src/gallium/auxiliary/vl/vl_mpeg12_decoder.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 Younes Manton.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include <math.h>
  29 #include <assert.h>
  30
  31 #include <util/u_memory.h>
  32 #include <util/u_rect.h>
  33 #include <util/u_video.h>
  34
  35 #include "vl_mpeg12_decoder.h"
  36 #include "vl_defines.h"
  37
  38 #define SCALE_FACTOR_SNORM (32768.0f / 256.0f)
  39 #define SCALE_FACTOR_SSCALED (1.0f / 256.0f)
  40
  41 static const unsigned const_empty_block_mask_420[3][2][2] = {
  42    { { 0x20, 0x10 },  { 0x08, 0x04 } },
  43    { { 0x02, 0x02 },  { 0x02, 0x02 } },
  44    { { 0x01, 0x01 },  { 0x01, 0x01 } }
  45 };
  46
  47 static const enum pipe_format const_idct_source_formats[] = {
  48    PIPE_FORMAT_R16G16B16A16_SNORM,
  49    PIPE_FORMAT_R16G16B16A16_SSCALED
  50 };
  51
  52 static const unsigned num_idct_source_formats =
  53    sizeof(const_idct_source_formats) / sizeof(enum pipe_format);
  54
  55 static const enum pipe_format const_idct_intermediate_formats[] = {
  56    PIPE_FORMAT_R16G16B16A16_FLOAT,
  57    PIPE_FORMAT_R16G16B16A16_SNORM,
  58    PIPE_FORMAT_R16G16B16A16_SSCALED,
  59    PIPE_FORMAT_R32G32B32A32_FLOAT
  60 };
  61
  62 static const unsigned num_idct_intermediate_formats =
  63    sizeof(const_idct_intermediate_formats) / sizeof(enum pipe_format);
  64
  65 static const enum pipe_format const_mc_source_formats[] = {
  66    PIPE_FORMAT_R16_SNORM,
  67    PIPE_FORMAT_R16_SSCALED
  68 };
  69
  70 static const unsigned num_mc_source_formats =
  71    sizeof(const_mc_source_formats) / sizeof(enum pipe_format);
  72
  73 static void
  74 map_buffers(struct vl_mpeg12_decoder *ctx, struct vl_mpeg12_buffer *buffer)
  75 {
  76    struct pipe_sampler_view **sampler_views;
  77    struct pipe_resource *tex;
  78    unsigned i;
  79
  80    assert(ctx && buffer);
  81
  82    if (ctx->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
  83       sampler_views = buffer->idct_source->get_sampler_views(buffer->idct_source);
  84    else
  85       sampler_views = buffer->mc_source->get_sampler_views(buffer->mc_source);
  86    assert(sampler_views);
  87
  88    for (i = 0; i < VL_MAX_PLANES; ++i) {
  89       tex = sampler_views[i]->texture;
  90
  91       struct pipe_box rect =
  92       {
  93          0, 0, 0,
  94          tex->width0,
  95          tex->height0,
  96          1
  97       };
  98
  99       buffer->tex_transfer[i] = ctx->pipe->get_transfer
 100       (
 101          ctx->pipe, tex,
 102          0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 103          &rect
 104       );
 105
 106       buffer->texels[i] = ctx->pipe->transfer_map(ctx->pipe, buffer->tex_transfer[i]);
 107    }
 108 }
 109
 110 static void
 111 upload_block(struct vl_mpeg12_buffer *buffer, unsigned plane, unsigned x, unsigned y, short *block)
 112 {
 113    unsigned tex_pitch;
 114    short *texels;
 115
 116    unsigned i;
 117
 118    assert(buffer);
 119    assert(block);
 120
 121    tex_pitch = buffer->tex_transfer[plane]->stride / sizeof(short);
 122    texels = buffer->texels[plane] + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 123
 124    for (i = 0; i < BLOCK_HEIGHT; ++i)
 125       memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 126 }
 127
 128 static void
 129 upload_buffer(struct vl_mpeg12_decoder *ctx,
 130               struct vl_mpeg12_buffer *buffer,
 131               struct pipe_mpeg12_macroblock *mb)
 132 {
 133    short *blocks;
 134    unsigned tb, x, y;
 135
 136    assert(ctx);
 137    assert(buffer);
 138    assert(mb);
 139
 140    blocks = mb->blocks;
 141
 142    for (y = 0; y < 2; ++y) {
 143       for (x = 0; x < 2; ++x, ++tb) {
 144          if (mb->cbp & (*ctx->empty_block_mask)[0][y][x]) {
 145             upload_block(buffer, 0, mb->mbx * 2 + x, mb->mby * 2 + y, blocks);
 146             blocks += BLOCK_WIDTH * BLOCK_HEIGHT;
 147          }
 148       }
 149    }
 150
 151    /* TODO: Implement 422, 444 */
 152    assert(ctx->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 153
 154    for (tb = 1; tb < 3; ++tb) {
 155       if (mb->cbp & (*ctx->empty_block_mask)[tb][0][0]) {
 156          upload_block(buffer, tb, mb->mbx, mb->mby, blocks);
 157          blocks += BLOCK_WIDTH * BLOCK_HEIGHT;
 158       }
 159    }
 160 }
 161
 162 static void
 163 unmap_buffers(struct vl_mpeg12_decoder *ctx, struct vl_mpeg12_buffer *buffer)
 164 {
 165    unsigned i;
 166
 167    assert(ctx && buffer);
 168
 169    for (i = 0; i < VL_MAX_PLANES; ++i) {
 170       ctx->pipe->transfer_unmap(ctx->pipe, buffer->tex_transfer[i]);
 171       ctx->pipe->transfer_destroy(ctx->pipe, buffer->tex_transfer[i]);
 172    }
 173 }
 174
 175 static void
 176 cleanup_idct_buffer(struct vl_mpeg12_buffer *buf)
 177 {
 178    struct vl_mpeg12_decoder *dec;
 179    assert(buf);
 180
 181    dec = (struct vl_mpeg12_decoder*)buf->base.decoder;
 182    assert(dec);
 183
 184    buf->idct_source->destroy(buf->idct_source);
 185    buf->idct_intermediate->destroy(buf->idct_intermediate);
 186    vl_idct_cleanup_buffer(&dec->idct_y, &buf->idct[0]);
 187    vl_idct_cleanup_buffer(&dec->idct_c, &buf->idct[1]);
 188    vl_idct_cleanup_buffer(&dec->idct_c, &buf->idct[2]);
 189 }
 190
 191 static void
 192 vl_mpeg12_buffer_destroy(struct pipe_video_decode_buffer *buffer)
 193 {
 194    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 195    struct vl_mpeg12_decoder *dec;
 196    unsigned i;
 197
 198    assert(buf);
 199
 200    dec = (struct vl_mpeg12_decoder*)buf->base.decoder;
 201    assert(dec);
 202
 203    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 204       cleanup_idct_buffer(buf);
 205
 206    buf->mc_source->destroy(buf->mc_source);
 207    vl_vb_cleanup(&buf->vertex_stream);
 208    for (i = 0; i < VL_MAX_PLANES; ++i)
 209       vl_mc_cleanup_buffer(&buf->mc[i]);
 210
 211    FREE(buf);
 212 }
 213
 214 static void
 215 vl_mpeg12_buffer_map(struct pipe_video_decode_buffer *buffer)
 216 {
 217    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 218    struct vl_mpeg12_decoder *dec;
 219    assert(buf);
 220
 221    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 222    assert(dec);
 223
 224    vl_vb_map(&buf->vertex_stream, dec->pipe);
 225    map_buffers(dec, buf);
 226 }
 227
 228 static void
 229 vl_mpeg12_buffer_add_macroblocks(struct pipe_video_decode_buffer *buffer,
 230                                  unsigned num_macroblocks,
 231                                  struct pipe_macroblock *macroblocks)
 232 {
 233    struct pipe_mpeg12_macroblock *mb = (struct pipe_mpeg12_macroblock*)macroblocks;
 234    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 235    struct vl_mpeg12_decoder *dec;
 236    unsigned i;
 237
 238    assert(buf);
 239
 240    dec =  (struct vl_mpeg12_decoder*)buf->base.decoder;
 241    assert(dec);
 242
 243    assert(num_macroblocks);
 244    assert(macroblocks);
 245    assert(macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
 246
 247    for ( i = 0; i < num_macroblocks; ++i ) {
 248       vl_vb_add_block(&buf->vertex_stream, &mb[i], dec->empty_block_mask);
 249       upload_buffer(dec, buf, &mb[i]);
 250    }
 251 }
 252
 253 static void
 254 vl_mpeg12_buffer_unmap(struct pipe_video_decode_buffer *buffer)
 255 {
 256    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 257    struct vl_mpeg12_decoder *dec;
 258    assert(buf);
 259
 260    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 261    assert(dec);
 262
 263    vl_vb_unmap(&buf->vertex_stream, dec->pipe);
 264    unmap_buffers(dec, buf);
 265 }
 266
 267 static void
 268 vl_mpeg12_destroy(struct pipe_video_decoder *decoder)
 269 {
 270    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 271
 272    assert(decoder);
 273
 274    /* Asserted in softpipe_delete_fs_state() for some reason */
 275    dec->pipe->bind_vs_state(dec->pipe, NULL);
 276    dec->pipe->bind_fs_state(dec->pipe, NULL);
 277
 278    dec->pipe->delete_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 279
 280    vl_mc_cleanup(&dec->mc_y);
 281    vl_mc_cleanup(&dec->mc_c);
 282
 283    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 284       vl_idct_cleanup(&dec->idct_y);
 285       vl_idct_cleanup(&dec->idct_c);
 286    }
 287
 288    dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_ycbcr);
 289    dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_mv);
 290
 291    pipe_resource_reference(&dec->quads.buffer, NULL);
 292    pipe_resource_reference(&dec->pos.buffer, NULL);
 293
 294    FREE(dec);
 295 }
 296
 297 static bool
 298 init_idct_buffer(struct vl_mpeg12_buffer *buffer)
 299 {
 300    enum pipe_format formats[3];
 301
 302    struct pipe_sampler_view **idct_source_sv, **idct_intermediate_sv;
 303    struct pipe_surface **idct_surfaces;
 304
 305    struct vl_mpeg12_decoder *dec;
 306
 307    unsigned i;
 308
 309    assert(buffer);
 310
 311    dec = (struct vl_mpeg12_decoder*)buffer->base.decoder;
 312
 313    formats[0] = formats[1] = formats[2] = dec->idct_source_format;
 314    buffer->idct_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 315                                               dec->base.width / 4, dec->base.height, 1,
 316                                               dec->base.chroma_format,
 317                                               formats, PIPE_USAGE_STREAM);
 318    if (!buffer->idct_source)
 319       goto error_source;
 320
 321    formats[0] = formats[1] = formats[2] = dec->idct_intermediate_format;
 322    buffer->idct_intermediate = vl_video_buffer_init(dec->base.context, dec->pipe,
 323                                                     dec->base.width / dec->nr_of_idct_render_targets,
 324                                                     dec->base.height / 4, dec->nr_of_idct_render_targets,
 325                                                     dec->base.chroma_format,
 326                                                     formats, PIPE_USAGE_STATIC);
 327
 328    if (!buffer->idct_intermediate)
 329       goto error_intermediate;
 330
 331    idct_source_sv = buffer->idct_source->get_sampler_views(buffer->idct_source);
 332    if (!idct_source_sv)
 333       goto error_source_sv;
 334
 335    idct_intermediate_sv = buffer->idct_intermediate->get_sampler_views(buffer->idct_intermediate);
 336    if (!idct_intermediate_sv)
 337       goto error_intermediate_sv;
 338
 339    idct_surfaces = buffer->mc_source->get_surfaces(buffer->mc_source);
 340    if (!idct_surfaces)
 341       goto error_surfaces;
 342
 343    for (i = 0; i < 3; ++i)
 344       if (!vl_idct_init_buffer(i == 0 ? &dec->idct_y : &dec->idct_c,
 345                                &buffer->idct[i], idct_source_sv[i],
 346                                idct_intermediate_sv[i], idct_surfaces[i]))
 347          goto error_plane;
 348
 349    return true;
 350
 351 error_plane:
 352    for (; i > 0; --i)
 353       vl_idct_cleanup_buffer(i == 1 ? &dec->idct_c : &dec->idct_y, &buffer->idct[i - 1]);
 354
 355 error_surfaces:
 356 error_intermediate_sv:
 357 error_source_sv:
 358    buffer->idct_intermediate->destroy(buffer->idct_intermediate);
 359
 360 error_intermediate:
 361    buffer->idct_source->destroy(buffer->idct_source);
 362
 363 error_source:
 364    return false;
 365 }
 366
 367 static struct pipe_video_decode_buffer *
 368 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 369 {
 370    enum pipe_format formats[3];
 371
 372    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 373    struct vl_mpeg12_buffer *buffer;
 374
 375    struct pipe_sampler_view **mc_source_sv;
 376
 377    assert(dec);
 378
 379    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 380    if (buffer == NULL)
 381       return NULL;
 382
 383    buffer->base.decoder = decoder;
 384    buffer->base.destroy = vl_mpeg12_buffer_destroy;
 385    buffer->base.map = vl_mpeg12_buffer_map;
 386    buffer->base.add_macroblocks = vl_mpeg12_buffer_add_macroblocks;
 387    buffer->base.unmap = vl_mpeg12_buffer_unmap;
 388
 389    vl_vb_init(&buffer->vertex_stream, dec->pipe,
 390               dec->base.width / MACROBLOCK_WIDTH,
 391               dec->base.height / MACROBLOCK_HEIGHT);
 392
 393    formats[0] = formats[1] = formats[2] =dec->mc_source_format;
 394    buffer->mc_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 395                                             dec->base.width, dec->base.height, 1,
 396                                             dec->base.chroma_format,
 397                                             formats, PIPE_USAGE_STATIC);
 398
 399    if (!buffer->mc_source)
 400       goto error_mc_source;
 401
 402    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 403       if (!init_idct_buffer(buffer))
 404          goto error_idct;
 405
 406    mc_source_sv = buffer->mc_source->get_sampler_views(buffer->mc_source);
 407    if (!mc_source_sv)
 408       goto error_mc_source_sv;
 409
 410    if(!vl_mc_init_buffer(&dec->mc_y, &buffer->mc[0], mc_source_sv[0]))
 411       goto error_mc_y;
 412
 413    if(!vl_mc_init_buffer(&dec->mc_c, &buffer->mc[1], mc_source_sv[1]))
 414       goto error_mc_cb;
 415
 416    if(!vl_mc_init_buffer(&dec->mc_c, &buffer->mc[2], mc_source_sv[2]))
 417       goto error_mc_cr;
 418
 419    return &buffer->base;
 420
 421 error_mc_cr:
 422    vl_mc_cleanup_buffer(&buffer->mc[1]);
 423
 424 error_mc_cb:
 425    vl_mc_cleanup_buffer(&buffer->mc[0]);
 426
 427 error_mc_y:
 428 error_mc_source_sv:
 429    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 430       cleanup_idct_buffer(buffer);
 431
 432 error_idct:
 433    buffer->mc_source->destroy(buffer->mc_source);
 434
 435 error_mc_source:
 436    vl_vb_cleanup(&buffer->vertex_stream);
 437
 438 error_vertex_stream:
 439    FREE(buffer);
 440    return NULL;
 441 }
 442
 443 static void
 444 vl_mpeg12_decoder_flush_buffer(struct pipe_video_decode_buffer *buffer,
 445                                struct pipe_video_buffer *refs[2],
 446                                struct pipe_video_buffer *dst,
 447                                struct pipe_fence_handle **fence)
 448 {
 449    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer *)buffer;
 450    struct vl_mpeg12_decoder *dec;
 451
 452    struct pipe_sampler_view **sv[2];
 453    struct pipe_surface **surfaces;
 454
 455    struct pipe_vertex_buffer vb[3];
 456
 457    unsigned i, j;
 458
 459    assert(buf);
 460
 461    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 462    assert(dec);
 463
 464    for (i = 0; i < 2; ++i)
 465       sv[i] = refs[i] ? refs[i]->get_sampler_views(refs[i]) : NULL;
 466
 467    surfaces = dst->get_surfaces(dst);
 468
 469    vb[0] = dec->quads;
 470    vb[1] = dec->pos;
 471
 472    dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_mv);
 473    for (i = 0; i < VL_MAX_PLANES; ++i) {
 474       vl_mc_set_surface(&buf->mc[i], surfaces[i]);
 475
 476       for (j = 0; j < 2; ++j) {
 477          if (sv[j] == NULL) continue;
 478
 479          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 480          dec->pipe->set_vertex_buffers(dec->pipe, 3, vb);
 481
 482          vl_mc_render_ref(&buf->mc[i], sv[j][i]);
 483       }
 484    }
 485
 486    dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_ycbcr);
 487    for (i = 0; i < VL_MAX_PLANES; ++i) {
 488       unsigned num_instances = vl_vb_restart(&buf->vertex_stream, i);
 489
 490       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 491       dec->pipe->set_vertex_buffers(dec->pipe, 2, vb);
 492
 493       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 494          vl_idct_flush(i == 0 ? &dec->idct_y : &dec->idct_c, &buf->idct[i], num_instances);
 495
 496       vl_mc_render_ycbcr(&buf->mc[i], num_instances);
 497    }
 498
 499    dec->pipe->flush(dec->pipe, fence);
 500 }
 501
 502 static void
 503 vl_mpeg12_decoder_clear_buffer(struct pipe_video_decode_buffer *buffer)
 504 {
 505    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer *)buffer;
 506    unsigned i;
 507
 508    assert(buf);
 509
 510    for (i = 0; i < VL_MAX_PLANES; ++i)
 511       vl_vb_restart(&buf->vertex_stream, i);
 512 }
 513
 514 static bool
 515 init_pipe_state(struct vl_mpeg12_decoder *dec)
 516 {
 517    struct pipe_depth_stencil_alpha_state dsa;
 518    unsigned i;
 519
 520    assert(dec);
 521
 522    memset(&dsa, 0, sizeof dsa);
 523    dsa.depth.enabled = 0;
 524    dsa.depth.writemask = 0;
 525    dsa.depth.func = PIPE_FUNC_ALWAYS;
 526    for (i = 0; i < 2; ++i) {
 527       dsa.stencil[i].enabled = 0;
 528       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 529       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 530       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 531       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 532       dsa.stencil[i].valuemask = 0;
 533       dsa.stencil[i].writemask = 0;
 534    }
 535    dsa.alpha.enabled = 0;
 536    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 537    dsa.alpha.ref_value = 0;
 538    dec->dsa = dec->pipe->create_depth_stencil_alpha_state(dec->pipe, &dsa);
 539    dec->pipe->bind_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 540
 541    return true;
 542 }
 543
 544 static enum pipe_format
 545 find_first_supported_format(struct vl_mpeg12_decoder *dec,
 546                             const enum pipe_format formats[],
 547                             unsigned num_formats,
 548                             enum pipe_texture_target target)
 549 {
 550    struct pipe_screen *screen;
 551    unsigned i;
 552
 553    assert(dec);
 554
 555    screen = dec->pipe->screen;
 556
 557    for (i = 0; i < num_formats; ++i)
 558       if (screen->is_format_supported(dec->pipe->screen, formats[i], target, 1,
 559                                       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 560          return formats[i];
 561
 562    return PIPE_FORMAT_NONE;
 563 }
 564
 565 static bool
 566 init_idct(struct vl_mpeg12_decoder *dec, unsigned buffer_width, unsigned buffer_height)
 567 {
 568    unsigned chroma_width, chroma_height, chroma_blocks_x, chroma_blocks_y;
 569    struct pipe_sampler_view *matrix, *transpose;
 570    float matrix_scale, transpose_scale;
 571
 572    dec->nr_of_idct_render_targets = dec->pipe->screen->get_param(dec->pipe->screen, PIPE_CAP_MAX_RENDER_TARGETS);
 573
 574    // more than 4 render targets usually doesn't makes any seens
 575    dec->nr_of_idct_render_targets = MIN2(dec->nr_of_idct_render_targets, 4);
 576
 577    dec->idct_source_format = find_first_supported_format(dec, const_idct_source_formats,
 578                                                          num_idct_source_formats, PIPE_TEXTURE_2D);
 579
 580    if (dec->idct_source_format == PIPE_FORMAT_NONE)
 581       return false;
 582
 583    dec->idct_intermediate_format = find_first_supported_format(dec, const_idct_intermediate_formats,
 584                                                                num_idct_intermediate_formats, PIPE_TEXTURE_3D);
 585
 586    if (dec->idct_intermediate_format == PIPE_FORMAT_NONE)
 587       return false;
 588
 589    switch (dec->idct_source_format) {
 590    case PIPE_FORMAT_R16G16B16A16_SSCALED:
 591       matrix_scale = SCALE_FACTOR_SSCALED;
 592       break;
 593
 594    case PIPE_FORMAT_R16G16B16A16_SNORM:
 595       matrix_scale = SCALE_FACTOR_SNORM;
 596       break;
 597
 598    default:
 599       assert(0);
 600       return false;
 601    }
 602
 603    if (dec->idct_intermediate_format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
 604        dec->idct_intermediate_format == PIPE_FORMAT_R32G32B32A32_FLOAT)
 605       transpose_scale = 1.0f;
 606    else
 607       transpose_scale = matrix_scale = sqrt(matrix_scale);
 608
 609    if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 610       transpose_scale /= SCALE_FACTOR_SSCALED;
 611
 612    if (!(matrix = vl_idct_upload_matrix(dec->pipe, matrix_scale)))
 613       goto error_matrix;
 614
 615    if (matrix_scale != transpose_scale) {
 616       if (!(transpose = vl_idct_upload_matrix(dec->pipe, transpose_scale)))
 617          goto error_transpose;
 618    } else
 619       pipe_sampler_view_reference(&transpose, matrix);
 620
 621    if (!vl_idct_init(&dec->idct_y, dec->pipe, buffer_width, buffer_height,
 622                      2, 2, dec->nr_of_idct_render_targets, matrix, transpose))
 623       goto error_y;
 624
 625    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
 626       chroma_width = buffer_width / 2;
 627       chroma_height = buffer_height / 2;
 628       chroma_blocks_x = 1;
 629       chroma_blocks_y = 1;
 630    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
 631       chroma_width = buffer_width;
 632       chroma_height = buffer_height / 2;
 633       chroma_blocks_x = 2;
 634       chroma_blocks_y = 1;
 635    } else {
 636       chroma_width = buffer_width;
 637       chroma_height = buffer_height;
 638       chroma_blocks_x = 2;
 639       chroma_blocks_y = 2;
 640    }
 641
 642    if(!vl_idct_init(&dec->idct_c, dec->pipe, chroma_width, chroma_height,
 643                     chroma_blocks_x, chroma_blocks_y,
 644                     dec->nr_of_idct_render_targets, matrix, transpose))
 645       goto error_c;
 646
 647    pipe_sampler_view_reference(&matrix, NULL);
 648    pipe_sampler_view_reference(&transpose, NULL);
 649    return true;
 650
 651 error_c:
 652    vl_idct_cleanup(&dec->idct_y);
 653
 654 error_y:
 655    pipe_sampler_view_reference(&transpose, NULL);
 656
 657 error_transpose:
 658    pipe_sampler_view_reference(&matrix, NULL);
 659
 660 error_matrix:
 661    return false;
 662 }
 663
 664 struct pipe_video_decoder *
 665 vl_create_mpeg12_decoder(struct pipe_video_context *context,
 666                          struct pipe_context *pipe,
 667                          enum pipe_video_profile profile,
 668                          enum pipe_video_entrypoint entrypoint,
 669                          enum pipe_video_chroma_format chroma_format,
 670                          unsigned width, unsigned height)
 671 {
 672    struct vl_mpeg12_decoder *dec;
 673    float mc_scale;
 674
 675    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
 676
 677    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
 678
 679    if (!dec)
 680       return NULL;
 681
 682    dec->base.context = context;
 683    dec->base.profile = profile;
 684    dec->base.entrypoint = entrypoint;
 685    dec->base.chroma_format = chroma_format;
 686    dec->base.width = width;
 687    dec->base.height = height;
 688
 689    dec->base.destroy = vl_mpeg12_destroy;
 690    dec->base.create_buffer = vl_mpeg12_create_buffer;
 691    dec->base.flush_buffer = vl_mpeg12_decoder_flush_buffer;
 692    dec->base.clear_buffer = vl_mpeg12_decoder_clear_buffer;
 693
 694    dec->base.width = align(width, MACROBLOCK_WIDTH);
 695    dec->base.height = align(height, MACROBLOCK_HEIGHT);
 696
 697    dec->pipe = pipe;
 698
 699    dec->quads = vl_vb_upload_quads(dec->pipe, 2, 2);
 700    dec->pos = vl_vb_upload_pos(
 701       dec->pipe,
 702       dec->base.width / MACROBLOCK_WIDTH,
 703       dec->base.height / MACROBLOCK_HEIGHT
 704    );
 705
 706    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->pipe);
 707    dec->ves_mv = vl_vb_get_ves_mv(dec->pipe);
 708
 709    /* TODO: Implement 422, 444 */
 710    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 711    dec->empty_block_mask = &const_empty_block_mask_420;
 712
 713    dec->mc_source_format = find_first_supported_format(dec, const_mc_source_formats,
 714                                                        num_mc_source_formats, PIPE_TEXTURE_3D);
 715
 716    if (dec->mc_source_format == PIPE_FORMAT_NONE)
 717       return NULL;
 718
 719    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 720       if (!init_idct(dec, dec->base.width, dec->base.height))
 721          goto error_idct;
 722       if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 723          mc_scale = SCALE_FACTOR_SSCALED;
 724       else
 725          mc_scale = 1.0f;
 726    } else {
 727       switch (dec->mc_source_format) {
 728       case PIPE_FORMAT_R16_SNORM:
 729          mc_scale = SCALE_FACTOR_SNORM;
 730          break;
 731
 732       case PIPE_FORMAT_R16_SSCALED:
 733          mc_scale = SCALE_FACTOR_SSCALED;
 734          break;
 735
 736       default:
 737          assert(0);
 738          return NULL;
 739       }
 740    }
 741
 742    if (!vl_mc_init(&dec->mc_y, dec->pipe, dec->base.width, dec->base.height, MACROBLOCK_HEIGHT, mc_scale))
 743       goto error_mc_y;
 744
 745    // TODO
 746    if (!vl_mc_init(&dec->mc_c, dec->pipe, dec->base.width, dec->base.height, BLOCK_HEIGHT, mc_scale))
 747       goto error_mc_c;
 748
 749    if (!init_pipe_state(dec))
 750       goto error_pipe_state;
 751
 752    return &dec->base;
 753
 754 error_pipe_state:
 755    vl_mc_cleanup(&dec->mc_c);
 756
 757 error_mc_c:
 758    vl_mc_cleanup(&dec->mc_y);
 759
 760 error_mc_y:
 761    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 762       vl_idct_cleanup(&dec->idct_y);
 763       vl_idct_cleanup(&dec->idct_c);
 764    }
 765
 766 error_idct:
 767    FREE(dec);
 768    return NULL;
 769 }