct_add_mlib(uint8_t *dest, int line_size, DCTELEM *data)
407 408
{
    mlib_VideoIDCT8x8_S16_S16 (data, data);
409
    mlib_VideoAddBlock_U8_S16(dest, (mlib_s16 *)data, line_size);
Fabrice Bellard's avatar
Fabrice Bellard committed
410 411
}

412
static void ff_idct_mlib(DCTELEM *data)
413 414 415 416
{
    mlib_VideoIDCT8x8_S16_S16 (data, data);
}

417
static void ff_fdct_mlib(DCTELEM *data)
Fabrice Bellard's avatar
Fabrice Bellard committed
418 419 420 421
{
    mlib_VideoDCT8x8_S16_S16 (data, data);
}

422
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx)
Fabrice Bellard's avatar
Fabrice Bellard committed
423
{
424 425 426 427
    c->get_pixels  = get_pixels_mlib;
    c->diff_pixels = diff_pixels_mlib;
    c->add_pixels_clamped = add_pixels_clamped_mlib;

428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448
    c->put_pixels_tab[0][0] = put_pixels16_mlib;
    c->put_pixels_tab[0][1] = put_pixels16_x2_mlib;
    c->put_pixels_tab[0][2] = put_pixels16_y2_mlib;
    c->put_pixels_tab[0][3] = put_pixels16_xy2_mlib;
    c->put_pixels_tab[1][0] = put_pixels8_mlib;
    c->put_pixels_tab[1][1] = put_pixels8_x2_mlib;
    c->put_pixels_tab[1][2] = put_pixels8_y2_mlib;
    c->put_pixels_tab[1][3] = put_pixels8_xy2_mlib;

    c->avg_pixels_tab[0][0] = avg_pixels16_mlib;
    c->avg_pixels_tab[0][1] = avg_pixels16_x2_mlib;
    c->avg_pixels_tab[0][2] = avg_pixels16_y2_mlib;
    c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mlib;
    c->avg_pixels_tab[1][0] = avg_pixels8_mlib;
    c->avg_pixels_tab[1][1] = avg_pixels8_x2_mlib;
    c->avg_pixels_tab[1][2] = avg_pixels8_y2_mlib;
    c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mlib;

    c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mlib;
    c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mlib;

449
    c->bswap_buf = bswap_buf_mlib;
Fabrice Bellard's avatar
Fabrice Bellard committed
450
}
451 452 453 454

void MPV_common_init_mlib(MpegEncContext *s)
{
    if(s->avctx->dct_algo==FF_DCT_AUTO || s->avctx->dct_algo==FF_DCT_MLIB){
455
        s->dsp.fdct = ff_fdct_mlib;
456
    }
457

458
    if(s->avctx->idct_algo==FF_IDCT_MLIB){
459 460
        s->dsp.idct_put= ff_idct_put_mlib;
        s->dsp.idct_add= ff_idct_add_mlib;
461
        s->dsp.idct    = ff_idct_mlib;
462
        s->dsp.idct_permutation_type= FF_NO_IDCT_PERM;
463
    }
464
}