src/gallium/auxiliary/vl/vl_mpeg12_decoder.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 Younes Manton.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include <math.h>
  29 #include <assert.h>
  30
  31 #include <util/u_memory.h>
  32 #include <util/u_rect.h>
  33 #include <util/u_video.h>
  34
  35 #include "vl_mpeg12_decoder.h"
  36 #include "vl_defines.h"
  37
  38 #define SCALE_FACTOR_SNORM (32768.0f / 256.0f)
  39 #define SCALE_FACTOR_SSCALED (1.0f / 256.0f)
  40
  41 static const enum pipe_format const_zscan_source_formats[] = {
  42    PIPE_FORMAT_R16_SNORM,
  43    PIPE_FORMAT_R16_SSCALED
  44 };
  45
  46 static const unsigned num_zscan_source_formats =
  47    sizeof(const_zscan_source_formats) / sizeof(enum pipe_format);
  48
  49 static const enum pipe_format const_idct_source_formats[] = {
  50    PIPE_FORMAT_R16G16B16A16_SNORM,
  51    PIPE_FORMAT_R16G16B16A16_SSCALED
  52 };
  53
  54 static const unsigned num_idct_source_formats =
  55    sizeof(const_idct_source_formats) / sizeof(enum pipe_format);
  56
  57 static const enum pipe_format const_idct_intermediate_formats[] = {
  58    PIPE_FORMAT_R16G16B16A16_FLOAT,
  59    PIPE_FORMAT_R16G16B16A16_SNORM,
  60    PIPE_FORMAT_R16G16B16A16_SSCALED,
  61    PIPE_FORMAT_R32G32B32A32_FLOAT
  62 };
  63
  64 static const unsigned num_idct_intermediate_formats =
  65    sizeof(const_idct_intermediate_formats) / sizeof(enum pipe_format);
  66
  67 static const enum pipe_format const_mc_source_formats[] = {
  68    PIPE_FORMAT_R16_SNORM,
  69    PIPE_FORMAT_R16_SSCALED
  70 };
  71
  72 static const unsigned num_mc_source_formats =
  73    sizeof(const_mc_source_formats) / sizeof(enum pipe_format);
  74
  75 static bool
  76 init_zscan_buffer(struct vl_mpeg12_buffer *buffer)
  77 {
  78    enum pipe_format formats[3];
  79
  80    struct pipe_sampler_view **source;
  81    struct pipe_surface **destination;
  82
  83    struct vl_mpeg12_decoder *dec;
  84
  85    unsigned i;
  86
  87    assert(buffer);
  88
  89    dec = (struct vl_mpeg12_decoder*)buffer->base.decoder;
  90
  91    formats[0] = formats[1] = formats[2] = dec->zscan_source_format;
  92    buffer->zscan_source = vl_video_buffer_init(dec->base.context, dec->pipe,
  93                                                dec->blocks_per_line * BLOCK_WIDTH * BLOCK_HEIGHT,
  94                                                dec->max_blocks / dec->blocks_per_line,
  95                                                1, PIPE_VIDEO_CHROMA_FORMAT_444,
  96                                                formats, PIPE_USAGE_STATIC);
  97    if (!buffer->zscan_source)
  98       goto error_source;
  99
 100    source = buffer->zscan_source->get_sampler_view_planes(buffer->zscan_source);
 101    if (!source)
 102       goto error_sampler;
 103
 104    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 105       destination = buffer->idct_source->get_surfaces(buffer->idct_source);
 106    else
 107       destination = buffer->mc_source->get_surfaces(buffer->mc_source);
 108
 109    if (!destination)
 110       goto error_surface;
 111
 112    for (i = 0; i < VL_MAX_PLANES; ++i)
 113       if (!vl_zscan_init_buffer(i == 0 ? &dec->zscan_y : &dec->zscan_c,
 114                                 &buffer->zscan[i], source[i], destination[i]))
 115          goto error_plane;
 116
 117    return true;
 118
 119 error_plane:
 120    for (; i > 0; --i)
 121       vl_zscan_cleanup_buffer(&buffer->zscan[i - 1]);
 122
 123 error_surface:
 124 error_sampler:
 125    buffer->zscan_source->destroy(buffer->zscan_source);
 126
 127 error_source:
 128    return false;
 129 }
 130
 131 static void
 132 cleanup_zscan_buffer(struct vl_mpeg12_buffer *buffer)
 133 {
 134    unsigned i;
 135
 136    assert(buffer);
 137
 138    for (i = 0; i < VL_MAX_PLANES; ++i)
 139       vl_zscan_cleanup_buffer(&buffer->zscan[i]);
 140    buffer->zscan_source->destroy(buffer->zscan_source);
 141 }
 142
 143 static bool
 144 init_idct_buffer(struct vl_mpeg12_buffer *buffer)
 145 {
 146    enum pipe_format formats[3];
 147
 148    struct pipe_sampler_view **idct_source_sv, **mc_source_sv;
 149    struct pipe_surface **idct_surfaces;
 150
 151    struct vl_mpeg12_decoder *dec;
 152
 153    unsigned i;
 154
 155    assert(buffer);
 156
 157    dec = (struct vl_mpeg12_decoder*)buffer->base.decoder;
 158
 159    formats[0] = formats[1] = formats[2] = dec->idct_source_format;
 160    buffer->idct_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 161                                               dec->base.width / 4, dec->base.height, 1,
 162                                               dec->base.chroma_format,
 163                                               formats, PIPE_USAGE_STATIC);
 164    if (!buffer->idct_source)
 165       goto error_source;
 166
 167    formats[0] = formats[1] = formats[2] = dec->mc_source_format;
 168    buffer->mc_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 169                                             dec->base.width / dec->nr_of_idct_render_targets,
 170                                             dec->base.height / 4, dec->nr_of_idct_render_targets,
 171                                             dec->base.chroma_format,
 172                                             formats, PIPE_USAGE_STATIC);
 173
 174    if (!buffer->mc_source)
 175       goto error_mc_source;
 176
 177    idct_source_sv = buffer->idct_source->get_sampler_view_planes(buffer->idct_source);
 178    if (!idct_source_sv)
 179       goto error_source_sv;
 180
 181    mc_source_sv = buffer->mc_source->get_sampler_view_planes(buffer->mc_source);
 182    if (!mc_source_sv)
 183       goto error_mc_source_sv;
 184
 185    idct_surfaces = buffer->mc_source->get_surfaces(buffer->mc_source);
 186    if (!idct_surfaces)
 187       goto error_surfaces;
 188
 189    for (i = 0; i < 3; ++i)
 190       if (!vl_idct_init_buffer(i == 0 ? &dec->idct_y : &dec->idct_c,
 191                                &buffer->idct[i], idct_source_sv[i],
 192                                mc_source_sv[i], idct_surfaces[i]))
 193          goto error_plane;
 194
 195    return true;
 196
 197 error_plane:
 198    for (; i > 0; --i)
 199       vl_idct_cleanup_buffer(i == 1 ? &dec->idct_c : &dec->idct_y, &buffer->idct[i - 1]);
 200
 201 error_surfaces:
 202 error_mc_source_sv:
 203 error_source_sv:
 204    buffer->mc_source->destroy(buffer->mc_source);
 205
 206 error_mc_source:
 207    buffer->idct_source->destroy(buffer->idct_source);
 208
 209 error_source:
 210    return false;
 211 }
 212
 213 static void
 214 cleanup_idct_buffer(struct vl_mpeg12_buffer *buf)
 215 {
 216    struct vl_mpeg12_decoder *dec;
 217    assert(buf);
 218
 219    dec = (struct vl_mpeg12_decoder*)buf->base.decoder;
 220    assert(dec);
 221
 222    vl_idct_cleanup_buffer(&dec->idct_y, &buf->idct[0]);
 223    vl_idct_cleanup_buffer(&dec->idct_c, &buf->idct[1]);
 224    vl_idct_cleanup_buffer(&dec->idct_c, &buf->idct[2]);
 225    buf->idct_source->destroy(buf->idct_source);
 226 }
 227
 228 static bool
 229 init_mc_buffer(struct vl_mpeg12_buffer *buf)
 230 {
 231    struct vl_mpeg12_decoder *dec;
 232    enum pipe_format formats[3];
 233
 234    assert(buf);
 235
 236    dec = (struct vl_mpeg12_decoder*)buf->base.decoder;
 237    assert(dec);
 238
 239    if (dec->base.entrypoint > PIPE_VIDEO_ENTRYPOINT_IDCT) {
 240       formats[0] = formats[1] = formats[2] =dec->mc_source_format;
 241       buf->mc_source = vl_video_buffer_init(dec->base.context, dec->pipe,
 242                                                dec->base.width, dec->base.height, 1,
 243                                                dec->base.chroma_format,
 244                                                formats, PIPE_USAGE_STATIC);
 245
 246       if (!buf->mc_source)
 247          goto error_mc_source;
 248    }
 249
 250    if(!vl_mc_init_buffer(&dec->mc_y, &buf->mc[0]))
 251       goto error_mc_y;
 252
 253    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[1]))
 254       goto error_mc_cb;
 255
 256    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[2]))
 257       goto error_mc_cr;
 258
 259    return true;
 260
 261 error_mc_cr:
 262    vl_mc_cleanup_buffer(&buf->mc[1]);
 263
 264 error_mc_cb:
 265    vl_mc_cleanup_buffer(&buf->mc[0]);
 266
 267 error_mc_y:
 268    buf->mc_source->destroy(buf->mc_source);
 269
 270 error_mc_source:
 271    return false;
 272 }
 273
 274 static void
 275 cleanup_mc_buffer(struct vl_mpeg12_buffer *buf)
 276 {
 277    unsigned i;
 278
 279    assert(buf);
 280
 281    for (i = 0; i < VL_MAX_PLANES; ++i)
 282       vl_mc_cleanup_buffer(&buf->mc[i]);
 283
 284    buf->mc_source->destroy(buf->mc_source);
 285 }
 286
 287 static void
 288 vl_mpeg12_buffer_destroy(struct pipe_video_decode_buffer *buffer)
 289 {
 290    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 291    struct vl_mpeg12_decoder *dec;
 292
 293    assert(buf);
 294
 295    dec = (struct vl_mpeg12_decoder*)buf->base.decoder;
 296    assert(dec);
 297
 298    cleanup_zscan_buffer(buf);
 299
 300    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 301       cleanup_idct_buffer(buf);
 302
 303    cleanup_mc_buffer(buf);
 304
 305    vl_vb_cleanup(&buf->vertex_stream);
 306
 307    FREE(buf);
 308 }
 309
 310 static void
 311 vl_mpeg12_buffer_map(struct pipe_video_decode_buffer *buffer)
 312 {
 313    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 314    struct vl_mpeg12_decoder *dec;
 315
 316    struct pipe_sampler_view **sampler_views;
 317    unsigned i;
 318
 319    assert(buf);
 320
 321    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 322    assert(dec);
 323
 324    vl_vb_map(&buf->vertex_stream, dec->pipe);
 325
 326    sampler_views = buf->zscan_source->get_sampler_view_planes(buf->zscan_source);
 327
 328    assert(sampler_views);
 329
 330    for (i = 0; i < VL_MAX_PLANES; ++i) {
 331       struct pipe_resource *tex = sampler_views[i]->texture;
 332       struct pipe_box rect =
 333       {
 334          0, 0, 0,
 335          tex->width0,
 336          tex->height0,
 337          1
 338       };
 339
 340       buf->tex_transfer[i] = dec->pipe->get_transfer
 341       (
 342          dec->pipe, tex,
 343          0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 344          &rect
 345       );
 346
 347       buf->texels[i] = dec->pipe->transfer_map(dec->pipe, buf->tex_transfer[i]);
 348    }
 349
 350    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 351       struct pipe_ycbcr_block *ycbcr_stream[VL_MAX_PLANES];
 352       struct pipe_motionvector *mv_stream[VL_MAX_REF_FRAMES];
 353
 354       for (i = 0; i < VL_MAX_PLANES; ++i)
 355          ycbcr_stream[i] = vl_vb_get_ycbcr_stream(&buf->vertex_stream, i);
 356
 357       for (i = 0; i < VL_MAX_REF_FRAMES; ++i)
 358          mv_stream[i] = vl_vb_get_mv_stream(&buf->vertex_stream, i);
 359
 360       vl_mpg12_bs_set_buffers(&buf->bs, ycbcr_stream, buf->texels, mv_stream);
 361    } else {
 362       for (i = 0; i < VL_MAX_PLANES; ++i)
 363          vl_zscan_set_layout(&buf->zscan[i], dec->zscan_linear);
 364    }
 365 }
 366
 367 static struct pipe_ycbcr_block *
 368 vl_mpeg12_buffer_get_ycbcr_stream(struct pipe_video_decode_buffer *buffer, int component)
 369 {
 370    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 371
 372    assert(buf);
 373
 374    return vl_vb_get_ycbcr_stream(&buf->vertex_stream, component);
 375 }
 376
 377 static short *
 378 vl_mpeg12_buffer_get_ycbcr_buffer(struct pipe_video_decode_buffer *buffer, int component)
 379 {
 380    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 381
 382    assert(buf);
 383    assert(component < VL_MAX_PLANES);
 384
 385    return buf->texels[component];
 386 }
 387
 388 static unsigned
 389 vl_mpeg12_buffer_get_mv_stream_stride(struct pipe_video_decode_buffer *buffer)
 390 {
 391    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 392
 393    assert(buf);
 394
 395    return vl_vb_get_mv_stream_stride(&buf->vertex_stream);
 396 }
 397
 398 static struct pipe_motionvector *
 399 vl_mpeg12_buffer_get_mv_stream(struct pipe_video_decode_buffer *buffer, int ref_frame)
 400 {
 401    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 402
 403    assert(buf);
 404
 405    return vl_vb_get_mv_stream(&buf->vertex_stream, ref_frame);
 406 }
 407
 408 static void
 409 vl_mpeg12_buffer_decode_bitstream(struct pipe_video_decode_buffer *buffer,
 410                                   unsigned num_bytes, const void *data,
 411                                   struct pipe_mpeg12_picture_desc *picture,
 412                                   unsigned num_ycbcr_blocks[3])
 413 {
 414    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 415    struct vl_mpeg12_decoder *dec;
 416    unsigned i;
 417
 418    assert(buf);
 419
 420    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 421    assert(dec);
 422
 423    for (i = 0; i < VL_MAX_PLANES; ++i)
 424       vl_zscan_set_layout(&buf->zscan[i], picture->alternate_scan ? dec->zscan_alternate : dec->zscan_normal);
 425
 426    vl_mpg12_bs_decode(&buf->bs, num_bytes, data, picture, num_ycbcr_blocks);
 427 }
 428
 429 static void
 430 vl_mpeg12_buffer_unmap(struct pipe_video_decode_buffer *buffer)
 431 {
 432    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer*)buffer;
 433    struct vl_mpeg12_decoder *dec;
 434    unsigned i;
 435
 436    assert(buf);
 437
 438    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 439    assert(dec);
 440
 441    vl_vb_unmap(&buf->vertex_stream, dec->pipe);
 442
 443    for (i = 0; i < VL_MAX_PLANES; ++i) {
 444       dec->pipe->transfer_unmap(dec->pipe, buf->tex_transfer[i]);
 445       dec->pipe->transfer_destroy(dec->pipe, buf->tex_transfer[i]);
 446    }
 447 }
 448
 449 static void
 450 vl_mpeg12_destroy(struct pipe_video_decoder *decoder)
 451 {
 452    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 453
 454    assert(decoder);
 455
 456    /* Asserted in softpipe_delete_fs_state() for some reason */
 457    dec->pipe->bind_vs_state(dec->pipe, NULL);
 458    dec->pipe->bind_fs_state(dec->pipe, NULL);
 459
 460    dec->pipe->delete_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 461    dec->pipe->delete_sampler_state(dec->pipe, dec->sampler_ycbcr);
 462
 463    vl_mc_cleanup(&dec->mc_y);
 464    vl_mc_cleanup(&dec->mc_c);
 465
 466    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 467       vl_idct_cleanup(&dec->idct_y);
 468       vl_idct_cleanup(&dec->idct_c);
 469    }
 470
 471    vl_zscan_cleanup(&dec->zscan_y);
 472    vl_zscan_cleanup(&dec->zscan_c);
 473
 474    dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_ycbcr);
 475    dec->pipe->delete_vertex_elements_state(dec->pipe, dec->ves_mv);
 476
 477    pipe_resource_reference(&dec->quads.buffer, NULL);
 478    pipe_resource_reference(&dec->pos.buffer, NULL);
 479
 480    pipe_sampler_view_reference(&dec->zscan_linear, NULL);
 481    pipe_sampler_view_reference(&dec->zscan_normal, NULL);
 482    pipe_sampler_view_reference(&dec->zscan_alternate, NULL);
 483
 484    FREE(dec);
 485 }
 486
 487 static struct pipe_video_decode_buffer *
 488 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 489 {
 490    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 491    struct vl_mpeg12_buffer *buffer;
 492
 493    assert(dec);
 494
 495    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 496    if (buffer == NULL)
 497       return NULL;
 498
 499    buffer->base.decoder = decoder;
 500    buffer->base.destroy = vl_mpeg12_buffer_destroy;
 501    buffer->base.map = vl_mpeg12_buffer_map;
 502    buffer->base.get_ycbcr_stream = vl_mpeg12_buffer_get_ycbcr_stream;
 503    buffer->base.get_ycbcr_buffer = vl_mpeg12_buffer_get_ycbcr_buffer;
 504    buffer->base.get_mv_stream_stride = vl_mpeg12_buffer_get_mv_stream_stride;
 505    buffer->base.get_mv_stream = vl_mpeg12_buffer_get_mv_stream;
 506    buffer->base.decode_bitstream = vl_mpeg12_buffer_decode_bitstream;
 507    buffer->base.unmap = vl_mpeg12_buffer_unmap;
 508
 509    if (!vl_vb_init(&buffer->vertex_stream, dec->pipe,
 510                    dec->base.width / MACROBLOCK_WIDTH,
 511                    dec->base.height / MACROBLOCK_HEIGHT))
 512       goto error_vertex_buffer;
 513
 514    if (!init_mc_buffer(buffer))
 515       goto error_mc;
 516
 517    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 518       if (!init_idct_buffer(buffer))
 519          goto error_idct;
 520
 521    if (!init_zscan_buffer(buffer))
 522       goto error_zscan;
 523
 524    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 525       vl_mpg12_bs_init(&buffer->bs,
 526                        dec->base.width / MACROBLOCK_WIDTH,
 527                        dec->base.height / MACROBLOCK_HEIGHT);
 528
 529    return &buffer->base;
 530
 531 error_zscan:
 532    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 533       cleanup_idct_buffer(buffer);
 534
 535 error_idct:
 536    cleanup_mc_buffer(buffer);
 537
 538 error_mc:
 539    vl_vb_cleanup(&buffer->vertex_stream);
 540
 541 error_vertex_buffer:
 542    FREE(buffer);
 543    return NULL;
 544 }
 545
 546 static void
 547 vl_mpeg12_decoder_flush_buffer(struct pipe_video_decode_buffer *buffer,
 548                                unsigned num_ycbcr_blocks[3],
 549                                struct pipe_video_buffer *refs[2],
 550                                struct pipe_video_buffer *dst)
 551 {
 552    struct vl_mpeg12_buffer *buf = (struct vl_mpeg12_buffer *)buffer;
 553    struct vl_mpeg12_decoder *dec;
 554
 555    struct pipe_sampler_view **sv[VL_MAX_REF_FRAMES], **mc_source_sv;
 556    struct pipe_surface **surfaces;
 557
 558    struct pipe_vertex_buffer vb[3];
 559
 560    unsigned i, j, component;
 561    unsigned nr_components;
 562
 563    assert(buf);
 564
 565    dec = (struct vl_mpeg12_decoder *)buf->base.decoder;
 566    assert(dec);
 567
 568    for (i = 0; i < 2; ++i)
 569       sv[i] = refs[i] ? refs[i]->get_sampler_view_planes(refs[i]) : NULL;
 570
 571    vb[0] = dec->quads;
 572    vb[1] = dec->pos;
 573
 574    surfaces = dst->get_surfaces(dst);
 575
 576    dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_mv);
 577    for (i = 0; i < VL_MAX_PLANES; ++i) {
 578       if (!surfaces[i]) continue;
 579
 580       vl_mc_set_surface(&buf->mc[i], surfaces[i]);
 581
 582       for (j = 0; j < VL_MAX_REF_FRAMES; ++j) {
 583          if (!sv[j]) continue;
 584
 585          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 586          dec->pipe->set_vertex_buffers(dec->pipe, 3, vb);
 587
 588          vl_mc_render_ref(&buf->mc[i], sv[j][i]);
 589       }
 590    }
 591
 592    dec->pipe->bind_vertex_elements_state(dec->pipe, dec->ves_ycbcr);
 593    for (i = 0; i < VL_MAX_PLANES; ++i) {
 594       if (!num_ycbcr_blocks[i]) continue;
 595
 596       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 597       dec->pipe->set_vertex_buffers(dec->pipe, 2, vb);
 598
 599       vl_zscan_render(&buf->zscan[i] , num_ycbcr_blocks[i]);
 600
 601       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 602          vl_idct_flush(i == 0 ? &dec->idct_y : &dec->idct_c, &buf->idct[i], num_ycbcr_blocks[i]);
 603    }
 604
 605    mc_source_sv = buf->mc_source->get_sampler_view_planes(buf->mc_source);
 606    for (i = 0, component = 0; i < VL_MAX_PLANES; ++i) {
 607       if (!surfaces[i]) continue;
 608
 609       nr_components = util_format_get_nr_components(surfaces[i]->texture->format);
 610       for (j = 0; j < nr_components; ++j, ++component) {
 611          if (!num_ycbcr_blocks[i]) continue;
 612
 613          vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, component);
 614          dec->pipe->set_vertex_buffers(dec->pipe, 2, vb);
 615
 616          if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 617             vl_idct_prepare_stage2(component == 0 ? &dec->idct_y : &dec->idct_c, &buf->idct[component]);
 618          else {
 619             dec->pipe->set_fragment_sampler_views(dec->pipe, 1, &mc_source_sv[component]);
 620             dec->pipe->bind_fragment_sampler_states(dec->pipe, 1, &dec->sampler_ycbcr);
 621          }
 622          vl_mc_render_ycbcr(&buf->mc[i], j, num_ycbcr_blocks[component]);
 623       }
 624    }
 625 }
 626
 627 static bool
 628 init_pipe_state(struct vl_mpeg12_decoder *dec)
 629 {
 630    struct pipe_depth_stencil_alpha_state dsa;
 631    struct pipe_sampler_state sampler;
 632    unsigned i;
 633
 634    assert(dec);
 635
 636    memset(&dsa, 0, sizeof dsa);
 637    dsa.depth.enabled = 0;
 638    dsa.depth.writemask = 0;
 639    dsa.depth.func = PIPE_FUNC_ALWAYS;
 640    for (i = 0; i < 2; ++i) {
 641       dsa.stencil[i].enabled = 0;
 642       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 643       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 644       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 645       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 646       dsa.stencil[i].valuemask = 0;
 647       dsa.stencil[i].writemask = 0;
 648    }
 649    dsa.alpha.enabled = 0;
 650    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 651    dsa.alpha.ref_value = 0;
 652    dec->dsa = dec->pipe->create_depth_stencil_alpha_state(dec->pipe, &dsa);
 653    dec->pipe->bind_depth_stencil_alpha_state(dec->pipe, dec->dsa);
 654
 655    memset(&sampler, 0, sizeof(sampler));
 656    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 657    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 658    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 659    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 660    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 661    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 662    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 663    sampler.compare_func = PIPE_FUNC_ALWAYS;
 664    sampler.normalized_coords = 1;
 665    dec->sampler_ycbcr = dec->pipe->create_sampler_state(dec->pipe, &sampler);
 666    if (!dec->sampler_ycbcr)
 667       return false;
 668
 669    return true;
 670 }
 671
 672 static enum pipe_format
 673 find_first_supported_format(struct vl_mpeg12_decoder *dec,
 674                             const enum pipe_format formats[],
 675                             unsigned num_formats,
 676                             enum pipe_texture_target target)
 677 {
 678    struct pipe_screen *screen;
 679    unsigned i;
 680
 681    assert(dec);
 682
 683    screen = dec->pipe->screen;
 684
 685    for (i = 0; i < num_formats; ++i)
 686       if (screen->is_format_supported(dec->pipe->screen, formats[i], target, 1,
 687                                       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 688          return formats[i];
 689
 690    return PIPE_FORMAT_NONE;
 691 }
 692
 693 static bool
 694 init_zscan(struct vl_mpeg12_decoder *dec)
 695 {
 696    unsigned num_channels;
 697
 698    assert(dec);
 699
 700    dec->blocks_per_line = 4;
 701    dec->max_blocks =
 702       (dec->base.width * dec->base.height) /
 703       (BLOCK_WIDTH * BLOCK_HEIGHT);
 704
 705    dec->zscan_source_format = find_first_supported_format(dec, const_zscan_source_formats,
 706                                                           num_zscan_source_formats, PIPE_TEXTURE_2D);
 707
 708    if (dec->zscan_source_format == PIPE_FORMAT_NONE)
 709       return false;
 710
 711    dec->zscan_linear = vl_zscan_layout(dec->pipe, vl_zscan_linear, dec->blocks_per_line);
 712    dec->zscan_normal = vl_zscan_layout(dec->pipe, vl_zscan_normal, dec->blocks_per_line);
 713    dec->zscan_alternate = vl_zscan_layout(dec->pipe, vl_zscan_alternate, dec->blocks_per_line);
 714
 715    num_channels = dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT ? 4 : 1;
 716
 717    if (!vl_zscan_init(&dec->zscan_y, dec->pipe, dec->base.width, dec->base.height,
 718                       dec->blocks_per_line, dec->max_blocks, num_channels))
 719       return false;
 720
 721    if (!vl_zscan_init(&dec->zscan_c, dec->pipe, dec->chroma_width, dec->chroma_height,
 722                       dec->blocks_per_line, dec->max_blocks, num_channels))
 723       return false;
 724
 725    return true;
 726 }
 727
 728 static bool
 729 init_idct(struct vl_mpeg12_decoder *dec)
 730 {
 731    struct pipe_sampler_view *matrix, *transpose = NULL;
 732    float matrix_scale, transpose_scale;
 733
 734    dec->nr_of_idct_render_targets = dec->pipe->screen->get_param(dec->pipe->screen, PIPE_CAP_MAX_RENDER_TARGETS);
 735
 736    // more than 4 render targets usually doesn't makes any seens
 737    dec->nr_of_idct_render_targets = MIN2(dec->nr_of_idct_render_targets, 4);
 738
 739    dec->idct_source_format = find_first_supported_format(dec, const_idct_source_formats,
 740                                                          num_idct_source_formats, PIPE_TEXTURE_2D);
 741
 742    if (dec->idct_source_format == PIPE_FORMAT_NONE)
 743       return false;
 744
 745    dec->mc_source_format = find_first_supported_format(dec, const_idct_intermediate_formats,
 746                                                        num_idct_intermediate_formats, PIPE_TEXTURE_3D);
 747
 748    if (dec->mc_source_format == PIPE_FORMAT_NONE)
 749       return false;
 750
 751    switch (dec->idct_source_format) {
 752    case PIPE_FORMAT_R16G16B16A16_SSCALED:
 753       matrix_scale = SCALE_FACTOR_SSCALED;
 754       break;
 755
 756    case PIPE_FORMAT_R16G16B16A16_SNORM:
 757       matrix_scale = SCALE_FACTOR_SNORM;
 758       break;
 759
 760    default:
 761       assert(0);
 762       return false;
 763    }
 764
 765    if (dec->mc_source_format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
 766        dec->mc_source_format == PIPE_FORMAT_R32G32B32A32_FLOAT)
 767       transpose_scale = 1.0f;
 768    else
 769       transpose_scale = matrix_scale = sqrt(matrix_scale);
 770
 771    if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 772       transpose_scale /= SCALE_FACTOR_SSCALED;
 773
 774    if (!(matrix = vl_idct_upload_matrix(dec->pipe, matrix_scale)))
 775       goto error_matrix;
 776
 777    if (matrix_scale != transpose_scale) {
 778       if (!(transpose = vl_idct_upload_matrix(dec->pipe, transpose_scale)))
 779          goto error_transpose;
 780    } else
 781       pipe_sampler_view_reference(&transpose, matrix);
 782
 783    if (!vl_idct_init(&dec->idct_y, dec->pipe, dec->base.width, dec->base.height,
 784                      dec->nr_of_idct_render_targets, matrix, transpose))
 785       goto error_y;
 786
 787    if(!vl_idct_init(&dec->idct_c, dec->pipe, dec->chroma_width, dec->chroma_height,
 788                     dec->nr_of_idct_render_targets, matrix, transpose))
 789       goto error_c;
 790
 791    pipe_sampler_view_reference(&matrix, NULL);
 792    pipe_sampler_view_reference(&transpose, NULL);
 793    return true;
 794
 795 error_c:
 796    vl_idct_cleanup(&dec->idct_y);
 797
 798 error_y:
 799    pipe_sampler_view_reference(&transpose, NULL);
 800
 801 error_transpose:
 802    pipe_sampler_view_reference(&matrix, NULL);
 803
 804 error_matrix:
 805    return false;
 806 }
 807
 808 static void
 809 mc_vert_shader_callback(void *priv, struct vl_mc *mc,
 810                         struct ureg_program *shader,
 811                         unsigned first_output,
 812                         struct ureg_dst tex)
 813 {
 814    struct vl_mpeg12_decoder *dec = priv;
 815    struct ureg_dst o_vtex;
 816
 817    assert(priv && mc);
 818    assert(shader);
 819
 820    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 821       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
 822       vl_idct_stage2_vert_shader(idct, shader, first_output, tex);
 823    } else {
 824       o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output);
 825       ureg_MOV(shader, ureg_writemask(o_vtex, TGSI_WRITEMASK_XY), ureg_src(tex));
 826    }
 827 }
 828
 829 static void
 830 mc_frag_shader_callback(void *priv, struct vl_mc *mc,
 831                         struct ureg_program *shader,
 832                         unsigned first_input,
 833                         struct ureg_dst dst)
 834 {
 835    struct vl_mpeg12_decoder *dec = priv;
 836    struct ureg_src src, sampler;
 837
 838    assert(priv && mc);
 839    assert(shader);
 840
 841    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 842       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
 843       vl_idct_stage2_frag_shader(idct, shader, first_input, dst);
 844    } else {
 845       src = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input, TGSI_INTERPOLATE_LINEAR);
 846       sampler = ureg_DECL_sampler(shader, 0);
 847       ureg_TEX(shader, dst, TGSI_TEXTURE_2D, src, sampler);
 848    }
 849 }
 850
 851 struct pipe_video_decoder *
 852 vl_create_mpeg12_decoder(struct pipe_video_context *context,
 853                          struct pipe_context *pipe,
 854                          enum pipe_video_profile profile,
 855                          enum pipe_video_entrypoint entrypoint,
 856                          enum pipe_video_chroma_format chroma_format,
 857                          unsigned width, unsigned height)
 858 {
 859    struct vl_mpeg12_decoder *dec;
 860    float mc_scale;
 861
 862    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
 863
 864    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
 865
 866    if (!dec)
 867       return NULL;
 868
 869    dec->base.context = context;
 870    dec->base.profile = profile;
 871    dec->base.entrypoint = entrypoint;
 872    dec->base.chroma_format = chroma_format;
 873    dec->base.width = width;
 874    dec->base.height = height;
 875
 876    dec->base.destroy = vl_mpeg12_destroy;
 877    dec->base.create_buffer = vl_mpeg12_create_buffer;
 878    dec->base.flush_buffer = vl_mpeg12_decoder_flush_buffer;
 879
 880    dec->pipe = pipe;
 881
 882    dec->quads = vl_vb_upload_quads(dec->pipe);
 883    dec->pos = vl_vb_upload_pos(
 884       dec->pipe,
 885       dec->base.width / MACROBLOCK_WIDTH,
 886       dec->base.height / MACROBLOCK_HEIGHT
 887    );
 888
 889    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->pipe);
 890    dec->ves_mv = vl_vb_get_ves_mv(dec->pipe);
 891
 892    /* TODO: Implement 422, 444 */
 893    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 894
 895    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
 896       dec->chroma_width = dec->base.width / 2;
 897       dec->chroma_height = dec->base.height / 2;
 898    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
 899       dec->chroma_width = dec->base.width;
 900       dec->chroma_height = dec->base.height / 2;
 901    } else {
 902       dec->chroma_width = dec->base.width;
 903       dec->chroma_height = dec->base.height;
 904    }
 905
 906    if (!init_zscan(dec))
 907       goto error_zscan;
 908
 909    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 910       if (!init_idct(dec))
 911          goto error_idct;
 912       if (dec->mc_source_format == PIPE_FORMAT_R16_SSCALED)
 913          mc_scale = SCALE_FACTOR_SSCALED;
 914       else
 915          mc_scale = 1.0f;
 916    } else {
 917       dec->mc_source_format = find_first_supported_format(dec, const_mc_source_formats,
 918                                                           num_mc_source_formats, PIPE_TEXTURE_3D);
 919
 920       if (dec->mc_source_format == PIPE_FORMAT_NONE)
 921          return NULL;
 922
 923       switch (dec->mc_source_format) {
 924       case PIPE_FORMAT_R16_SNORM:
 925          mc_scale = SCALE_FACTOR_SNORM;
 926          break;
 927
 928       case PIPE_FORMAT_R16_SSCALED:
 929          mc_scale = SCALE_FACTOR_SSCALED;
 930          break;
 931
 932       default:
 933          assert(0);
 934          return NULL;
 935       }
 936    }
 937
 938    if (!vl_mc_init(&dec->mc_y, dec->pipe, dec->base.width, dec->base.height, MACROBLOCK_HEIGHT, mc_scale,
 939                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
 940       goto error_mc_y;
 941
 942    // TODO
 943    if (!vl_mc_init(&dec->mc_c, dec->pipe, dec->base.width, dec->base.height, BLOCK_HEIGHT, mc_scale,
 944                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
 945       goto error_mc_c;
 946
 947    if (!init_pipe_state(dec))
 948       goto error_pipe_state;
 949
 950    return &dec->base;
 951
 952 error_pipe_state:
 953    vl_mc_cleanup(&dec->mc_c);
 954
 955 error_mc_c:
 956    vl_mc_cleanup(&dec->mc_y);
 957
 958 error_mc_y:
 959    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 960       vl_idct_cleanup(&dec->idct_y);
 961       vl_idct_cleanup(&dec->idct_c);
 962    }
 963
 964 error_idct:
 965    vl_zscan_cleanup(&dec->zscan_y);
 966    vl_zscan_cleanup(&dec->zscan_c);
 967
 968 error_zscan:
 969    FREE(dec);
 970    return NULL;
 971 }