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    unsigned i;
 272
 273    assert(decoder);
 274
 275    /* Asserted in softpipe_delete_fs_state() for some reason */
 276    dec->pipe->bind_vs_state(dec->pipe, NULL);
 277    dec->pipe->bind_fs_state(dec->pipe, NULL);
 278
 279    dec->pipe->delete_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 280
 281    vl_mc_cleanup(&dec->mc_y);
 282    vl_mc_cleanup(&dec->mc_c);
 283
 284    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 285       vl_idct_cleanup(&dec->idct_y);
 286       vl_idct_cleanup(&dec->idct_c);
 287    }
 288
 289    for (i = 0; i < VL_MAX_PLANES; ++i)
 290       dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_eb[i]);
 291
 292    dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_mv);
 293
 294    pipe_resource_reference(&dec->quads.buffer, NULL);
 295    pipe_resource_reference(&dec->pos.buffer, NULL);
 296
 297    FREE(dec);
 298 }
 299
 300 static bool
 301 init_idct_buffer(struct vl_mpeg12_buffer *buffer)
 302 {
 303    enum pipe_format formats[3];
 304
 305    struct pipe_sampler_view **idct_source_sv, **idct_intermediate_sv;
 306    struct pipe_surface **idct_surfaces;
 307
 308    struct vl_mpeg12_decoder *dec;
 309
 310    unsigned i;
 311
 312    assert(buffer);
 313
 314    dec = (struct vl_mpeg12_decoder*)buffer->base.decoder;
 315
 316    formats[0] = formats[1] = formats[2] = dec->idct_source_format;
 317    buffer->idct_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 318                                               dec->base.width / 4, dec->base.height, 1,
 319                                               dec->base.chroma_format,
 320                                               formats, PIPE_USAGE_STREAM);
 321    if (!buffer->idct_source)
 322       goto error_source;
 323
 324    formats[0] = formats[1] = formats[2] = dec->idct_intermediate_format;
 325    buffer->idct_intermediate = vl_video_buffer_init(dec->base.context, dec->pipe,
 326                                                     dec->base.width / dec->nr_of_idct_render_targets,
 327                                                     dec->base.height / 4, dec->nr_of_idct_render_targets,
 328                                                     dec->base.chroma_format,
 329                                                     formats, PIPE_USAGE_STATIC);
 330
 331    if (!buffer->idct_intermediate)
 332       goto error_intermediate;
 333
 334    idct_source_sv = buffer->idct_source->get_sampler_views(buffer->idct_source);
 335    if (!idct_source_sv)
 336       goto error_source_sv;
 337
 338    idct_intermediate_sv = buffer->idct_intermediate->get_sampler_views(buffer->idct_intermediate);
 339    if (!idct_intermediate_sv)
 340       goto error_intermediate_sv;
 341
 342    idct_surfaces = buffer->mc_source->get_surfaces(buffer->mc_source);
 343    if (!idct_surfaces)
 344       goto error_surfaces;
 345
 346    for (i = 0; i < 3; ++i)
 347       if (!vl_idct_init_buffer(i == 0 ? &dec->idct_y : &dec->idct_c,
 348                                &buffer->idct[i], idct_source_sv[i],
 349                                idct_intermediate_sv[i], idct_surfaces[i]))
 350          goto error_plane;
 351
 352    return true;
 353
 354 error_plane:
 355    for (; i > 0; --i)
 356       vl_idct_cleanup_buffer(i == 1 ? &dec->idct_c : &dec->idct_y, &buffer->idct[i - 1]);
 357
 358 error_surfaces:
 359 error_intermediate_sv:
 360 error_source_sv:
 361    buffer->idct_intermediate->destroy(buffer->idct_intermediate);
 362
 363 error_intermediate:
 364    buffer->idct_source->destroy(buffer->idct_source);
 365
 366 error_source:
 367    return false;
 368 }
 369
 370 static struct pipe_video_decode_buffer *
 371 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 372 {
 373    enum pipe_format formats[3];
 374
 375    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 376    struct vl_mpeg12_buffer *buffer;
 377
 378    struct pipe_sampler_view **mc_source_sv;
 379
 380    assert(dec);
 381
 382    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 383    if (buffer == NULL)
 384       return NULL;
 385
 386    buffer->base.decoder = decoder;
 387    buffer->base.destroy = vl_mpeg12_buffer_destroy;
 388    buffer->base.map = vl_mpeg12_buffer_map;
 389    buffer->base.add_macroblocks = vl_mpeg12_buffer_add_macroblocks;
 390    buffer->base.unmap = vl_mpeg12_buffer_unmap;
 391
 392    vl_vb_init(&buffer->vertex_stream, dec->pipe,
 393               dec->base.width / MACROBLOCK_WIDTH,
 394               dec->base.height / MACROBLOCK_HEIGHT);
 395
 396    formats[0] = formats[1] = formats[2] =dec->mc_source_format;
 397    buffer->mc_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 398                                             dec->base.width, dec->base.height, 1,
 399                                             dec->base.chroma_format,
 400                                             formats, PIPE_USAGE_STATIC);
 401
 402    if (!buffer->mc_source)
 403       goto error_mc_source;
 404
 405    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 406       if (!init_idct_buffer(buffer))
 407          goto error_idct;
 408
 409    mc_source_sv = buffer->mc_source->get_sampler_views(buffer->mc_source);
 410    if (!mc_source_sv)
 411       goto error_mc_source_sv;
 412
 413    if(!vl_mc_init_buffer(&dec->mc_y, &buffer->mc[0], mc_source_sv[0]))
 414       goto error_mc_y;
 415
 416    if(!vl_mc_init_buffer(&dec->mc_c, &buffer->mc[1], mc_source_sv[1]))
 417       goto error_mc_cb;
 418
 419    if(!vl_mc_init_buffer(&dec->mc_c, &buffer->mc[2], mc_source_sv[2]))
 420       goto error_mc_cr;
 421
 422    return &buffer->base;
 423
 424 error_mc_cr:
 425    vl_mc_cleanup_buffer(&buffer->mc[1]);
 426
 427 error_mc_cb:
 428    vl_mc_cleanup_buffer(&buffer->mc[0]);
 429
 430 error_mc_y:
 431 error_mc_source_sv:
 432    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 433       cleanup_idct_buffer(buffer);
 434
 435 error_idct:
 436    buffer->mc_source->destroy(buffer->mc_source);
 437
 438 error_mc_source:
 439    vl_vb_cleanup(&buffer->vertex_stream);
 440
 441 error_vertex_stream:
 442    FREE(buffer);
 443    return NULL;
 444 }
 445
 446 static void
 447 vl_mpeg12_decoder_flush_buffer(struct pipe_video_decode_buffer *buffer,
 448                                struct pipe_video_buffer *refs[2],
 449                                struct pipe_video_buffer *dst,
 450                                struct pipe_fence_handle **fence)
 451 {
 452    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer *)buffer;
 453    struct vl_mpeg12_decoder *dec;
 454
 455    struct pipe_sampler_view **sv[2];
 456    struct pipe_surface **surfaces;
 457
 458    struct pipe_vertex_buffer vb[3];
 459
 460    unsigned num_instances;
 461    unsigned i, j;
 462
 463    assert(buf);
 464
 465    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 466    assert(dec);
 467
 468    for (i = 0; i < 2; ++i)
 469       sv[i] = refs[i] ? refs[i]->get_sampler_views(refs[i]) : NULL;
 470
 471    surfaces = dst->get_surfaces(dst);
 472
 473    num_instances = vl_vb_restart(&buf->vertex_stream);
 474
 475    vb[0] = dec->quads;
 476    vb[1] = dec->pos;
 477
 478    dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_mv);
 479    for (i = 0; i < VL_MAX_PLANES; ++i) {
 480       vl_mc_set_surface(&buf->mc[i], surfaces[i]);
 481
 482       for (j = 0; j < 2; ++j) {
 483          if (sv[j] == NULL) continue;
 484
 485          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 486          dec->pipe->set_vertex_buffers(dec->pipe, 3, vb);
 487
 488          vl_mc_render_ref(&buf->mc[i], sv[j][i]);
 489       }
 490    }
 491
 492    vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream);
 493    dec->pipe->set_vertex_buffers(dec->pipe, 2, vb);
 494
 495    for (i = 0; i < VL_MAX_PLANES; ++i) {
 496       dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_eb[i]);
 497       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 498          vl_idct_flush(i == 0 ? &dec->idct_y : &dec->idct_c, &buf->idct[i], num_instances);
 499
 500       vl_mc_render_ycbcr(&buf->mc[i], num_instances);
 501    }
 502
 503    dec->pipe->flush(dec->pipe, fence);
 504 }
 505
 506 static void
 507 vl_mpeg12_decoder_clear_buffer(struct pipe_video_decode_buffer *buffer)
 508 {
 509    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer *)buffer;
 510
 511    assert(buf);
 512
 513    vl_vb_restart(&buf->vertex_stream);
 514 }
 515
 516 static bool
 517 init_pipe_state(struct vl_mpeg12_decoder *dec)
 518 {
 519    struct pipe_depth_stencil_alpha_state dsa;
 520    unsigned i;
 521
 522    assert(dec);
 523
 524    memset(&dsa, 0, sizeof dsa);
 525    dsa.depth.enabled = 0;
 526    dsa.depth.writemask = 0;
 527    dsa.depth.func = PIPE_FUNC_ALWAYS;
 528    for (i = 0; i < 2; ++i) {
 529       dsa.stencil[i].enabled = 0;
 530       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 531       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 532       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 533       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 534       dsa.stencil[i].valuemask = 0;
 535       dsa.stencil[i].writemask = 0;
 536    }
 537    dsa.alpha.enabled = 0;
 538    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 539    dsa.alpha.ref_value = 0;
 540    dec->dsa = dec->pipe->create_depth_stencil_alpha_state(dec->pipe, &dsa);
 541    dec->pipe->bind_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 542
 543    return true;
 544 }
 545
 546 static enum pipe_format
 547 find_first_supported_format(struct vl_mpeg12_decoder *dec,
 548                             const enum pipe_format formats[],
 549                             unsigned num_formats,
 550                             enum pipe_texture_target target)
 551 {
 552    struct pipe_screen *screen;
 553    unsigned i;
 554
 555    assert(dec);
 556
 557    screen = dec->pipe->screen;
 558
 559    for (i = 0; i < num_formats; ++i)
 560       if (screen->is_format_supported(dec->pipe->screen, formats[i], target, 1,
 561                                       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 562          return formats[i];
 563
 564    return PIPE_FORMAT_NONE;
 565 }
 566
 567 static bool
 568 init_idct(struct vl_mpeg12_decoder *dec, unsigned buffer_width, unsigned buffer_height)
 569 {
 570    unsigned chroma_width, chroma_height, chroma_blocks_x, chroma_blocks_y;
 571    struct pipe_sampler_view *matrix, *transpose;
 572    float matrix_scale, transpose_scale;
 573
 574    dec->nr_of_idct_render_targets = dec->pipe->screen->get_param(dec->pipe->screen, PIPE_CAP_MAX_RENDER_TARGETS);
 575
 576    // more than 4 render targets usually doesn't makes any seens
 577    dec->nr_of_idct_render_targets = MIN2(dec->nr_of_idct_render_targets, 4);
 578
 579    dec->idct_source_format = find_first_supported_format(dec, const_idct_source_formats,
 580                                                          num_idct_source_formats, PIPE_TEXTURE_2D);
 581
 582    if (dec->idct_source_format == PIPE_FORMAT_NONE)
 583       return false;
 584
 585    dec->idct_intermediate_format = find_first_supported_format(dec, const_idct_intermediate_formats,
 586                                                                num_idct_intermediate_formats, PIPE_TEXTURE_3D);
 587
 588    if (dec->idct_intermediate_format == PIPE_FORMAT_NONE)
 589       return false;
 590
 591    switch (dec->idct_source_format) {
 592    case PIPE_FORMAT_R16G16B16A16_SSCALED:
 593       matrix_scale = SCALE_FACTOR_SSCALED;
 594       break;
 595
 596    case PIPE_FORMAT_R16G16B16A16_SNORM:
 597       matrix_scale = SCALE_FACTOR_SNORM;
 598       break;
 599
 600    default:
 601       assert(0);
 602       return false;
 603    }
 604
 605    if (dec->idct_intermediate_format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
 606        dec->idct_intermediate_format == PIPE_FORMAT_R32G32B32A32_FLOAT)
 607       transpose_scale = 1.0f;
 608    else
 609       transpose_scale = matrix_scale = sqrt(matrix_scale);
 610
 611    if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 612       transpose_scale /= SCALE_FACTOR_SSCALED;
 613
 614    if (!(matrix = vl_idct_upload_matrix(dec->pipe, matrix_scale)))
 615       goto error_matrix;
 616
 617    if (matrix_scale != transpose_scale) {
 618       if (!(transpose = vl_idct_upload_matrix(dec->pipe, transpose_scale)))
 619          goto error_transpose;
 620    } else
 621       pipe_sampler_view_reference(&transpose, matrix);
 622
 623    if (!vl_idct_init(&dec->idct_y, dec->pipe, buffer_width, buffer_height,
 624                      2, 2, dec->nr_of_idct_render_targets, matrix, transpose))
 625       goto error_y;
 626
 627    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
 628       chroma_width = buffer_width / 2;
 629       chroma_height = buffer_height / 2;
 630       chroma_blocks_x = 1;
 631       chroma_blocks_y = 1;
 632    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
 633       chroma_width = buffer_width;
 634       chroma_height = buffer_height / 2;
 635       chroma_blocks_x = 2;
 636       chroma_blocks_y = 1;
 637    } else {
 638       chroma_width = buffer_width;
 639       chroma_height = buffer_height;
 640       chroma_blocks_x = 2;
 641       chroma_blocks_y = 2;
 642    }
 643
 644    if(!vl_idct_init(&dec->idct_c, dec->pipe, chroma_width, chroma_height,
 645                     chroma_blocks_x, chroma_blocks_y,
 646                     dec->nr_of_idct_render_targets, matrix, transpose))
 647       goto error_c;
 648
 649    pipe_sampler_view_reference(&matrix, NULL);
 650    pipe_sampler_view_reference(&transpose, NULL);
 651    return true;
 652
 653 error_c:
 654    vl_idct_cleanup(&dec->idct_y);
 655
 656 error_y:
 657    pipe_sampler_view_reference(&transpose, NULL);
 658
 659 error_transpose:
 660    pipe_sampler_view_reference(&matrix, NULL);
 661
 662 error_matrix:
 663    return false;
 664 }
 665
 666 struct pipe_video_decoder *
 667 vl_create_mpeg12_decoder(struct pipe_video_context *context,
 668                          struct pipe_context *pipe,
 669                          enum pipe_video_profile profile,
 670                          enum pipe_video_entrypoint entrypoint,
 671                          enum pipe_video_chroma_format chroma_format,
 672                          unsigned width, unsigned height)
 673 {
 674    struct vl_mpeg12_decoder *dec;
 675    float mc_scale;
 676    unsigned i;
 677
 678    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
 679
 680    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
 681
 682    if (!dec)
 683       return NULL;
 684
 685    dec->base.context = context;
 686    dec->base.profile = profile;
 687    dec->base.entrypoint = entrypoint;
 688    dec->base.chroma_format = chroma_format;
 689    dec->base.width = width;
 690    dec->base.height = height;
 691
 692    dec->base.destroy = vl_mpeg12_destroy;
 693    dec->base.create_buffer = vl_mpeg12_create_buffer;
 694    dec->base.flush_buffer = vl_mpeg12_decoder_flush_buffer;
 695    dec->base.clear_buffer = vl_mpeg12_decoder_clear_buffer;
 696
 697    dec->base.width = align(width, MACROBLOCK_WIDTH);
 698    dec->base.height = align(height, MACROBLOCK_HEIGHT);
 699
 700    dec->pipe = pipe;
 701
 702    dec->quads = vl_vb_upload_quads(dec->pipe, 2, 2);
 703    dec->pos = vl_vb_upload_pos(
 704       dec->pipe,
 705       dec->base.width / MACROBLOCK_WIDTH,
 706       dec->base.height / MACROBLOCK_HEIGHT
 707    );
 708
 709    for (i = 0; i < VL_MAX_PLANES; ++i)
 710       dec->ves_eb[i] = vl_vb_get_ves_eb(dec->pipe, i);
 711
 712    dec->ves_mv = vl_vb_get_ves_mv(dec->pipe);
 713
 714    /* TODO: Implement 422, 444 */
 715    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 716    dec->empty_block_mask = &const_empty_block_mask_420;
 717
 718    dec->mc_source_format = find_first_supported_format(dec, const_mc_source_formats,
 719                                                        num_mc_source_formats, PIPE_TEXTURE_3D);
 720
 721    if (dec->mc_source_format == PIPE_FORMAT_NONE)
 722       return NULL;
 723
 724    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 725       if (!init_idct(dec, dec->base.width, dec->base.height))
 726          goto error_idct;
 727       if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 728          mc_scale = SCALE_FACTOR_SSCALED;
 729       else
 730          mc_scale = 1.0f;
 731    } else {
 732       switch (dec->mc_source_format) {
 733       case PIPE_FORMAT_R16_SNORM:
 734          mc_scale = SCALE_FACTOR_SNORM;
 735          break;
 736
 737       case PIPE_FORMAT_R16_SSCALED:
 738          mc_scale = SCALE_FACTOR_SSCALED;
 739          break;
 740
 741       default:
 742          assert(0);
 743          return NULL;
 744       }
 745    }
 746
 747    if (!vl_mc_init(&dec->mc_y, dec->pipe, dec->base.width, dec->base.height, MACROBLOCK_HEIGHT, mc_scale))
 748       goto error_mc_y;
 749
 750    // TODO
 751    if (!vl_mc_init(&dec->mc_c, dec->pipe, dec->base.width, dec->base.height, BLOCK_HEIGHT, mc_scale))
 752       goto error_mc_c;
 753
 754    if (!init_pipe_state(dec))
 755       goto error_pipe_state;
 756
 757    return &dec->base;
 758
 759 error_pipe_state:
 760    vl_mc_cleanup(&dec->mc_c);
 761
 762 error_mc_c:
 763    vl_mc_cleanup(&dec->mc_y);
 764
 765 error_mc_y:
 766    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 767       vl_idct_cleanup(&dec->idct_y);
 768       vl_idct_cleanup(&dec->idct_c);
 769    }
 770
 771 error_idct:
 772    FREE(dec);
 773    return NULL;
 774 }