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_sampler.h"
  34 #include "util/u_video.h"
  35
  36 #include "vl_mpeg12_decoder.h"
  37 #include "vl_defines.h"
  38
  39 #define SCALE_FACTOR_SNORM (32768.0f / 256.0f)
  40 #define SCALE_FACTOR_SSCALED (1.0f / 256.0f)
  41
  42 struct format_config {
  43    enum pipe_format zscan_source_format;
  44    enum pipe_format idct_source_format;
  45    enum pipe_format mc_source_format;
  46
  47    float idct_scale;
  48    float mc_scale;
  49 };
  50
  51 static const struct format_config bitstream_format_config[] = {
  52 //   { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SSCALED },
  53 //   { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, 1.0f, SCALE_FACTOR_SSCALED },
  54    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SNORM },
  55    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, 1.0f, SCALE_FACTOR_SNORM }
  56 };
  57
  58 static const unsigned num_bitstream_format_configs =
  59    sizeof(bitstream_format_config) / sizeof(struct format_config);
  60
  61 static const struct format_config idct_format_config[] = {
  62 //   { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SSCALED },
  63 //   { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, 1.0f, SCALE_FACTOR_SSCALED },
  64    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SNORM },
  65    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, 1.0f, SCALE_FACTOR_SNORM }
  66 };
  67
  68 static const unsigned num_idct_format_configs =
  69    sizeof(idct_format_config) / sizeof(struct format_config);
  70
  71 static const struct format_config mc_format_config[] = {
  72    //{ PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_NONE, PIPE_FORMAT_R16_SSCALED, 0.0f, SCALE_FACTOR_SSCALED },
  73    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_NONE, PIPE_FORMAT_R16_SNORM, 0.0f, SCALE_FACTOR_SNORM }
  74 };
  75
  76 static const unsigned num_mc_format_configs =
  77    sizeof(mc_format_config) / sizeof(struct format_config);
  78
  79 static const unsigned const_empty_block_mask_420[3][2][2] = {
  80    { { 0x20, 0x10 },  { 0x08, 0x04 } },
  81    { { 0x02, 0x02 },  { 0x02, 0x02 } },
  82    { { 0x01, 0x01 },  { 0x01, 0x01 } }
  83 };
  84
  85 static bool
  86 init_zscan_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buffer)
  87 {
  88    struct pipe_resource *res, res_tmpl;
  89    struct pipe_sampler_view sv_tmpl;
  90    struct pipe_surface **destination;
  91
  92    unsigned i;
  93
  94    assert(dec && buffer);
  95
  96    memset(&res_tmpl, 0, sizeof(res_tmpl));
  97    res_tmpl.target = PIPE_TEXTURE_2D;
  98    res_tmpl.format = dec->zscan_source_format;
  99    res_tmpl.width0 = dec->blocks_per_line * BLOCK_WIDTH * BLOCK_HEIGHT;
 100    res_tmpl.height0 = align(dec->num_blocks, dec->blocks_per_line) / dec->blocks_per_line;
 101    res_tmpl.depth0 = 1;
 102    res_tmpl.array_size = 1;
 103    res_tmpl.usage = PIPE_USAGE_STREAM;
 104    res_tmpl.bind = PIPE_BIND_SAMPLER_VIEW;
 105
 106    res = dec->base.context->screen->resource_create(dec->base.context->screen, &res_tmpl);
 107    if (!res)
 108       goto error_source;
 109
 110
 111    memset(&sv_tmpl, 0, sizeof(sv_tmpl));
 112    u_sampler_view_default_template(&sv_tmpl, res, res->format);
 113    sv_tmpl.swizzle_r = sv_tmpl.swizzle_g = sv_tmpl.swizzle_b = sv_tmpl.swizzle_a = PIPE_SWIZZLE_RED;
 114    buffer->zscan_source = dec->base.context->create_sampler_view(dec->base.context, res, &sv_tmpl);
 115    pipe_resource_reference(&res, NULL);
 116    if (!buffer->zscan_source)
 117       goto error_sampler;
 118
 119    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 120       destination = dec->idct_source->get_surfaces(dec->idct_source);
 121    else
 122       destination = dec->mc_source->get_surfaces(dec->mc_source);
 123
 124    if (!destination)
 125       goto error_surface;
 126
 127    for (i = 0; i < VL_MAX_PLANES; ++i)
 128       if (!vl_zscan_init_buffer(i == 0 ? &dec->zscan_y : &dec->zscan_c,
 129                                 &buffer->zscan[i], buffer->zscan_source, destination[i]))
 130          goto error_plane;
 131
 132    return true;
 133
 134 error_plane:
 135    for (; i > 0; --i)
 136       vl_zscan_cleanup_buffer(&buffer->zscan[i - 1]);
 137
 138 error_surface:
 139 error_sampler:
 140    pipe_sampler_view_reference(&buffer->zscan_source, NULL);
 141
 142 error_source:
 143    return false;
 144 }
 145
 146 static void
 147 cleanup_zscan_buffer(struct vl_mpeg12_buffer *buffer)
 148 {
 149    unsigned i;
 150
 151    assert(buffer);
 152
 153    for (i = 0; i < VL_MAX_PLANES; ++i)
 154       vl_zscan_cleanup_buffer(&buffer->zscan[i]);
 155
 156    pipe_sampler_view_reference(&buffer->zscan_source, NULL);
 157 }
 158
 159 static bool
 160 init_idct_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buffer)
 161 {
 162    struct pipe_sampler_view **idct_source_sv, **mc_source_sv;
 163
 164    unsigned i;
 165
 166    assert(dec && buffer);
 167
 168    idct_source_sv = dec->idct_source->get_sampler_view_planes(dec->idct_source);
 169    if (!idct_source_sv)
 170       goto error_source_sv;
 171
 172    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 173    if (!mc_source_sv)
 174       goto error_mc_source_sv;
 175
 176    for (i = 0; i < 3; ++i)
 177       if (!vl_idct_init_buffer(i == 0 ? &dec->idct_y : &dec->idct_c,
 178                                &buffer->idct[i], idct_source_sv[i],
 179                                mc_source_sv[i]))
 180          goto error_plane;
 181
 182    return true;
 183
 184 error_plane:
 185    for (; i > 0; --i)
 186       vl_idct_cleanup_buffer(&buffer->idct[i - 1]);
 187
 188 error_mc_source_sv:
 189 error_source_sv:
 190    return false;
 191 }
 192
 193 static void
 194 cleanup_idct_buffer(struct vl_mpeg12_buffer *buf)
 195 {
 196    unsigned i;
 197
 198    assert(buf);
 199
 200    for (i = 0; i < 3; ++i)
 201       vl_idct_cleanup_buffer(&buf->idct[0]);
 202 }
 203
 204 static bool
 205 init_mc_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buf)
 206 {
 207    assert(dec && buf);
 208
 209    if(!vl_mc_init_buffer(&dec->mc_y, &buf->mc[0]))
 210       goto error_mc_y;
 211
 212    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[1]))
 213       goto error_mc_cb;
 214
 215    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[2]))
 216       goto error_mc_cr;
 217
 218    return true;
 219
 220 error_mc_cr:
 221    vl_mc_cleanup_buffer(&buf->mc[1]);
 222
 223 error_mc_cb:
 224    vl_mc_cleanup_buffer(&buf->mc[0]);
 225
 226 error_mc_y:
 227    return false;
 228 }
 229
 230 static void
 231 cleanup_mc_buffer(struct vl_mpeg12_buffer *buf)
 232 {
 233    unsigned i;
 234
 235    assert(buf);
 236
 237    for (i = 0; i < VL_MAX_PLANES; ++i)
 238       vl_mc_cleanup_buffer(&buf->mc[i]);
 239 }
 240
 241 static INLINE void
 242 MacroBlockTypeToPipeWeights(const struct pipe_mpeg12_macroblock *mb, unsigned weights[2])
 243 {
 244    assert(mb);
 245
 246    switch (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_MOTION_FORWARD | PIPE_MPEG12_MB_TYPE_MOTION_BACKWARD)) {
 247    case PIPE_MPEG12_MB_TYPE_MOTION_FORWARD:
 248       weights[0] = PIPE_VIDEO_MV_WEIGHT_MAX;
 249       weights[1] = PIPE_VIDEO_MV_WEIGHT_MIN;
 250       break;
 251
 252    case (PIPE_MPEG12_MB_TYPE_MOTION_FORWARD | PIPE_MPEG12_MB_TYPE_MOTION_BACKWARD):
 253       weights[0] = PIPE_VIDEO_MV_WEIGHT_HALF;
 254       weights[1] = PIPE_VIDEO_MV_WEIGHT_HALF;
 255       break;
 256
 257    case PIPE_MPEG12_MB_TYPE_MOTION_BACKWARD:
 258       weights[0] = PIPE_VIDEO_MV_WEIGHT_MIN;
 259       weights[1] = PIPE_VIDEO_MV_WEIGHT_MAX;
 260       break;
 261
 262    default:
 263       if (mb->macroblock_type & PIPE_MPEG12_MB_TYPE_INTRA) {
 264          weights[0] = PIPE_VIDEO_MV_WEIGHT_MIN;
 265          weights[1] = PIPE_VIDEO_MV_WEIGHT_MIN;
 266       } else {
 267          /* no motion vector, but also not intra mb ->
 268             just copy the old frame content */
 269          weights[0] = PIPE_VIDEO_MV_WEIGHT_MAX;
 270          weights[1] = PIPE_VIDEO_MV_WEIGHT_MIN;
 271       }
 272       break;
 273    }
 274 }
 275
 276 static INLINE struct vl_motionvector
 277 MotionVectorToPipe(const struct pipe_mpeg12_macroblock *mb, unsigned vector,
 278                    unsigned field_select_mask, unsigned weight)
 279 {
 280    struct vl_motionvector mv;
 281
 282    assert(mb);
 283
 284    if (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_MOTION_FORWARD | PIPE_MPEG12_MB_TYPE_MOTION_BACKWARD)) {
 285       switch (mb->macroblock_modes.bits.frame_motion_type) {
 286       case PIPE_MPEG12_MO_TYPE_FRAME:
 287          mv.top.x = mb->PMV[0][vector][0];
 288          mv.top.y = mb->PMV[0][vector][1];
 289          mv.top.field_select = PIPE_VIDEO_FRAME;
 290          mv.top.weight = weight;
 291
 292          mv.bottom.x = mb->PMV[0][vector][0];
 293          mv.bottom.y = mb->PMV[0][vector][1];
 294          mv.bottom.weight = weight;
 295          mv.bottom.field_select = PIPE_VIDEO_FRAME;
 296          break;
 297
 298       case PIPE_MPEG12_MO_TYPE_FIELD:
 299          mv.top.x = mb->PMV[0][vector][0];
 300          mv.top.y = mb->PMV[0][vector][1];
 301          mv.top.field_select = (mb->motion_vertical_field_select & field_select_mask) ?
 302             PIPE_VIDEO_BOTTOM_FIELD : PIPE_VIDEO_TOP_FIELD;
 303          mv.top.weight = weight;
 304
 305          mv.bottom.x = mb->PMV[1][vector][0];
 306          mv.bottom.y = mb->PMV[1][vector][1];
 307          mv.bottom.field_select = (mb->motion_vertical_field_select & (field_select_mask << 2)) ?
 308             PIPE_VIDEO_BOTTOM_FIELD : PIPE_VIDEO_TOP_FIELD;
 309          mv.bottom.weight = weight;
 310          break;
 311
 312       default: // TODO: Support DUALPRIME and 16x8
 313          break;
 314       }
 315    } else {
 316       mv.top.x = mv.top.y = 0;
 317       mv.top.field_select = PIPE_VIDEO_FRAME;
 318       mv.top.weight = weight;
 319
 320       mv.bottom.x = mv.bottom.y = 0;
 321       mv.bottom.field_select = PIPE_VIDEO_FRAME;
 322       mv.bottom.weight = weight;
 323    }
 324    return mv;
 325 }
 326
 327 static INLINE void
 328 UploadYcbcrBlocks(struct vl_mpeg12_decoder *dec,
 329                   struct vl_mpeg12_buffer *buf,
 330                   const struct pipe_mpeg12_macroblock *mb)
 331 {
 332    unsigned intra;
 333    unsigned tb, x, y, num_blocks = 0;
 334
 335    assert(dec && buf);
 336    assert(mb);
 337
 338    if (!mb->coded_block_pattern)
 339       return;
 340
 341    intra = mb->macroblock_type & PIPE_MPEG12_MB_TYPE_INTRA ? 1 : 0;
 342
 343    for (y = 0; y < 2; ++y) {
 344       for (x = 0; x < 2; ++x) {
 345          if (mb->coded_block_pattern & const_empty_block_mask_420[0][y][x]) {
 346
 347             struct vl_ycbcr_block *stream = buf->ycbcr_stream[0];
 348             stream->x = mb->x * 2 + x;
 349             stream->y = mb->y * 2 + y;
 350             stream->intra = intra;
 351             stream->coding = mb->macroblock_modes.bits.dct_type;
 352             stream->block_num = buf->block_num++;
 353
 354             buf->num_ycbcr_blocks[0]++;
 355             buf->ycbcr_stream[0]++;
 356
 357             num_blocks++;
 358          }
 359       }
 360    }
 361
 362    /* TODO: Implement 422, 444 */
 363    //assert(ctx->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 364
 365    for (tb = 1; tb < 3; ++tb) {
 366       if (mb->coded_block_pattern & const_empty_block_mask_420[tb][0][0]) {
 367
 368          struct vl_ycbcr_block *stream = buf->ycbcr_stream[tb];
 369          stream->x = mb->x;
 370          stream->y = mb->y;
 371          stream->intra = intra;
 372          stream->coding = 0;
 373          stream->block_num = buf->block_num++;
 374
 375          buf->num_ycbcr_blocks[tb]++;
 376          buf->ycbcr_stream[tb]++;
 377
 378          num_blocks++;
 379       }
 380    }
 381
 382    memcpy(buf->texels, mb->blocks, 64 * sizeof(short) * num_blocks);
 383    buf->texels += 64 * num_blocks;
 384 }
 385
 386 static void
 387 vl_mpeg12_destroy_buffer(void *buffer)
 388 {
 389    struct vl_mpeg12_buffer *buf = buffer;
 390
 391    assert(buf);
 392
 393    cleanup_zscan_buffer(buf);
 394    cleanup_idct_buffer(buf);
 395    cleanup_mc_buffer(buf);
 396    vl_vb_cleanup(&buf->vertex_stream);
 397
 398    FREE(buf);
 399 }
 400
 401 static void
 402 vl_mpeg12_destroy(struct pipe_video_decoder *decoder)
 403 {
 404    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 405    unsigned i;
 406
 407    assert(decoder);
 408
 409    /* Asserted in softpipe_delete_fs_state() for some reason */
 410    dec->base.context->bind_vs_state(dec->base.context, NULL);
 411    dec->base.context->bind_fs_state(dec->base.context, NULL);
 412
 413    dec->base.context->delete_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 414    dec->base.context->delete_sampler_state(dec->base.context, dec->sampler_ycbcr);
 415
 416    vl_mc_cleanup(&dec->mc_y);
 417    vl_mc_cleanup(&dec->mc_c);
 418    dec->mc_source->destroy(dec->mc_source);
 419
 420    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 421       vl_idct_cleanup(&dec->idct_y);
 422       vl_idct_cleanup(&dec->idct_c);
 423       dec->idct_source->destroy(dec->idct_source);
 424    }
 425
 426    vl_zscan_cleanup(&dec->zscan_y);
 427    vl_zscan_cleanup(&dec->zscan_c);
 428
 429    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 430    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_mv);
 431
 432    pipe_resource_reference(&dec->quads.buffer, NULL);
 433    pipe_resource_reference(&dec->pos.buffer, NULL);
 434
 435    pipe_sampler_view_reference(&dec->zscan_linear, NULL);
 436    pipe_sampler_view_reference(&dec->zscan_normal, NULL);
 437    pipe_sampler_view_reference(&dec->zscan_alternate, NULL);
 438
 439    for (i = 0; i < 4; ++i)
 440       if (dec->dec_buffers[i])
 441          vl_mpeg12_destroy_buffer(dec->dec_buffers[i]);
 442
 443    FREE(dec);
 444 }
 445
 446 static struct vl_mpeg12_buffer *
 447 vl_mpeg12_get_decode_buffer(struct vl_mpeg12_decoder *dec)
 448 {
 449    struct vl_mpeg12_buffer *buffer;
 450
 451    assert(dec);
 452
 453    buffer = vl_video_buffer_get_associated_data(dec->target, &dec->base);
 454    if (buffer)
 455       return buffer;
 456
 457    buffer = dec->dec_buffers[dec->current_buffer];
 458    if (buffer)
 459       return buffer;
 460
 461    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 462    if (buffer == NULL)
 463       return NULL;
 464
 465    if (!vl_vb_init(&buffer->vertex_stream, dec->base.context,
 466                    dec->base.width / MACROBLOCK_WIDTH,
 467                    dec->base.height / MACROBLOCK_HEIGHT))
 468       goto error_vertex_buffer;
 469
 470    if (!init_mc_buffer(dec, buffer))
 471       goto error_mc;
 472
 473    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 474       if (!init_idct_buffer(dec, buffer))
 475          goto error_idct;
 476
 477    if (!init_zscan_buffer(dec, buffer))
 478       goto error_zscan;
 479
 480    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 481       vl_mpg12_bs_init(&buffer->bs, &dec->base);
 482
 483    if (dec->expect_chunked_decode)
 484       vl_video_buffer_set_associated_data(dec->target, &dec->base,
 485                                           buffer, vl_mpeg12_destroy_buffer);
 486    else
 487       dec->dec_buffers[dec->current_buffer] = buffer;
 488
 489    return buffer;
 490
 491 error_zscan:
 492    cleanup_idct_buffer(buffer);
 493
 494 error_idct:
 495    cleanup_mc_buffer(buffer);
 496
 497 error_mc:
 498    vl_vb_cleanup(&buffer->vertex_stream);
 499
 500 error_vertex_buffer:
 501    FREE(buffer);
 502    return NULL;
 503 }
 504
 505 static void
 506 vl_mpeg12_set_picture_parameters(struct pipe_video_decoder *decoder,
 507                                  struct pipe_picture_desc *picture)
 508 {
 509    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 510    struct pipe_mpeg12_picture_desc *pic = (struct pipe_mpeg12_picture_desc *)picture;
 511
 512    assert(dec && pic);
 513
 514    dec->picture_desc = *pic;
 515 }
 516
 517 static void
 518 vl_mpeg12_set_quant_matrix(struct pipe_video_decoder *decoder,
 519                            const struct pipe_quant_matrix *matrix)
 520 {
 521    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 522    const struct pipe_mpeg12_quant_matrix *m = (const struct pipe_mpeg12_quant_matrix *)matrix;
 523
 524    assert(dec);
 525    assert(matrix->codec == PIPE_VIDEO_CODEC_MPEG12);
 526
 527    memcpy(dec->intra_matrix, m->intra_matrix, 64);
 528    memcpy(dec->non_intra_matrix, m->non_intra_matrix, 64);
 529 }
 530
 531 static void
 532 vl_mpeg12_set_decode_target(struct pipe_video_decoder *decoder,
 533                             struct pipe_video_buffer *target)
 534 {
 535    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 536    struct pipe_surface **surfaces;
 537    unsigned i;
 538
 539    assert(dec);
 540
 541    dec->target = target;
 542    surfaces = target->get_surfaces(target);
 543    for (i = 0; i < VL_MAX_PLANES; ++i)
 544       pipe_surface_reference(&dec->target_surfaces[i], surfaces[i]);
 545 }
 546
 547 static void
 548 vl_mpeg12_set_reference_frames(struct pipe_video_decoder *decoder,
 549                                struct pipe_video_buffer **ref_frames,
 550                                unsigned num_ref_frames)
 551 {
 552    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 553    struct pipe_sampler_view **sv;
 554    unsigned i,j;
 555
 556    assert(dec);
 557    assert(num_ref_frames <= VL_MAX_REF_FRAMES);
 558
 559    for (i = 0; i < num_ref_frames; ++i) {
 560       sv = ref_frames[i]->get_sampler_view_planes(ref_frames[i]);
 561       for (j = 0; j < VL_MAX_PLANES; ++j)
 562          pipe_sampler_view_reference(&dec->ref_frames[i][j], sv[j]);
 563    }
 564
 565    for (; i < VL_MAX_REF_FRAMES; ++i)
 566       for (j = 0; j < VL_MAX_PLANES; ++j)
 567          pipe_sampler_view_reference(&dec->ref_frames[i][j], NULL);
 568 }
 569
 570 static void
 571 vl_mpeg12_begin_frame(struct pipe_video_decoder *decoder)
 572 {
 573    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 574    struct vl_mpeg12_buffer *buf;
 575
 576    struct pipe_resource *tex;
 577    struct pipe_box rect = { 0, 0, 0, 1, 1, 1 };
 578
 579    unsigned i;
 580
 581    assert(dec && dec->target);
 582
 583    buf = vl_mpeg12_get_decode_buffer(dec);
 584    assert(buf);
 585
 586    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 587       dec->intra_matrix[0] = 1 << (7 - dec->picture_desc.intra_dc_precision);
 588
 589    for (i = 0; i < VL_MAX_PLANES; ++i) {
 590       struct vl_zscan *zscan = i == 0 ? &dec->zscan_y : &dec->zscan_c;
 591       vl_zscan_upload_quant(zscan, &buf->zscan[i], dec->intra_matrix, true);
 592       vl_zscan_upload_quant(zscan, &buf->zscan[i], dec->non_intra_matrix, false);
 593    }
 594
 595    vl_vb_map(&buf->vertex_stream, dec->base.context);
 596
 597    tex = buf->zscan_source->texture;
 598    rect.width = tex->width0;
 599    rect.height = tex->height0;
 600
 601    buf->tex_transfer = dec->base.context->get_transfer
 602    (
 603       dec->base.context, tex,
 604       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_RANGE,
 605       &rect
 606    );
 607
 608    buf->block_num = 0;
 609    buf->texels = dec->base.context->transfer_map(dec->base.context, buf->tex_transfer);
 610
 611    for (i = 0; i < VL_MAX_PLANES; ++i) {
 612       buf->ycbcr_stream[i] = vl_vb_get_ycbcr_stream(&buf->vertex_stream, i);
 613       buf->num_ycbcr_blocks[i] = 0;
 614    }
 615
 616    for (i = 0; i < VL_MAX_REF_FRAMES; ++i)
 617       buf->mv_stream[i] = vl_vb_get_mv_stream(&buf->vertex_stream, i);
 618
 619    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 620       vl_mpg12_bs_set_picture_desc(&buf->bs, &dec->picture_desc);
 621
 622    } else {
 623
 624       for (i = 0; i < VL_MAX_PLANES; ++i)
 625          vl_zscan_set_layout(&buf->zscan[i], dec->zscan_linear);
 626    }
 627 }
 628
 629 static void
 630 vl_mpeg12_decode_macroblock(struct pipe_video_decoder *decoder,
 631                             const struct pipe_macroblock *macroblocks,
 632                             unsigned num_macroblocks)
 633 {
 634    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 635    const struct pipe_mpeg12_macroblock *mb = (const struct pipe_mpeg12_macroblock *)macroblocks;
 636    struct vl_mpeg12_buffer *buf;
 637
 638    unsigned i, j, mv_weights[2];
 639
 640    assert(dec && dec->target);
 641    assert(macroblocks && macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
 642
 643    buf = vl_mpeg12_get_decode_buffer(dec);
 644    assert(buf);
 645
 646    for (; num_macroblocks > 0; --num_macroblocks) {
 647       unsigned mb_addr = mb->y * dec->width_in_macroblocks + mb->x;
 648
 649       if (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_PATTERN | PIPE_MPEG12_MB_TYPE_INTRA))
 650          UploadYcbcrBlocks(dec, buf, mb);
 651
 652       MacroBlockTypeToPipeWeights(mb, mv_weights);
 653
 654       for (i = 0; i < 2; ++i) {
 655           if (!dec->ref_frames[i][0]) continue;
 656
 657          buf->mv_stream[i][mb_addr] = MotionVectorToPipe
 658          (
 659             mb, i,
 660             i ? PIPE_MPEG12_FS_FIRST_BACKWARD : PIPE_MPEG12_FS_FIRST_FORWARD,
 661             mv_weights[i]
 662          );
 663       }
 664
 665       /* see section 7.6.6 of the spec */
 666       if (mb->num_skipped_macroblocks > 0) {
 667          struct vl_motionvector skipped_mv[2];
 668
 669          if (dec->ref_frames[0][0] && !dec->ref_frames[1][0]) {
 670             skipped_mv[0].top.x = skipped_mv[0].top.y = 0;
 671             skipped_mv[0].top.weight = PIPE_VIDEO_MV_WEIGHT_MAX;
 672          } else {
 673            skipped_mv[0] = buf->mv_stream[0][mb_addr];
 674            skipped_mv[1] = buf->mv_stream[1][mb_addr];
 675          }
 676          skipped_mv[0].top.field_select = PIPE_VIDEO_FRAME;
 677          skipped_mv[1].top.field_select = PIPE_VIDEO_FRAME;
 678
 679          skipped_mv[0].bottom = skipped_mv[0].top;
 680          skipped_mv[1].bottom = skipped_mv[1].top;
 681
 682          ++mb_addr;
 683          for (i = 0; i < mb->num_skipped_macroblocks; ++i, ++mb_addr) {
 684             for (j = 0; j < 2; ++j) {
 685                if (!dec->ref_frames[j][0]) continue;
 686                buf->mv_stream[j][mb_addr] = skipped_mv[j];
 687
 688             }
 689          }
 690       }
 691
 692       ++mb;
 693    }
 694 }
 695
 696 static void
 697 vl_mpeg12_decode_bitstream(struct pipe_video_decoder *decoder,
 698                            unsigned num_buffers,
 699                            const void * const *buffers,
 700                            const unsigned *sizes)
 701 {
 702    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 703    struct vl_mpeg12_buffer *buf;
 704
 705    unsigned i;
 706
 707    assert(dec && dec->target);
 708
 709    buf = vl_mpeg12_get_decode_buffer(dec);
 710    assert(buf);
 711
 712    for (i = 0; i < VL_MAX_PLANES; ++i)
 713       vl_zscan_set_layout(&buf->zscan[i], dec->picture_desc.alternate_scan ?
 714                           dec->zscan_alternate : dec->zscan_normal);
 715
 716    vl_mpg12_bs_decode(&buf->bs, num_buffers, buffers, sizes);
 717 }
 718
 719 static void
 720 vl_mpeg12_end_frame(struct pipe_video_decoder *decoder)
 721 {
 722    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 723    struct pipe_sampler_view **mc_source_sv;
 724    struct pipe_vertex_buffer vb[3];
 725    struct vl_mpeg12_buffer *buf;
 726
 727    unsigned i, j, component;
 728    unsigned nr_components;
 729
 730    assert(dec && dec->target);
 731
 732    buf = vl_mpeg12_get_decode_buffer(dec);
 733
 734    vl_vb_unmap(&buf->vertex_stream, dec->base.context);
 735
 736    dec->base.context->transfer_unmap(dec->base.context, buf->tex_transfer);
 737    dec->base.context->transfer_destroy(dec->base.context, buf->tex_transfer);
 738
 739    vb[0] = dec->quads;
 740    vb[1] = dec->pos;
 741
 742    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_mv);
 743    for (i = 0; i < VL_MAX_PLANES; ++i) {
 744       if (!dec->target_surfaces[i]) continue;
 745
 746       vl_mc_set_surface(&buf->mc[i], dec->target_surfaces[i]);
 747
 748       for (j = 0; j < VL_MAX_REF_FRAMES; ++j) {
 749          if (!dec->ref_frames[j][i]) continue;
 750
 751          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 752          dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 753
 754          vl_mc_render_ref(i ? &dec->mc_c : &dec->mc_y, &buf->mc[i], dec->ref_frames[j][i]);
 755       }
 756    }
 757
 758    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 759    for (i = 0; i < VL_MAX_PLANES; ++i) {
 760       if (!buf->num_ycbcr_blocks[i]) continue;
 761
 762       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 763       dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 764
 765       vl_zscan_render(i ? &dec->zscan_c : & dec->zscan_y, &buf->zscan[i] , buf->num_ycbcr_blocks[i]);
 766
 767       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 768          vl_idct_flush(i ? &dec->idct_c : &dec->idct_y, &buf->idct[i], buf->num_ycbcr_blocks[i]);
 769    }
 770
 771    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 772    for (i = 0, component = 0; i < VL_MAX_PLANES; ++i) {
 773       if (!dec->target_surfaces[i]) continue;
 774
 775       nr_components = util_format_get_nr_components(dec->target_surfaces[i]->texture->format);
 776       for (j = 0; j < nr_components; ++j, ++component) {
 777          if (!buf->num_ycbcr_blocks[i]) continue;
 778
 779          vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, component);
 780          dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 781
 782          if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 783             vl_idct_prepare_stage2(i ? &dec->idct_c : &dec->idct_y, &buf->idct[component]);
 784          else {
 785             dec->base.context->set_fragment_sampler_views(dec->base.context, 1, &mc_source_sv[component]);
 786             dec->base.context->bind_fragment_sampler_states(dec->base.context, 1, &dec->sampler_ycbcr);
 787          }
 788          vl_mc_render_ycbcr(i ? &dec->mc_c : &dec->mc_y, &buf->mc[i], j, buf->num_ycbcr_blocks[component]);
 789       }
 790    }
 791    ++dec->current_buffer;
 792    dec->current_buffer %= 4;
 793 }
 794
 795 static void
 796 vl_mpeg12_flush(struct pipe_video_decoder *decoder)
 797 {
 798    assert(decoder);
 799
 800    //Noop, for shaders it is much faster to flush everything in end_frame
 801 }
 802
 803 static bool
 804 init_pipe_state(struct vl_mpeg12_decoder *dec)
 805 {
 806    struct pipe_depth_stencil_alpha_state dsa;
 807    struct pipe_sampler_state sampler;
 808    unsigned i;
 809
 810    assert(dec);
 811
 812    memset(&dsa, 0, sizeof dsa);
 813    dsa.depth.enabled = 0;
 814    dsa.depth.writemask = 0;
 815    dsa.depth.func = PIPE_FUNC_ALWAYS;
 816    for (i = 0; i < 2; ++i) {
 817       dsa.stencil[i].enabled = 0;
 818       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 819       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 820       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 821       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 822       dsa.stencil[i].valuemask = 0;
 823       dsa.stencil[i].writemask = 0;
 824    }
 825    dsa.alpha.enabled = 0;
 826    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 827    dsa.alpha.ref_value = 0;
 828    dec->dsa = dec->base.context->create_depth_stencil_alpha_state(dec->base.context, &dsa);
 829    dec->base.context->bind_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 830
 831    memset(&sampler, 0, sizeof(sampler));
 832    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 833    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 834    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 835    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 836    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 837    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 838    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 839    sampler.compare_func = PIPE_FUNC_ALWAYS;
 840    sampler.normalized_coords = 1;
 841    dec->sampler_ycbcr = dec->base.context->create_sampler_state(dec->base.context, &sampler);
 842    if (!dec->sampler_ycbcr)
 843       return false;
 844
 845    return true;
 846 }
 847
 848 static const struct format_config*
 849 find_format_config(struct vl_mpeg12_decoder *dec, const struct format_config configs[], unsigned num_configs)
 850 {
 851    struct pipe_screen *screen;
 852    unsigned i;
 853
 854    assert(dec);
 855
 856    screen = dec->base.context->screen;
 857
 858    for (i = 0; i < num_configs; ++i) {
 859       if (!screen->is_format_supported(screen, configs[i].zscan_source_format, PIPE_TEXTURE_2D,
 860                                        1, PIPE_BIND_SAMPLER_VIEW))
 861          continue;
 862
 863       if (configs[i].idct_source_format != PIPE_FORMAT_NONE) {
 864          if (!screen->is_format_supported(screen, configs[i].idct_source_format, PIPE_TEXTURE_2D,
 865                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 866             continue;
 867
 868          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_3D,
 869                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 870             continue;
 871       } else {
 872          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_2D,
 873                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 874             continue;
 875       }
 876       return &configs[i];
 877    }
 878
 879    return NULL;
 880 }
 881
 882 static bool
 883 init_zscan(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 884 {
 885    unsigned num_channels;
 886
 887    assert(dec);
 888
 889    dec->zscan_source_format = format_config->zscan_source_format;
 890    dec->zscan_linear = vl_zscan_layout(dec->base.context, vl_zscan_linear, dec->blocks_per_line);
 891    dec->zscan_normal = vl_zscan_layout(dec->base.context, vl_zscan_normal, dec->blocks_per_line);
 892    dec->zscan_alternate = vl_zscan_layout(dec->base.context, vl_zscan_alternate, dec->blocks_per_line);
 893
 894    num_channels = dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT ? 4 : 1;
 895
 896    if (!vl_zscan_init(&dec->zscan_y, dec->base.context, dec->base.width, dec->base.height,
 897                       dec->blocks_per_line, dec->num_blocks, num_channels))
 898       return false;
 899
 900    if (!vl_zscan_init(&dec->zscan_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 901                       dec->blocks_per_line, dec->num_blocks, num_channels))
 902       return false;
 903
 904    return true;
 905 }
 906
 907 static bool
 908 init_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 909 {
 910    unsigned nr_of_idct_render_targets, max_inst;
 911    enum pipe_format formats[3];
 912
 913    struct pipe_sampler_view *matrix = NULL;
 914
 915    nr_of_idct_render_targets = dec->base.context->screen->get_param
 916    (
 917       dec->base.context->screen, PIPE_CAP_MAX_RENDER_TARGETS
 918    );
 919
 920    max_inst = dec->base.context->screen->get_shader_param
 921    (
 922       dec->base.context->screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_INSTRUCTIONS
 923    );
 924
 925    // Just assume we need 32 inst per render target, not 100% true, but should work in most cases
 926    if (nr_of_idct_render_targets >= 4 && max_inst >= 32*4)
 927       // more than 4 render targets usually doesn't makes any seens
 928       nr_of_idct_render_targets = 4;
 929    else
 930       nr_of_idct_render_targets = 1;
 931
 932    formats[0] = formats[1] = formats[2] = format_config->idct_source_format;
 933    dec->idct_source = vl_video_buffer_create_ex
 934    (
 935       dec->base.context, dec->base.width / 4, dec->base.height, 1,
 936       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 937    );
 938
 939    if (!dec->idct_source)
 940       goto error_idct_source;
 941
 942    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 943    dec->mc_source = vl_video_buffer_create_ex
 944    (
 945       dec->base.context, dec->base.width / nr_of_idct_render_targets,
 946       dec->base.height / 4, nr_of_idct_render_targets,
 947       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 948    );
 949
 950    if (!dec->mc_source)
 951       goto error_mc_source;
 952
 953    if (!(matrix = vl_idct_upload_matrix(dec->base.context, format_config->idct_scale)))
 954       goto error_matrix;
 955
 956    if (!vl_idct_init(&dec->idct_y, dec->base.context, dec->base.width, dec->base.height,
 957                      nr_of_idct_render_targets, matrix, matrix))
 958       goto error_y;
 959
 960    if(!vl_idct_init(&dec->idct_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 961                     nr_of_idct_render_targets, matrix, matrix))
 962       goto error_c;
 963
 964    pipe_sampler_view_reference(&matrix, NULL);
 965
 966    return true;
 967
 968 error_c:
 969    vl_idct_cleanup(&dec->idct_y);
 970
 971 error_y:
 972    pipe_sampler_view_reference(&matrix, NULL);
 973
 974 error_matrix:
 975    dec->mc_source->destroy(dec->mc_source);
 976
 977 error_mc_source:
 978    dec->idct_source->destroy(dec->idct_source);
 979
 980 error_idct_source:
 981    return false;
 982 }
 983
 984 static bool
 985 init_mc_source_widthout_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 986 {
 987    enum pipe_format formats[3];
 988
 989    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 990    dec->mc_source = vl_video_buffer_create_ex
 991    (
 992       dec->base.context, dec->base.width, dec->base.height, 1,
 993       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 994    );
 995
 996    return dec->mc_source != NULL;
 997 }
 998
 999 static void
1000 mc_vert_shader_callback(void *priv, struct vl_mc *mc,
1001                         struct ureg_program *shader,
1002                         unsigned first_output,
1003                         struct ureg_dst tex)
1004 {
1005    struct vl_mpeg12_decoder *dec = priv;
1006    struct ureg_dst o_vtex;
1007
1008    assert(priv && mc);
1009    assert(shader);
1010
1011    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1012       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1013       vl_idct_stage2_vert_shader(idct, shader, first_output, tex);
1014    } else {
1015       o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output);
1016       ureg_MOV(shader, ureg_writemask(o_vtex, TGSI_WRITEMASK_XY), ureg_src(tex));
1017    }
1018 }
1019
1020 static void
1021 mc_frag_shader_callback(void *priv, struct vl_mc *mc,
1022                         struct ureg_program *shader,
1023                         unsigned first_input,
1024                         struct ureg_dst dst)
1025 {
1026    struct vl_mpeg12_decoder *dec = priv;
1027    struct ureg_src src, sampler;
1028
1029    assert(priv && mc);
1030    assert(shader);
1031
1032    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1033       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1034       vl_idct_stage2_frag_shader(idct, shader, first_input, dst);
1035    } else {
1036       src = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input, TGSI_INTERPOLATE_LINEAR);
1037       sampler = ureg_DECL_sampler(shader, 0);
1038       ureg_TEX(shader, dst, TGSI_TEXTURE_2D, src, sampler);
1039    }
1040 }
1041
1042 struct pipe_video_decoder *
1043 vl_create_mpeg12_decoder(struct pipe_context *context,
1044                          enum pipe_video_profile profile,
1045                          enum pipe_video_entrypoint entrypoint,
1046                          enum pipe_video_chroma_format chroma_format,
1047                          unsigned width, unsigned height, unsigned max_references,
1048                          bool expect_chunked_decode)
1049 {
1050    const unsigned block_size_pixels = BLOCK_WIDTH * BLOCK_HEIGHT;
1051    const struct format_config *format_config;
1052    struct vl_mpeg12_decoder *dec;
1053
1054    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
1055
1056    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
1057
1058    if (!dec)
1059       return NULL;
1060
1061    dec->base.context = context;
1062    dec->base.profile = profile;
1063    dec->base.entrypoint = entrypoint;
1064    dec->base.chroma_format = chroma_format;
1065    dec->base.width = width;
1066    dec->base.height = height;
1067    dec->base.max_references = max_references;
1068
1069    dec->base.destroy = vl_mpeg12_destroy;
1070    dec->base.set_picture_parameters = vl_mpeg12_set_picture_parameters;
1071    dec->base.set_quant_matrix = vl_mpeg12_set_quant_matrix;
1072    dec->base.set_decode_target = vl_mpeg12_set_decode_target;
1073    dec->base.set_reference_frames = vl_mpeg12_set_reference_frames;
1074    dec->base.begin_frame = vl_mpeg12_begin_frame;
1075    dec->base.decode_macroblock = vl_mpeg12_decode_macroblock;
1076    dec->base.decode_bitstream = vl_mpeg12_decode_bitstream;
1077    dec->base.end_frame = vl_mpeg12_end_frame;
1078    dec->base.flush = vl_mpeg12_flush;
1079
1080    dec->blocks_per_line = MAX2(util_next_power_of_two(dec->base.width) / block_size_pixels, 4);
1081    dec->num_blocks = (dec->base.width * dec->base.height) / block_size_pixels;
1082    dec->width_in_macroblocks = align(dec->base.width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
1083    dec->expect_chunked_decode = expect_chunked_decode;
1084
1085    /* TODO: Implement 422, 444 */
1086    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
1087
1088    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
1089       dec->chroma_width = dec->base.width / 2;
1090       dec->chroma_height = dec->base.height / 2;
1091       dec->num_blocks = dec->num_blocks * 2;
1092    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
1093       dec->chroma_width = dec->base.width;
1094       dec->chroma_height = dec->base.height / 2;
1095       dec->num_blocks = dec->num_blocks * 2 + dec->num_blocks;
1096    } else {
1097       dec->chroma_width = dec->base.width;
1098       dec->chroma_height = dec->base.height;
1099       dec->num_blocks = dec->num_blocks * 3;
1100    }
1101
1102    dec->quads = vl_vb_upload_quads(dec->base.context);
1103    dec->pos = vl_vb_upload_pos(
1104       dec->base.context,
1105       dec->base.width / MACROBLOCK_WIDTH,
1106       dec->base.height / MACROBLOCK_HEIGHT
1107    );
1108
1109    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->base.context);
1110    dec->ves_mv = vl_vb_get_ves_mv(dec->base.context);
1111
1112    switch (entrypoint) {
1113    case PIPE_VIDEO_ENTRYPOINT_BITSTREAM:
1114       format_config = find_format_config(dec, bitstream_format_config, num_bitstream_format_configs);
1115       break;
1116
1117    case PIPE_VIDEO_ENTRYPOINT_IDCT:
1118       format_config = find_format_config(dec, idct_format_config, num_idct_format_configs);
1119       break;
1120
1121    case PIPE_VIDEO_ENTRYPOINT_MC:
1122       format_config = find_format_config(dec, mc_format_config, num_mc_format_configs);
1123       break;
1124
1125    default:
1126       assert(0);
1127       FREE(dec);
1128       return NULL;
1129    }
1130
1131    if (!format_config) {
1132       FREE(dec);
1133       return NULL;
1134    }
1135
1136    if (!init_zscan(dec, format_config))
1137       goto error_zscan;
1138
1139    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1140       if (!init_idct(dec, format_config))
1141          goto error_sources;
1142    } else {
1143       if (!init_mc_source_widthout_idct(dec, format_config))
1144          goto error_sources;
1145    }
1146
1147    if (!vl_mc_init(&dec->mc_y, dec->base.context, dec->base.width, dec->base.height,
1148                    MACROBLOCK_HEIGHT, format_config->mc_scale,
1149                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1150       goto error_mc_y;
1151
1152    // TODO
1153    if (!vl_mc_init(&dec->mc_c, dec->base.context, dec->base.width, dec->base.height,
1154                    BLOCK_HEIGHT, format_config->mc_scale,
1155                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1156       goto error_mc_c;
1157
1158    if (!init_pipe_state(dec))
1159       goto error_pipe_state;
1160
1161    memset(dec->intra_matrix, 0x10, 64);
1162    memset(dec->non_intra_matrix, 0x10, 64);
1163
1164    return &dec->base;
1165
1166 error_pipe_state:
1167    vl_mc_cleanup(&dec->mc_c);
1168
1169 error_mc_c:
1170    vl_mc_cleanup(&dec->mc_y);
1171
1172 error_mc_y:
1173    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1174       vl_idct_cleanup(&dec->idct_y);
1175       vl_idct_cleanup(&dec->idct_c);
1176       dec->idct_source->destroy(dec->idct_source);
1177    }
1178    dec->mc_source->destroy(dec->mc_source);
1179
1180 error_sources:
1181    vl_zscan_cleanup(&dec->zscan_y);
1182    vl_zscan_cleanup(&dec->zscan_c);
1183
1184 error_zscan:
1185    FREE(dec);
1186    return NULL;
1187 }