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_PATTERN) {
 264          /* patern without a motion vector, just copy the old frame content */
 265          weights[0] = PIPE_VIDEO_MV_WEIGHT_MAX;
 266          weights[1] = PIPE_VIDEO_MV_WEIGHT_MIN;
 267       } else {
 268          weights[0] = PIPE_VIDEO_MV_WEIGHT_MIN;
 269          weights[1] = PIPE_VIDEO_MV_WEIGHT_MIN;
 270       }
 271       break;
 272    }
 273 }
 274
 275 static INLINE struct vl_motionvector
 276 MotionVectorToPipe(const struct pipe_mpeg12_macroblock *mb, unsigned vector,
 277                    unsigned field_select_mask, unsigned weight)
 278 {
 279    struct vl_motionvector mv;
 280
 281    assert(mb);
 282
 283    if (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_MOTION_FORWARD | PIPE_MPEG12_MB_TYPE_MOTION_BACKWARD)) {
 284       switch (mb->macroblock_modes.bits.frame_motion_type) {
 285       case PIPE_MPEG12_MO_TYPE_FRAME:
 286          mv.top.x = mb->PMV[0][vector][0];
 287          mv.top.y = mb->PMV[0][vector][1];
 288          mv.top.field_select = PIPE_VIDEO_FRAME;
 289          mv.top.weight = weight;
 290
 291          mv.bottom.x = mb->PMV[0][vector][0];
 292          mv.bottom.y = mb->PMV[0][vector][1];
 293          mv.bottom.weight = weight;
 294          mv.bottom.field_select = PIPE_VIDEO_FRAME;
 295          break;
 296
 297       case PIPE_MPEG12_MO_TYPE_FIELD:
 298          mv.top.x = mb->PMV[0][vector][0];
 299          mv.top.y = mb->PMV[0][vector][1];
 300          mv.top.field_select = (mb->motion_vertical_field_select & field_select_mask) ?
 301             PIPE_VIDEO_BOTTOM_FIELD : PIPE_VIDEO_TOP_FIELD;
 302          mv.top.weight = weight;
 303
 304          mv.bottom.x = mb->PMV[1][vector][0];
 305          mv.bottom.y = mb->PMV[1][vector][1];
 306          mv.bottom.field_select = (mb->motion_vertical_field_select & (field_select_mask << 2)) ?
 307             PIPE_VIDEO_BOTTOM_FIELD : PIPE_VIDEO_TOP_FIELD;
 308          mv.bottom.weight = weight;
 309          break;
 310
 311       default: // TODO: Support DUALPRIME and 16x8
 312          break;
 313       }
 314    } else {
 315       mv.top.x = mv.top.y = 0;
 316       mv.top.field_select = PIPE_VIDEO_FRAME;
 317       mv.top.weight = weight;
 318
 319       mv.bottom.x = mv.bottom.y = 0;
 320       mv.bottom.field_select = PIPE_VIDEO_FRAME;
 321       mv.bottom.weight = weight;
 322    }
 323    return mv;
 324 }
 325
 326 static INLINE void
 327 UploadYcbcrBlocks(struct vl_mpeg12_decoder *dec,
 328                   struct vl_mpeg12_buffer *buf,
 329                   const struct pipe_mpeg12_macroblock *mb)
 330 {
 331    unsigned intra;
 332    unsigned tb, x, y, num_blocks = 0;
 333
 334    assert(dec && buf);
 335    assert(mb);
 336
 337    if (!mb->coded_block_pattern)
 338       return;
 339
 340    intra = mb->macroblock_type & PIPE_MPEG12_MB_TYPE_INTRA ? 1 : 0;
 341
 342    for (y = 0; y < 2; ++y) {
 343       for (x = 0; x < 2; ++x) {
 344          if (mb->coded_block_pattern & const_empty_block_mask_420[0][y][x]) {
 345
 346             struct vl_ycbcr_block *stream = buf->ycbcr_stream[0];
 347             stream->x = mb->x * 2 + x;
 348             stream->y = mb->y * 2 + y;
 349             stream->intra = intra;
 350             stream->coding = mb->macroblock_modes.bits.dct_type;
 351             stream->block_num = buf->block_num++;
 352
 353             buf->num_ycbcr_blocks[0]++;
 354             buf->ycbcr_stream[0]++;
 355
 356             num_blocks++;
 357          }
 358       }
 359    }
 360
 361    /* TODO: Implement 422, 444 */
 362    //assert(ctx->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 363
 364    for (tb = 1; tb < 3; ++tb) {
 365       if (mb->coded_block_pattern & const_empty_block_mask_420[tb][0][0]) {
 366
 367          struct vl_ycbcr_block *stream = buf->ycbcr_stream[tb];
 368          stream->x = mb->x;
 369          stream->y = mb->y;
 370          stream->intra = intra;
 371          stream->coding = 0;
 372          stream->block_num = buf->block_num++;
 373
 374          buf->num_ycbcr_blocks[tb]++;
 375          buf->ycbcr_stream[tb]++;
 376
 377          num_blocks++;
 378       }
 379    }
 380
 381    memcpy(buf->texels, mb->blocks, 64 * sizeof(short) * num_blocks);
 382    buf->texels += 64 * num_blocks;
 383 }
 384
 385 static void
 386 vl_mpeg12_destroy(struct pipe_video_decoder *decoder)
 387 {
 388    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 389
 390    assert(decoder);
 391
 392    /* Asserted in softpipe_delete_fs_state() for some reason */
 393    dec->base.context->bind_vs_state(dec->base.context, NULL);
 394    dec->base.context->bind_fs_state(dec->base.context, NULL);
 395
 396    dec->base.context->delete_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 397    dec->base.context->delete_sampler_state(dec->base.context, dec->sampler_ycbcr);
 398
 399    vl_mc_cleanup(&dec->mc_y);
 400    vl_mc_cleanup(&dec->mc_c);
 401    dec->mc_source->destroy(dec->mc_source);
 402
 403    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 404       vl_idct_cleanup(&dec->idct_y);
 405       vl_idct_cleanup(&dec->idct_c);
 406       dec->idct_source->destroy(dec->idct_source);
 407    }
 408
 409    vl_zscan_cleanup(&dec->zscan_y);
 410    vl_zscan_cleanup(&dec->zscan_c);
 411
 412    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 413    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_mv);
 414
 415    pipe_resource_reference(&dec->quads.buffer, NULL);
 416    pipe_resource_reference(&dec->pos.buffer, NULL);
 417
 418    pipe_sampler_view_reference(&dec->zscan_linear, NULL);
 419    pipe_sampler_view_reference(&dec->zscan_normal, NULL);
 420    pipe_sampler_view_reference(&dec->zscan_alternate, NULL);
 421
 422    FREE(dec);
 423 }
 424
 425 static void *
 426 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 427 {
 428    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 429    struct vl_mpeg12_buffer *buffer;
 430
 431    assert(dec);
 432
 433    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 434    if (buffer == NULL)
 435       return NULL;
 436
 437    if (!vl_vb_init(&buffer->vertex_stream, dec->base.context,
 438                    dec->base.width / MACROBLOCK_WIDTH,
 439                    dec->base.height / MACROBLOCK_HEIGHT))
 440       goto error_vertex_buffer;
 441
 442    if (!init_mc_buffer(dec, buffer))
 443       goto error_mc;
 444
 445    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 446       if (!init_idct_buffer(dec, buffer))
 447          goto error_idct;
 448
 449    if (!init_zscan_buffer(dec, buffer))
 450       goto error_zscan;
 451
 452    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 453       vl_mpg12_bs_init(&buffer->bs, decoder);
 454
 455    return buffer;
 456
 457 error_zscan:
 458    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 459       cleanup_idct_buffer(buffer);
 460
 461 error_idct:
 462    cleanup_mc_buffer(buffer);
 463
 464 error_mc:
 465    vl_vb_cleanup(&buffer->vertex_stream);
 466
 467 error_vertex_buffer:
 468    FREE(buffer);
 469    return NULL;
 470 }
 471
 472 static void
 473 vl_mpeg12_destroy_buffer(struct pipe_video_decoder *decoder, void *buffer)
 474 {
 475    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 476    struct vl_mpeg12_buffer *buf = buffer;
 477
 478    assert(dec && buf);
 479
 480    cleanup_zscan_buffer(buf);
 481
 482    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 483       cleanup_idct_buffer(buf);
 484
 485    cleanup_mc_buffer(buf);
 486
 487    vl_vb_cleanup(&buf->vertex_stream);
 488
 489    FREE(buf);
 490 }
 491
 492 static void
 493 vl_mpeg12_set_decode_buffer(struct pipe_video_decoder *decoder, void *buffer)
 494 {
 495    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 496
 497    assert(dec && buffer);
 498
 499    dec->current_buffer = buffer;
 500 }
 501
 502 static void
 503 vl_mpeg12_set_picture_parameters(struct pipe_video_decoder *decoder,
 504                                  struct pipe_picture_desc *picture)
 505 {
 506    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 507    struct pipe_mpeg12_picture_desc *pic = (struct pipe_mpeg12_picture_desc *)picture;
 508
 509    assert(dec && pic);
 510
 511    dec->picture_desc = *pic;
 512 }
 513
 514 static void
 515 vl_mpeg12_set_quant_matrix(struct pipe_video_decoder *decoder,
 516                            const struct pipe_quant_matrix *matrix)
 517 {
 518    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 519    const struct pipe_mpeg12_quant_matrix *m = (const struct pipe_mpeg12_quant_matrix *)matrix;
 520
 521    assert(dec);
 522    assert(matrix->codec == PIPE_VIDEO_CODEC_MPEG12);
 523
 524    memcpy(dec->intra_matrix, m->intra_matrix, 64);
 525    memcpy(dec->non_intra_matrix, m->non_intra_matrix, 64);
 526 }
 527
 528 static void
 529 vl_mpeg12_set_decode_target(struct pipe_video_decoder *decoder,
 530                             struct pipe_video_buffer *target)
 531 {
 532    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 533    struct pipe_surface **surfaces;
 534    unsigned i;
 535
 536    assert(dec);
 537
 538    surfaces = target->get_surfaces(target);
 539    for (i = 0; i < VL_MAX_PLANES; ++i)
 540       pipe_surface_reference(&dec->target_surfaces[i], surfaces[i]);
 541 }
 542
 543 static void
 544 vl_mpeg12_set_reference_frames(struct pipe_video_decoder *decoder,
 545                                struct pipe_video_buffer **ref_frames,
 546                                unsigned num_ref_frames)
 547 {
 548    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 549    struct pipe_sampler_view **sv;
 550    unsigned i,j;
 551
 552    assert(dec);
 553    assert(num_ref_frames <= VL_MAX_REF_FRAMES);
 554
 555    for (i = 0; i < num_ref_frames; ++i) {
 556       sv = ref_frames[i]->get_sampler_view_planes(ref_frames[i]);
 557       for (j = 0; j < VL_MAX_PLANES; ++j)
 558          pipe_sampler_view_reference(&dec->ref_frames[i][j], sv[j]);
 559    }
 560
 561    for (; i < VL_MAX_REF_FRAMES; ++i)
 562       for (j = 0; j < VL_MAX_PLANES; ++j)
 563          pipe_sampler_view_reference(&dec->ref_frames[i][j], NULL);
 564 }
 565
 566 static void
 567 vl_mpeg12_begin_frame(struct pipe_video_decoder *decoder)
 568 {
 569    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 570    struct vl_mpeg12_buffer *buf;
 571
 572    struct pipe_resource *tex;
 573    struct pipe_box rect = { 0, 0, 0, 1, 1, 1 };
 574
 575    unsigned i;
 576
 577    assert(dec);
 578
 579    buf = dec->current_buffer;
 580    assert(buf);
 581
 582    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 583       dec->intra_matrix[0] = 1 << (7 - dec->picture_desc.intra_dc_precision);
 584
 585    for (i = 0; i < VL_MAX_PLANES; ++i) {
 586       vl_zscan_upload_quant(&buf->zscan[i], dec->intra_matrix, true);
 587       vl_zscan_upload_quant(&buf->zscan[i], dec->non_intra_matrix, false);
 588    }
 589
 590    vl_vb_map(&buf->vertex_stream, dec->base.context);
 591
 592    tex = buf->zscan_source->texture;
 593    rect.width = tex->width0;
 594    rect.height = tex->height0;
 595
 596    buf->tex_transfer = dec->base.context->get_transfer
 597    (
 598       dec->base.context, tex,
 599       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 600       &rect
 601    );
 602
 603    buf->block_num = 0;
 604    buf->texels = dec->base.context->transfer_map(dec->base.context, buf->tex_transfer);
 605
 606    for (i = 0; i < VL_MAX_PLANES; ++i) {
 607       buf->ycbcr_stream[i] = vl_vb_get_ycbcr_stream(&buf->vertex_stream, i);
 608       buf->num_ycbcr_blocks[i] = 0;
 609    }
 610
 611    for (i = 0; i < VL_MAX_REF_FRAMES; ++i)
 612       buf->mv_stream[i] = vl_vb_get_mv_stream(&buf->vertex_stream, i);
 613
 614    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 615       vl_mpg12_bs_set_picture_desc(&buf->bs, &dec->picture_desc);
 616
 617    } else {
 618
 619       for (i = 0; i < VL_MAX_PLANES; ++i)
 620          vl_zscan_set_layout(&buf->zscan[i], dec->zscan_linear);
 621    }
 622 }
 623
 624 static void
 625 vl_mpeg12_decode_macroblock(struct pipe_video_decoder *decoder,
 626                             const struct pipe_macroblock *macroblocks,
 627                             unsigned num_macroblocks)
 628 {
 629    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 630    const struct pipe_mpeg12_macroblock *mb = (const struct pipe_mpeg12_macroblock *)macroblocks;
 631    struct vl_mpeg12_buffer *buf;
 632
 633    unsigned i, j, mv_weights[2];
 634
 635    assert(dec && dec->current_buffer);
 636    assert(macroblocks && macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
 637
 638    buf = dec->current_buffer;
 639    assert(buf);
 640
 641    for (; num_macroblocks > 0; --num_macroblocks) {
 642       unsigned mb_addr = mb->y * dec->width_in_macroblocks + mb->x;
 643
 644       if (mb->macroblock_type & (PIPE_MPEG12_MB_TYPE_PATTERN | PIPE_MPEG12_MB_TYPE_INTRA))
 645          UploadYcbcrBlocks(dec, buf, mb);
 646
 647       MacroBlockTypeToPipeWeights(mb, mv_weights);
 648
 649       for (i = 0; i < 2; ++i) {
 650           if (!dec->ref_frames[i][0]) continue;
 651
 652          buf->mv_stream[i][mb_addr] = MotionVectorToPipe
 653          (
 654             mb, i,
 655             i ? PIPE_MPEG12_FS_FIRST_BACKWARD : PIPE_MPEG12_FS_FIRST_FORWARD,
 656             mv_weights[i]
 657          );
 658       }
 659
 660       /* see section 7.6.6 of the spec */
 661       if (mb->num_skipped_macroblocks > 0) {
 662          struct vl_motionvector skipped_mv[2];
 663
 664          if (dec->ref_frames[0][0] && !dec->ref_frames[1][0]) {
 665             skipped_mv[0].top.x = skipped_mv[0].top.y = 0;
 666             skipped_mv[0].top.weight = PIPE_VIDEO_MV_WEIGHT_MAX;
 667          } else {
 668            skipped_mv[0] = buf->mv_stream[0][mb_addr];
 669            skipped_mv[1] = buf->mv_stream[1][mb_addr];
 670          }
 671          skipped_mv[0].top.field_select = PIPE_VIDEO_FRAME;
 672          skipped_mv[1].top.field_select = PIPE_VIDEO_FRAME;
 673
 674          skipped_mv[0].bottom = skipped_mv[0].top;
 675          skipped_mv[1].bottom = skipped_mv[1].top;
 676
 677          ++mb_addr;
 678          for (i = 0; i < mb->num_skipped_macroblocks; ++i, ++mb_addr) {
 679             for (j = 0; j < 2; ++j) {
 680                if (!dec->ref_frames[j][0]) continue;
 681                buf->mv_stream[j][mb_addr] = skipped_mv[j];
 682
 683             }
 684          }
 685       }
 686
 687       ++mb;
 688    }
 689 }
 690
 691 static void
 692 vl_mpeg12_decode_bitstream(struct pipe_video_decoder *decoder,
 693                            unsigned num_bytes, const void *data)
 694 {
 695    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 696    struct vl_mpeg12_buffer *buf;
 697
 698    unsigned i;
 699
 700    assert(dec && dec->current_buffer);
 701
 702    buf = dec->current_buffer;
 703    assert(buf);
 704
 705    for (i = 0; i < VL_MAX_PLANES; ++i)
 706       vl_zscan_set_layout(&buf->zscan[i], dec->picture_desc.alternate_scan ?
 707                           dec->zscan_alternate : dec->zscan_normal);
 708
 709    vl_mpg12_bs_decode(&buf->bs, num_bytes, data);
 710 }
 711
 712 static void
 713 vl_mpeg12_end_frame(struct pipe_video_decoder *decoder)
 714 {
 715    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 716    struct pipe_sampler_view **mc_source_sv;
 717    struct pipe_vertex_buffer vb[3];
 718    struct vl_mpeg12_buffer *buf;
 719
 720    unsigned i, j, component;
 721    unsigned nr_components;
 722
 723    assert(dec && dec->current_buffer);
 724
 725    buf = dec->current_buffer;
 726
 727    vl_vb_unmap(&buf->vertex_stream, dec->base.context);
 728
 729    dec->base.context->transfer_unmap(dec->base.context, buf->tex_transfer);
 730    dec->base.context->transfer_destroy(dec->base.context, buf->tex_transfer);
 731
 732    vb[0] = dec->quads;
 733    vb[1] = dec->pos;
 734
 735    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_mv);
 736    for (i = 0; i < VL_MAX_PLANES; ++i) {
 737       if (!dec->target_surfaces[i]) continue;
 738
 739       vl_mc_set_surface(&buf->mc[i], dec->target_surfaces[i]);
 740
 741       for (j = 0; j < VL_MAX_REF_FRAMES; ++j) {
 742          if (!dec->ref_frames[j][i]) continue;
 743
 744          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 745          dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 746
 747          vl_mc_render_ref(&buf->mc[i], dec->ref_frames[j][i]);
 748       }
 749    }
 750
 751    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 752    for (i = 0; i < VL_MAX_PLANES; ++i) {
 753       if (!buf->num_ycbcr_blocks[i]) continue;
 754
 755       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 756       dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 757
 758       vl_zscan_render(&buf->zscan[i] , buf->num_ycbcr_blocks[i]);
 759
 760       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 761          vl_idct_flush(&buf->idct[i], buf->num_ycbcr_blocks[i]);
 762    }
 763
 764    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 765    for (i = 0, component = 0; i < VL_MAX_PLANES; ++i) {
 766       if (!dec->target_surfaces[i]) continue;
 767
 768       nr_components = util_format_get_nr_components(dec->target_surfaces[i]->texture->format);
 769       for (j = 0; j < nr_components; ++j, ++component) {
 770          if (!buf->num_ycbcr_blocks[i]) continue;
 771
 772          vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, component);
 773          dec->base.context->set_vertex_buffers(dec->base.context, 2, vb);
 774
 775          if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 776             vl_idct_prepare_stage2(&buf->idct[component]);
 777          else {
 778             dec->base.context->set_fragment_sampler_views(dec->base.context, 1, &mc_source_sv[component]);
 779             dec->base.context->bind_fragment_sampler_states(dec->base.context, 1, &dec->sampler_ycbcr);
 780          }
 781          vl_mc_render_ycbcr(&buf->mc[i], j, buf->num_ycbcr_blocks[component]);
 782       }
 783    }
 784 }
 785
 786 static void
 787 vl_mpeg12_flush(struct pipe_video_decoder *decoder)
 788 {
 789    assert(decoder);
 790
 791    //Noop, for shaders it is much faster to flush everything in end_frame
 792 }
 793
 794 static bool
 795 init_pipe_state(struct vl_mpeg12_decoder *dec)
 796 {
 797    struct pipe_depth_stencil_alpha_state dsa;
 798    struct pipe_sampler_state sampler;
 799    unsigned i;
 800
 801    assert(dec);
 802
 803    memset(&dsa, 0, sizeof dsa);
 804    dsa.depth.enabled = 0;
 805    dsa.depth.writemask = 0;
 806    dsa.depth.func = PIPE_FUNC_ALWAYS;
 807    for (i = 0; i < 2; ++i) {
 808       dsa.stencil[i].enabled = 0;
 809       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 810       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 811       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 812       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 813       dsa.stencil[i].valuemask = 0;
 814       dsa.stencil[i].writemask = 0;
 815    }
 816    dsa.alpha.enabled = 0;
 817    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 818    dsa.alpha.ref_value = 0;
 819    dec->dsa = dec->base.context->create_depth_stencil_alpha_state(dec->base.context, &dsa);
 820    dec->base.context->bind_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 821
 822    memset(&sampler, 0, sizeof(sampler));
 823    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 824    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 825    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 826    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 827    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 828    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 829    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 830    sampler.compare_func = PIPE_FUNC_ALWAYS;
 831    sampler.normalized_coords = 1;
 832    dec->sampler_ycbcr = dec->base.context->create_sampler_state(dec->base.context, &sampler);
 833    if (!dec->sampler_ycbcr)
 834       return false;
 835
 836    return true;
 837 }
 838
 839 static const struct format_config*
 840 find_format_config(struct vl_mpeg12_decoder *dec, const struct format_config configs[], unsigned num_configs)
 841 {
 842    struct pipe_screen *screen;
 843    unsigned i;
 844
 845    assert(dec);
 846
 847    screen = dec->base.context->screen;
 848
 849    for (i = 0; i < num_configs; ++i) {
 850       if (!screen->is_format_supported(screen, configs[i].zscan_source_format, PIPE_TEXTURE_2D,
 851                                        1, PIPE_BIND_SAMPLER_VIEW))
 852          continue;
 853
 854       if (configs[i].idct_source_format != PIPE_FORMAT_NONE) {
 855          if (!screen->is_format_supported(screen, configs[i].idct_source_format, PIPE_TEXTURE_2D,
 856                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 857             continue;
 858
 859          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_3D,
 860                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 861             continue;
 862       } else {
 863          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_2D,
 864                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 865             continue;
 866       }
 867       return &configs[i];
 868    }
 869
 870    return NULL;
 871 }
 872
 873 static bool
 874 init_zscan(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 875 {
 876    unsigned num_channels;
 877
 878    assert(dec);
 879
 880    dec->zscan_source_format = format_config->zscan_source_format;
 881    dec->zscan_linear = vl_zscan_layout(dec->base.context, vl_zscan_linear, dec->blocks_per_line);
 882    dec->zscan_normal = vl_zscan_layout(dec->base.context, vl_zscan_normal, dec->blocks_per_line);
 883    dec->zscan_alternate = vl_zscan_layout(dec->base.context, vl_zscan_alternate, dec->blocks_per_line);
 884
 885    num_channels = dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT ? 4 : 1;
 886
 887    if (!vl_zscan_init(&dec->zscan_y, dec->base.context, dec->base.width, dec->base.height,
 888                       dec->blocks_per_line, dec->num_blocks, num_channels))
 889       return false;
 890
 891    if (!vl_zscan_init(&dec->zscan_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 892                       dec->blocks_per_line, dec->num_blocks, num_channels))
 893       return false;
 894
 895    return true;
 896 }
 897
 898 static bool
 899 init_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 900 {
 901    unsigned nr_of_idct_render_targets, max_inst;
 902    enum pipe_format formats[3];
 903
 904    struct pipe_sampler_view *matrix = NULL;
 905
 906    nr_of_idct_render_targets = dec->base.context->screen->get_param
 907    (
 908       dec->base.context->screen, PIPE_CAP_MAX_RENDER_TARGETS
 909    );
 910
 911    max_inst = dec->base.context->screen->get_shader_param
 912    (
 913       dec->base.context->screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_INSTRUCTIONS
 914    );
 915
 916    // Just assume we need 32 inst per render target, not 100% true, but should work in most cases
 917    if (nr_of_idct_render_targets >= 4 && max_inst >= 32*4)
 918       // more than 4 render targets usually doesn't makes any seens
 919       nr_of_idct_render_targets = 4;
 920    else
 921       nr_of_idct_render_targets = 1;
 922
 923    formats[0] = formats[1] = formats[2] = format_config->idct_source_format;
 924    dec->idct_source = vl_video_buffer_create_ex
 925    (
 926       dec->base.context, dec->base.width / 4, dec->base.height, 1,
 927       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 928    );
 929
 930    if (!dec->idct_source)
 931       goto error_idct_source;
 932
 933    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 934    dec->mc_source = vl_video_buffer_create_ex
 935    (
 936       dec->base.context, dec->base.width / nr_of_idct_render_targets,
 937       dec->base.height / 4, nr_of_idct_render_targets,
 938       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 939    );
 940
 941    if (!dec->mc_source)
 942       goto error_mc_source;
 943
 944    if (!(matrix = vl_idct_upload_matrix(dec->base.context, format_config->idct_scale)))
 945       goto error_matrix;
 946
 947    if (!vl_idct_init(&dec->idct_y, dec->base.context, dec->base.width, dec->base.height,
 948                      nr_of_idct_render_targets, matrix, matrix))
 949       goto error_y;
 950
 951    if(!vl_idct_init(&dec->idct_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 952                     nr_of_idct_render_targets, matrix, matrix))
 953       goto error_c;
 954
 955    pipe_sampler_view_reference(&matrix, NULL);
 956
 957    return true;
 958
 959 error_c:
 960    vl_idct_cleanup(&dec->idct_y);
 961
 962 error_y:
 963    pipe_sampler_view_reference(&matrix, NULL);
 964
 965 error_matrix:
 966    dec->mc_source->destroy(dec->mc_source);
 967
 968 error_mc_source:
 969    dec->idct_source->destroy(dec->idct_source);
 970
 971 error_idct_source:
 972    return false;
 973 }
 974
 975 static bool
 976 init_mc_source_widthout_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 977 {
 978    enum pipe_format formats[3];
 979
 980    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 981    dec->mc_source = vl_video_buffer_create_ex
 982    (
 983       dec->base.context, dec->base.width, dec->base.height, 1,
 984       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 985    );
 986
 987    return dec->mc_source != NULL;
 988 }
 989
 990 static void
 991 mc_vert_shader_callback(void *priv, struct vl_mc *mc,
 992                         struct ureg_program *shader,
 993                         unsigned first_output,
 994                         struct ureg_dst tex)
 995 {
 996    struct vl_mpeg12_decoder *dec = priv;
 997    struct ureg_dst o_vtex;
 998
 999    assert(priv && mc);
1000    assert(shader);
1001
1002    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1003       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1004       vl_idct_stage2_vert_shader(idct, shader, first_output, tex);
1005    } else {
1006       o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output);
1007       ureg_MOV(shader, ureg_writemask(o_vtex, TGSI_WRITEMASK_XY), ureg_src(tex));
1008    }
1009 }
1010
1011 static void
1012 mc_frag_shader_callback(void *priv, struct vl_mc *mc,
1013                         struct ureg_program *shader,
1014                         unsigned first_input,
1015                         struct ureg_dst dst)
1016 {
1017    struct vl_mpeg12_decoder *dec = priv;
1018    struct ureg_src src, sampler;
1019
1020    assert(priv && mc);
1021    assert(shader);
1022
1023    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1024       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
1025       vl_idct_stage2_frag_shader(idct, shader, first_input, dst);
1026    } else {
1027       src = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input, TGSI_INTERPOLATE_LINEAR);
1028       sampler = ureg_DECL_sampler(shader, 0);
1029       ureg_TEX(shader, dst, TGSI_TEXTURE_2D, src, sampler);
1030    }
1031 }
1032
1033 struct pipe_video_decoder *
1034 vl_create_mpeg12_decoder(struct pipe_context *context,
1035                          enum pipe_video_profile profile,
1036                          enum pipe_video_entrypoint entrypoint,
1037                          enum pipe_video_chroma_format chroma_format,
1038                          unsigned width, unsigned height, unsigned max_references)
1039 {
1040    const unsigned block_size_pixels = BLOCK_WIDTH * BLOCK_HEIGHT;
1041    const struct format_config *format_config;
1042    struct vl_mpeg12_decoder *dec;
1043
1044    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
1045
1046    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
1047
1048    if (!dec)
1049       return NULL;
1050
1051    dec->base.context = context;
1052    dec->base.profile = profile;
1053    dec->base.entrypoint = entrypoint;
1054    dec->base.chroma_format = chroma_format;
1055    dec->base.width = width;
1056    dec->base.height = height;
1057    dec->base.max_references = max_references;
1058
1059    dec->base.destroy = vl_mpeg12_destroy;
1060    dec->base.create_buffer = vl_mpeg12_create_buffer;
1061    dec->base.destroy_buffer = vl_mpeg12_destroy_buffer;
1062    dec->base.set_decode_buffer = vl_mpeg12_set_decode_buffer;
1063    dec->base.set_picture_parameters = vl_mpeg12_set_picture_parameters;
1064    dec->base.set_quant_matrix = vl_mpeg12_set_quant_matrix;
1065    dec->base.set_decode_target = vl_mpeg12_set_decode_target;
1066    dec->base.set_reference_frames = vl_mpeg12_set_reference_frames;
1067    dec->base.begin_frame = vl_mpeg12_begin_frame;
1068    dec->base.decode_macroblock = vl_mpeg12_decode_macroblock;
1069    dec->base.decode_bitstream = vl_mpeg12_decode_bitstream;
1070    dec->base.end_frame = vl_mpeg12_end_frame;
1071    dec->base.flush = vl_mpeg12_flush;
1072
1073    dec->blocks_per_line = MAX2(util_next_power_of_two(dec->base.width) / block_size_pixels, 4);
1074    dec->num_blocks = (dec->base.width * dec->base.height) / block_size_pixels;
1075    dec->width_in_macroblocks = align(dec->base.width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
1076
1077    /* TODO: Implement 422, 444 */
1078    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
1079
1080    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
1081       dec->chroma_width = dec->base.width / 2;
1082       dec->chroma_height = dec->base.height / 2;
1083       dec->num_blocks = dec->num_blocks * 2;
1084    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
1085       dec->chroma_width = dec->base.width;
1086       dec->chroma_height = dec->base.height / 2;
1087       dec->num_blocks = dec->num_blocks * 2 + dec->num_blocks;
1088    } else {
1089       dec->chroma_width = dec->base.width;
1090       dec->chroma_height = dec->base.height;
1091       dec->num_blocks = dec->num_blocks * 3;
1092    }
1093
1094    dec->quads = vl_vb_upload_quads(dec->base.context);
1095    dec->pos = vl_vb_upload_pos(
1096       dec->base.context,
1097       dec->base.width / MACROBLOCK_WIDTH,
1098       dec->base.height / MACROBLOCK_HEIGHT
1099    );
1100
1101    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->base.context);
1102    dec->ves_mv = vl_vb_get_ves_mv(dec->base.context);
1103
1104    switch (entrypoint) {
1105    case PIPE_VIDEO_ENTRYPOINT_BITSTREAM:
1106       format_config = find_format_config(dec, bitstream_format_config, num_bitstream_format_configs);
1107       break;
1108
1109    case PIPE_VIDEO_ENTRYPOINT_IDCT:
1110       format_config = find_format_config(dec, idct_format_config, num_idct_format_configs);
1111       break;
1112
1113    case PIPE_VIDEO_ENTRYPOINT_MC:
1114       format_config = find_format_config(dec, mc_format_config, num_mc_format_configs);
1115       break;
1116
1117    default:
1118       assert(0);
1119       FREE(dec);
1120       return NULL;
1121    }
1122
1123    if (!format_config) {
1124       FREE(dec);
1125       return NULL;
1126    }
1127
1128    if (!init_zscan(dec, format_config))
1129       goto error_zscan;
1130
1131    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1132       if (!init_idct(dec, format_config))
1133          goto error_sources;
1134    } else {
1135       if (!init_mc_source_widthout_idct(dec, format_config))
1136          goto error_sources;
1137    }
1138
1139    if (!vl_mc_init(&dec->mc_y, dec->base.context, dec->base.width, dec->base.height,
1140                    MACROBLOCK_HEIGHT, format_config->mc_scale,
1141                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1142       goto error_mc_y;
1143
1144    // TODO
1145    if (!vl_mc_init(&dec->mc_c, dec->base.context, dec->base.width, dec->base.height,
1146                    BLOCK_HEIGHT, format_config->mc_scale,
1147                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
1148       goto error_mc_c;
1149
1150    if (!init_pipe_state(dec))
1151       goto error_pipe_state;
1152
1153    memset(dec->intra_matrix, 0x10, 64);
1154    memset(dec->non_intra_matrix, 0x10, 64);
1155
1156    return &dec->base;
1157
1158 error_pipe_state:
1159    vl_mc_cleanup(&dec->mc_c);
1160
1161 error_mc_c:
1162    vl_mc_cleanup(&dec->mc_y);
1163
1164 error_mc_y:
1165    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
1166       vl_idct_cleanup(&dec->idct_y);
1167       vl_idct_cleanup(&dec->idct_c);
1168       dec->idct_source->destroy(dec->idct_source);
1169    }
1170    dec->mc_source->destroy(dec->mc_source);
1171
1172 error_sources:
1173    vl_zscan_cleanup(&dec->zscan_y);
1174    vl_zscan_cleanup(&dec->zscan_c);
1175
1176 error_zscan:
1177    FREE(dec);
1178    return NULL;
1179 }