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(struct pipe_video_decoder *decoder)
 388 {
 389    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 390
 391    assert(decoder);
 392
 393    /* Asserted in softpipe_delete_fs_state() for some reason */
 394    dec->base.context->bind_vs_state(dec->base.context, NULL);
 395    dec->base.context->bind_fs_state(dec->base.context, NULL);
 396
 397    dec->base.context->delete_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 398    dec->base.context->delete_sampler_state(dec->base.context, dec->sampler_ycbcr);
 399
 400    vl_mc_cleanup(&dec->mc_y);
 401    vl_mc_cleanup(&dec->mc_c);
 402    dec->mc_source->destroy(dec->mc_source);
 403
 404    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 405       vl_idct_cleanup(&dec->idct_y);
 406       vl_idct_cleanup(&dec->idct_c);
 407       dec->idct_source->destroy(dec->idct_source);
 408    }
 409
 410    vl_zscan_cleanup(&dec->zscan_y);
 411    vl_zscan_cleanup(&dec->zscan_c);
 412
 413    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 414    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_mv);
 415
 416    pipe_resource_reference(&dec->quads.buffer, NULL);
 417    pipe_resource_reference(&dec->pos.buffer, NULL);
 418
 419    pipe_sampler_view_reference(&dec->zscan_linear, NULL);
 420    pipe_sampler_view_reference(&dec->zscan_normal, NULL);
 421    pipe_sampler_view_reference(&dec->zscan_alternate, NULL);
 422
 423    FREE(dec);
 424 }
 425
 426 static void *
 427 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 428 {
 429    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 430    struct vl_mpeg12_buffer *buffer;
 431
 432    assert(dec);
 433
 434    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 435    if (buffer == NULL)
 436       return NULL;
 437
 438    if (!vl_vb_init(&buffer->vertex_stream, dec->base.context,
 439                    dec->base.width / MACROBLOCK_WIDTH,
 440                    dec->base.height / MACROBLOCK_HEIGHT))
 441       goto error_vertex_buffer;
 442
 443    if (!init_mc_buffer(dec, buffer))
 444       goto error_mc;
 445
 446    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 447       if (!init_idct_buffer(dec, buffer))
 448          goto error_idct;
 449
 450    if (!init_zscan_buffer(dec, buffer))
 451       goto error_zscan;
 452
 453    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 454       vl_mpg12_bs_init(&buffer->bs, decoder);
 455
 456    return buffer;
 457
 458 error_zscan:
 459    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 460       cleanup_idct_buffer(buffer);
 461
 462 error_idct:
 463    cleanup_mc_buffer(buffer);
 464
 465 error_mc:
 466    vl_vb_cleanup(&buffer->vertex_stream);
 467
 468 error_vertex_buffer:
 469    FREE(buffer);
 470    return NULL;
 471 }
 472
 473 static void
 474 vl_mpeg12_destroy_buffer(struct pipe_video_decoder *decoder, void *buffer)
 475 {
 476    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 477    struct vl_mpeg12_buffer *buf = buffer;
 478
 479    assert(dec && buf);
 480
 481    cleanup_zscan_buffer(buf);
 482
 483    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 484       cleanup_idct_buffer(buf);
 485
 486    cleanup_mc_buffer(buf);
 487
 488    vl_vb_cleanup(&buf->vertex_stream);
 489
 490    FREE(buf);
 491 }
 492
 493 static void
 494 vl_mpeg12_set_decode_buffer(struct pipe_video_decoder *decoder, void *buffer)
 495 {
 496    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 497
 498    assert(dec && buffer);
 499
 500    dec->current_buffer = buffer;
 501 }
 502
 503 static void
 504 vl_mpeg12_set_picture_parameters(struct pipe_video_decoder *decoder,
 505                                  struct pipe_picture_desc *picture)
 506 {
 507    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 508    struct pipe_mpeg12_picture_desc *pic = (struct pipe_mpeg12_picture_desc *)picture;
 509
 510    assert(dec && pic);
 511
 512    dec->picture_desc = *pic;
 513 }
 514
 515 static void
 516 vl_mpeg12_set_quant_matrix(struct pipe_video_decoder *decoder,
 517                            const struct pipe_quant_matrix *matrix)
 518 {
 519    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 520    const struct pipe_mpeg12_quant_matrix *m = (const struct pipe_mpeg12_quant_matrix *)matrix;
 521
 522    assert(dec);
 523    assert(matrix->codec == PIPE_VIDEO_CODEC_MPEG12);
 524
 525    memcpy(dec->intra_matrix, m->intra_matrix, 64);
 526    memcpy(dec->non_intra_matrix, m->non_intra_matrix, 64);
 527 }
 528
 529 static void
 530 vl_mpeg12_set_decode_target(struct pipe_video_decoder *decoder,
 531                             struct pipe_video_buffer *target)
 532 {
 533    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 534    struct pipe_surface **surfaces;
 535    unsigned i;
 536
 537    assert(dec);
 538
 539    surfaces = target->get_surfaces(target);
 540    for (i = 0; i < VL_MAX_PLANES; ++i)
 541       pipe_surface_reference(&dec->target_surfaces[i], surfaces[i]);
 542 }
 543
 544 static void
 545 vl_mpeg12_set_reference_frames(struct pipe_video_decoder *decoder,
 546                                struct pipe_video_buffer **ref_frames,
 547                                unsigned num_ref_frames)
 548 {
 549    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 550    struct pipe_sampler_view **sv;
 551    unsigned i,j;
 552
 553    assert(dec);
 554    assert(num_ref_frames <= VL_MAX_REF_FRAMES);
 555
 556    for (i = 0; i < num_ref_frames; ++i) {
 557       sv = ref_frames[i]->get_sampler_view_planes(ref_frames[i]);
 558       for (j = 0; j < VL_MAX_PLANES; ++j)
 559          pipe_sampler_view_reference(&dec->ref_frames[i][j], sv[j]);
 560    }
 561
 562    for (; i < VL_MAX_REF_FRAMES; ++i)
 563       for (j = 0; j < VL_MAX_PLANES; ++j)
 564          pipe_sampler_view_reference(&dec->ref_frames[i][j], NULL);
 565 }
 566
 567 static void
 568 vl_mpeg12_begin_frame(struct pipe_video_decoder *decoder)
 569 {
 570    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 571    struct vl_mpeg12_buffer *buf;
 572
 573    struct pipe_resource *tex;
 574    struct pipe_box rect = { 0, 0, 0, 1, 1, 1 };
 575
 576    unsigned i;
 577
 578    assert(dec);
 579
 580    buf = dec->current_buffer;
 581    assert(buf);
 582
 583    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 584       dec->intra_matrix[0] = 1 << (7 - dec->picture_desc.intra_dc_precision);
 585
 586    for (i = 0; i < VL_MAX_PLANES; ++i) {
 587       struct vl_zscan *zscan = i == 0 ? &dec->zscan_y : &dec->zscan_c;
 588       vl_zscan_upload_quant(zscan, &buf->zscan[i], dec->intra_matrix, true);
 589       vl_zscan_upload_quant(zscan, &buf->zscan[i], dec->non_intra_matrix, false);
 590    }
 591
 592    vl_vb_map(&buf->vertex_stream, dec->base.context);
 593
 594    tex = buf->zscan_source->texture;
 595    rect.width = tex->width0;
 596    rect.height = tex->height0;
 597
 598    buf->tex_transfer = dec->base.context->get_transfer
 599    (
 600       dec->base.context, tex,
 601       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 602       &rect
 603    );
 604
 605    buf->block_num = 0;
 606    buf->texels = dec->base.context->transfer_map(dec->base.context, buf->tex_transfer);
 607
 608    for (i = 0; i < VL_MAX_PLANES; ++i) {
 609       buf->ycbcr_stream[i] = vl_vb_get_ycbcr_stream(&buf->vertex_stream, i);
 610       buf->num_ycbcr_blocks[i] = 0;
 611    }
 612
 613    for (i = 0; i < VL_MAX_REF_FRAMES; ++i)
 614       buf->mv_stream[i] = vl_vb_get_mv_stream(&buf->vertex_stream, i);
 615
 616    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 617       vl_mpg12_bs_set_picture_desc(&buf->bs, &dec->picture_desc);
 618
 619    } else {
 620
 621       for (i = 0; i < VL_MAX_PLANES; ++i)
 622          vl_zscan_set_layout(&buf->zscan[i], dec->zscan_linear);
 623    }
 624 }
 625
 626 static void
 627 vl_mpeg12_decode_macroblock(struct pipe_video_decoder *decoder,
 628                             const struct pipe_macroblock *macroblocks,
 629                             unsigned num_macroblocks)
 630 {
 631    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 632    const struct pipe_mpeg12_macroblock *mb = (const struct pipe_mpeg12_macroblock *)macroblocks;
 633    struct vl_mpeg12_buffer *buf;
 634
 635    unsigned i, j, mv_weights[2];
 636
 637    assert(dec && dec->current_buffer);
 638    assert(macroblocks && macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
 639
 640    buf = dec->current_buffer;
 641    assert(buf);
 642
 643    for (; num_macroblocks > 0; --num_macroblocks) {
 644       unsigned mb_addr = mb->y * dec->width_in_macroblocks + mb->x;
 645
 646       if (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_PATTERN | PIPE_MPEG12_MB_TYPE_INTRA))
 647          UploadYcbcrBlocks(dec, buf, mb);
 648
 649       MacroBlockTypeToPipeWeights(mb, mv_weights);
 650
 651       for (i = 0; i < 2; ++i) {
 652           if (!dec->ref_frames[i][0]) continue;
 653
 654          buf->mv_stream[i][mb_addr] = MotionVectorToPipe
 655          (
 656             mb, i,
 657             i ? PIPE_MPEG12_FS_FIRST_BACKWARD : PIPE_MPEG12_FS_FIRST_FORWARD,
 658             mv_weights[i]
 659          );
 660       }
 661
 662       /* see section 7.6.6 of the spec */
 663       if (mb->num_skipped_macroblocks > 0) {
 664          struct vl_motionvector skipped_mv[2];
 665
 666          if (dec->ref_frames[0][0] && !dec->ref_frames[1][0]) {
 667             skipped_mv[0].top.x = skipped_mv[0].top.y = 0;
 668             skipped_mv[0].top.weight = PIPE_VIDEO_MV_WEIGHT_MAX;
 669          } else {
 670            skipped_mv[0] = buf->mv_stream[0][mb_addr];
 671            skipped_mv[1] = buf->mv_stream[1][mb_addr];
 672          }
 673          skipped_mv[0].top.field_select = PIPE_VIDEO_FRAME;
 674          skipped_mv[1].top.field_select = PIPE_VIDEO_FRAME;
 675
 676          skipped_mv[0].bottom = skipped_mv[0].top;
 677          skipped_mv[1].bottom = skipped_mv[1].top;
 678
 679          ++mb_addr;
 680          for (i = 0; i < mb->num_skipped_macroblocks; ++i, ++mb_addr) {
 681             for (j = 0; j < 2; ++j) {
 682                if (!dec->ref_frames[j][0]) continue;
 683                buf->mv_stream[j][mb_addr] = skipped_mv[j];
 684
 685             }
 686          }
 687       }
 688
 689       ++mb;
 690    }
 691 }
 692
 693 static void
 694 vl_mpeg12_decode_bitstream(struct pipe_video_decoder *decoder,
 695                            unsigned num_buffers,
 696                            const void * const *buffers,
 697                            const unsigned *sizes)
 698 {
 699    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 700    struct vl_mpeg12_buffer *buf;
 701
 702    unsigned i;
 703
 704    assert(dec && dec->current_buffer);
 705
 706    buf = dec->current_buffer;
 707    assert(buf);
 708
 709    for (i = 0; i < VL_MAX_PLANES; ++i)
 710       vl_zscan_set_layout(&buf->zscan[i], dec->picture_desc.alternate_scan ?
 711                           dec->zscan_alternate : dec->zscan_normal);
 712
 713    vl_mpg12_bs_decode(&buf->bs, num_buffers, buffers, sizes);
 714 }
 715
 716 static void
 717 vl_mpeg12_end_frame(struct pipe_video_decoder *decoder)
 718 {
 719    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 720    struct pipe_sampler_view **mc_source_sv;
 721    struct pipe_vertex_buffer vb[3];
 722    struct vl_mpeg12_buffer *buf;
 723
 724    unsigned i, j, component;
 725    unsigned nr_components;
 726
 727    assert(dec && dec->current_buffer);
 728
 729    buf = dec->current_buffer;
 730
 731    vl_vb_unmap(&buf->vertex_stream, dec->base.context);
 732
 733    dec->base.context->transfer_unmap(dec->base.context, buf->tex_transfer);
 734    dec->base.context->transfer_destroy(dec->base.context, buf->tex_transfer);
 735
 736    vb[0] = dec->quads;
 737    vb[1] = dec->pos;
 738
 739    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_mv);
 740    for (i = 0; i < VL_MAX_PLANES; ++i) {
 741       if (!dec->target_surfaces[i]) continue;
 742
 743       vl_mc_set_surface(&buf->mc[i], dec->target_surfaces[i]);
 744
 745       for (j = 0; j < VL_MAX_REF_FRAMES; ++j) {
 746          if (!dec->ref_frames[j][i]) continue;
 747
 748          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 749          dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 750
 751          vl_mc_render_ref(i ? &dec->mc_c : &dec->mc_y, &buf->mc[i], dec->ref_frames[j][i]);
 752       }
 753    }
 754
 755    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 756    for (i = 0; i < VL_MAX_PLANES; ++i) {
 757       if (!buf->num_ycbcr_blocks[i]) continue;
 758
 759       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 760       dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 761
 762       vl_zscan_render(i ? &dec->zscan_c : & dec->zscan_y, &buf->zscan[i] , buf->num_ycbcr_blocks[i]);
 763
 764       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 765          vl_idct_flush(i ? &dec->idct_c : &dec->idct_y, &buf->idct[i], buf->num_ycbcr_blocks[i]);
 766    }
 767
 768    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 769    for (i = 0, component = 0; i < VL_MAX_PLANES; ++i) {
 770       if (!dec->target_surfaces[i]) continue;
 771
 772       nr_components = util_format_get_nr_components(dec->target_surfaces[i]->texture->format);
 773       for (j = 0; j < nr_components; ++j, ++component) {
 774          if (!buf->num_ycbcr_blocks[i]) continue;
 775
 776          vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, component);
 777          dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 778
 779          if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 780             vl_idct_prepare_stage2(i ? &dec->idct_c : &dec->idct_y, &buf->idct[component]);
 781          else {
 782             dec->base.context->set_fragment_sampler_views(dec->base.context, 1, &mc_source_sv[component]);
 783             dec->base.context->bind_fragment_sampler_states(dec->base.context, 1, &dec->sampler_ycbcr);
 784          }
 785          vl_mc_render_ycbcr(i ? &dec->mc_c : &dec->mc_y, &buf->mc[i], j, buf->num_ycbcr_blocks[component]);
 786       }
 787    }
 788 }
 789
 790 static void
 791 vl_mpeg12_flush(struct pipe_video_decoder *decoder)
 792 {
 793    assert(decoder);
 794
 795    //Noop, for shaders it is much faster to flush everything in end_frame
 796 }
 797
 798 static bool
 799 init_pipe_state(struct vl_mpeg12_decoder *dec)
 800 {
 801    struct pipe_depth_stencil_alpha_state dsa;
 802    struct pipe_sampler_state sampler;
 803    unsigned i;
 804
 805    assert(dec);
 806
 807    memset(&dsa, 0, sizeof dsa);
 808    dsa.depth.enabled = 0;
 809    dsa.depth.writemask = 0;
 810    dsa.depth.func = PIPE_FUNC_ALWAYS;
 811    for (i = 0; i < 2; ++i) {
 812       dsa.stencil[i].enabled = 0;
 813       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 814       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 815       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 816       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 817       dsa.stencil[i].valuemask = 0;
 818       dsa.stencil[i].writemask = 0;
 819    }
 820    dsa.alpha.enabled = 0;
 821    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 822    dsa.alpha.ref_value = 0;
 823    dec->dsa = dec->base.context->create_depth_stencil_alpha_state(dec->base.context, &dsa);
 824    dec->base.context->bind_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 825
 826    memset(&sampler, 0, sizeof(sampler));
 827    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 828    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 829    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 830    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 831    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 832    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 833    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 834    sampler.compare_func = PIPE_FUNC_ALWAYS;
 835    sampler.normalized_coords = 1;
 836    dec->sampler_ycbcr = dec->base.context->create_sampler_state(dec->base.context, &sampler);
 837    if (!dec->sampler_ycbcr)
 838       return false;
 839
 840    return true;
 841 }
 842
 843 static const struct format_config*
 844 find_format_config(struct vl_mpeg12_decoder *dec, const struct format_config configs[], unsigned num_configs)
 845 {
 846    struct pipe_screen *screen;
 847    unsigned i;
 848
 849    assert(dec);
 850
 851    screen = dec->base.context->screen;
 852
 853    for (i = 0; i < num_configs; ++i) {
 854       if (!screen->is_format_supported(screen, configs[i].zscan_source_format, PIPE_TEXTURE_2D,
 855                                        1, PIPE_BIND_SAMPLER_VIEW))
 856          continue;
 857
 858       if (configs[i].idct_source_format != PIPE_FORMAT_NONE) {
 859          if (!screen->is_format_supported(screen, configs[i].idct_source_format, PIPE_TEXTURE_2D,
 860                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 861             continue;
 862
 863          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_3D,
 864                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 865             continue;
 866       } else {
 867          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_2D,
 868                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 869             continue;
 870       }
 871       return &configs[i];
 872    }
 873
 874    return NULL;
 875 }
 876
 877 static bool
 878 init_zscan(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 879 {
 880    unsigned num_channels;
 881
 882    assert(dec);
 883
 884    dec->zscan_source_format = format_config->zscan_source_format;
 885    dec->zscan_linear = vl_zscan_layout(dec->base.context, vl_zscan_linear, dec->blocks_per_line);
 886    dec->zscan_normal = vl_zscan_layout(dec->base.context, vl_zscan_normal, dec->blocks_per_line);
 887    dec->zscan_alternate = vl_zscan_layout(dec->base.context, vl_zscan_alternate, dec->blocks_per_line);
 888
 889    num_channels = dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT ? 4 : 1;
 890
 891    if (!vl_zscan_init(&dec->zscan_y, dec->base.context, dec->base.width, dec->base.height,
 892                       dec->blocks_per_line, dec->num_blocks, num_channels))
 893       return false;
 894
 895    if (!vl_zscan_init(&dec->zscan_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 896                       dec->blocks_per_line, dec->num_blocks, num_channels))
 897       return false;
 898
 899    return true;
 900 }
 901
 902 static bool
 903 init_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 904 {
 905    unsigned nr_of_idct_render_targets, max_inst;
 906    enum pipe_format formats[3];
 907
 908    struct pipe_sampler_view *matrix = NULL;
 909
 910    nr_of_idct_render_targets = dec->base.context->screen->get_param
 911    (
 912       dec->base.context->screen, PIPE_CAP_MAX_RENDER_TARGETS
 913    );
 914
 915    max_inst = dec->base.context->screen->get_shader_param
 916    (
 917       dec->base.context->screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_INSTRUCTIONS
 918    );
 919
 920    // Just assume we need 32 inst per render target, not 100% true, but should work in most cases
 921    if (nr_of_idct_render_targets >= 4 && max_inst >= 32*4)
 922       // more than 4 render targets usually doesn't makes any seens
 923       nr_of_idct_render_targets = 4;
 924    else
 925       nr_of_idct_render_targets = 1;
 926
 927    formats[0] = formats[1] = formats[2] = format_config->idct_source_format;
 928    dec->idct_source = vl_video_buffer_create_ex
 929    (
 930       dec->base.context, dec->base.width / 4, dec->base.height, 1,
 931       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 932    );
 933
 934    if (!dec->idct_source)
 935       goto error_idct_source;
 936
 937    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 938    dec->mc_source = vl_video_buffer_create_ex
 939    (
 940       dec->base.context, dec->base.width / nr_of_idct_render_targets,
 941       dec->base.height / 4, nr_of_idct_render_targets,
 942       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 943    );
 944
 945    if (!dec->mc_source)
 946       goto error_mc_source;
 947
 948    if (!(matrix = vl_idct_upload_matrix(dec->base.context, format_config->idct_scale)))
 949       goto error_matrix;
 950
 951    if (!vl_idct_init(&dec->idct_y, dec->base.context, dec->base.width, dec->base.height,
 952                      nr_of_idct_render_targets, matrix, matrix))
 953       goto error_y;
 954
 955    if(!vl_idct_init(&dec->idct_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 956                     nr_of_idct_render_targets, matrix, matrix))
 957       goto error_c;
 958
 959    pipe_sampler_view_reference(&matrix, NULL);
 960
 961    return true;
 962
 963 error_c:
 964    vl_idct_cleanup(&dec->idct_y);
 965
 966 error_y:
 967    pipe_sampler_view_reference(&matrix, NULL);
 968
 969 error_matrix:
 970    dec->mc_source->destroy(dec->mc_source);
 971
 972 error_mc_source:
 973    dec->idct_source->destroy(dec->idct_source);
 974
 975 error_idct_source:
 976    return false;
 977 }
 978
 979 static bool
 980 init_mc_source_widthout_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 981 {
 982    enum pipe_format formats[3];
 983
 984    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 985    dec->mc_source = vl_video_buffer_create_ex
 986    (
 987       dec->base.context, dec->base.width, dec->base.height, 1,
 988       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 989    );
 990
 991    return dec->mc_source != NULL;
 992 }
 993
 994 static void
 995 mc_vert_shader_callback(void *priv, struct vl_mc *mc,
 996                         struct ureg_program *shader,
 997                         unsigned first_output,
 998                         struct ureg_dst tex)
 999 {
1000    struct vl_mpeg12_decoder *dec = priv;
1001    struct ureg_dst o_vtex;
1002
1003    assert(priv && mc);
1004    assert(shader);
1005
1006    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1007       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1008       vl_idct_stage2_vert_shader(idct, shader, first_output, tex);
1009    } else {
1010       o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output);
1011       ureg_MOV(shader, ureg_writemask(o_vtex, TGSI_WRITEMASK_XY), ureg_src(tex));
1012    }
1013 }
1014
1015 static void
1016 mc_frag_shader_callback(void *priv, struct vl_mc *mc,
1017                         struct ureg_program *shader,
1018                         unsigned first_input,
1019                         struct ureg_dst dst)
1020 {
1021    struct vl_mpeg12_decoder *dec = priv;
1022    struct ureg_src src, sampler;
1023
1024    assert(priv && mc);
1025    assert(shader);
1026
1027    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1028       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1029       vl_idct_stage2_frag_shader(idct, shader, first_input, dst);
1030    } else {
1031       src = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input, TGSI_INTERPOLATE_LINEAR);
1032       sampler = ureg_DECL_sampler(shader, 0);
1033       ureg_TEX(shader, dst, TGSI_TEXTURE_2D, src, sampler);
1034    }
1035 }
1036
1037 struct pipe_video_decoder *
1038 vl_create_mpeg12_decoder(struct pipe_context *context,
1039                          enum pipe_video_profile profile,
1040                          enum pipe_video_entrypoint entrypoint,
1041                          enum pipe_video_chroma_format chroma_format,
1042                          unsigned width, unsigned height, unsigned max_references)
1043 {
1044    const unsigned block_size_pixels = BLOCK_WIDTH * BLOCK_HEIGHT;
1045    const struct format_config *format_config;
1046    struct vl_mpeg12_decoder *dec;
1047
1048    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
1049
1050    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
1051
1052    if (!dec)
1053       return NULL;
1054
1055    dec->base.context = context;
1056    dec->base.profile = profile;
1057    dec->base.entrypoint = entrypoint;
1058    dec->base.chroma_format = chroma_format;
1059    dec->base.width = width;
1060    dec->base.height = height;
1061    dec->base.max_references = max_references;
1062
1063    dec->base.destroy = vl_mpeg12_destroy;
1064    dec->base.create_buffer = vl_mpeg12_create_buffer;
1065    dec->base.destroy_buffer = vl_mpeg12_destroy_buffer;
1066    dec->base.set_decode_buffer = vl_mpeg12_set_decode_buffer;
1067    dec->base.set_picture_parameters = vl_mpeg12_set_picture_parameters;
1068    dec->base.set_quant_matrix = vl_mpeg12_set_quant_matrix;
1069    dec->base.set_decode_target = vl_mpeg12_set_decode_target;
1070    dec->base.set_reference_frames = vl_mpeg12_set_reference_frames;
1071    dec->base.begin_frame = vl_mpeg12_begin_frame;
1072    dec->base.decode_macroblock = vl_mpeg12_decode_macroblock;
1073    dec->base.decode_bitstream = vl_mpeg12_decode_bitstream;
1074    dec->base.end_frame = vl_mpeg12_end_frame;
1075    dec->base.flush = vl_mpeg12_flush;
1076
1077    dec->blocks_per_line = MAX2(util_next_power_of_two(dec->base.width) / block_size_pixels, 4);
1078    dec->num_blocks = (dec->base.width * dec->base.height) / block_size_pixels;
1079    dec->width_in_macroblocks = align(dec->base.width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
1080
1081    /* TODO: Implement 422, 444 */
1082    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
1083
1084    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
1085       dec->chroma_width = dec->base.width / 2;
1086       dec->chroma_height = dec->base.height / 2;
1087       dec->num_blocks = dec->num_blocks * 2;
1088    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
1089       dec->chroma_width = dec->base.width;
1090       dec->chroma_height = dec->base.height / 2;
1091       dec->num_blocks = dec->num_blocks * 2 + dec->num_blocks;
1092    } else {
1093       dec->chroma_width = dec->base.width;
1094       dec->chroma_height = dec->base.height;
1095       dec->num_blocks = dec->num_blocks * 3;
1096    }
1097
1098    dec->quads = vl_vb_upload_quads(dec->base.context);
1099    dec->pos = vl_vb_upload_pos(
1100       dec->base.context,
1101       dec->base.width / MACROBLOCK_WIDTH,
1102       dec->base.height / MACROBLOCK_HEIGHT
1103    );
1104
1105    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->base.context);
1106    dec->ves_mv = vl_vb_get_ves_mv(dec->base.context);
1107
1108    switch (entrypoint) {
1109    case PIPE_VIDEO_ENTRYPOINT_BITSTREAM:
1110       format_config = find_format_config(dec, bitstream_format_config, num_bitstream_format_configs);
1111       break;
1112
1113    case PIPE_VIDEO_ENTRYPOINT_IDCT:
1114       format_config = find_format_config(dec, idct_format_config, num_idct_format_configs);
1115       break;
1116
1117    case PIPE_VIDEO_ENTRYPOINT_MC:
1118       format_config = find_format_config(dec, mc_format_config, num_mc_format_configs);
1119       break;
1120
1121    default:
1122       assert(0);
1123       FREE(dec);
1124       return NULL;
1125    }
1126
1127    if (!format_config) {
1128       FREE(dec);
1129       return NULL;
1130    }
1131
1132    if (!init_zscan(dec, format_config))
1133       goto error_zscan;
1134
1135    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1136       if (!init_idct(dec, format_config))
1137          goto error_sources;
1138    } else {
1139       if (!init_mc_source_widthout_idct(dec, format_config))
1140          goto error_sources;
1141    }
1142
1143    if (!vl_mc_init(&dec->mc_y, dec->base.context, dec->base.width, dec->base.height,
1144                    MACROBLOCK_HEIGHT, format_config->mc_scale,
1145                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1146       goto error_mc_y;
1147
1148    // TODO
1149    if (!vl_mc_init(&dec->mc_c, dec->base.context, dec->base.width, dec->base.height,
1150                    BLOCK_HEIGHT, format_config->mc_scale,
1151                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1152       goto error_mc_c;
1153
1154    if (!init_pipe_state(dec))
1155       goto error_pipe_state;
1156
1157    memset(dec->intra_matrix, 0x10, 64);
1158    memset(dec->non_intra_matrix, 0x10, 64);
1159
1160    return &dec->base;
1161
1162 error_pipe_state:
1163    vl_mc_cleanup(&dec->mc_c);
1164
1165 error_mc_c:
1166    vl_mc_cleanup(&dec->mc_y);
1167
1168 error_mc_y:
1169    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1170       vl_idct_cleanup(&dec->idct_y);
1171       vl_idct_cleanup(&dec->idct_c);
1172       dec->idct_source->destroy(dec->idct_source);
1173    }
1174    dec->mc_source->destroy(dec->mc_source);
1175
1176 error_sources:
1177    vl_zscan_cleanup(&dec->zscan_y);
1178    vl_zscan_cleanup(&dec->zscan_c);
1179
1180 error_zscan:
1181    FREE(dec);
1182    return NULL;
1183 }