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 struct format_config {
  42    enum pipe_format zscan_source_format;
  43    enum pipe_format idct_source_format;
  44    enum pipe_format mc_source_format;
  45
  46    float idct_scale;
  47    float mc_scale;
  48 };
  49
  50 static const struct format_config bitstream_format_config[] = {
  51    { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SSCALED },
  52    { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, 1.0f, SCALE_FACTOR_SSCALED },
  53    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SNORM },
  54    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, 1.0f, SCALE_FACTOR_SNORM }
  55 };
  56
  57 static const unsigned num_bitstream_format_configs =
  58    sizeof(bitstream_format_config) / sizeof(struct format_config);
  59
  60 static const struct format_config idct_format_config[] = {
  61    { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SSCALED },
  62    { PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, PIPE_FORMAT_R16G16B16A16_SSCALED, 1.0f, SCALE_FACTOR_SSCALED },
  63    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_FLOAT, 1.0f, SCALE_FACTOR_SNORM },
  64    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, PIPE_FORMAT_R16G16B16A16_SNORM, 1.0f, SCALE_FACTOR_SNORM }
  65 };
  66
  67 static const unsigned num_idct_format_configs =
  68    sizeof(idct_format_config) / sizeof(struct format_config);
  69
  70 static const struct format_config mc_format_config[] = {
  71    //{ PIPE_FORMAT_R16_SSCALED, PIPE_FORMAT_NONE, PIPE_FORMAT_R16_SSCALED, 0.0f, SCALE_FACTOR_SSCALED },
  72    { PIPE_FORMAT_R16_SNORM, PIPE_FORMAT_NONE, PIPE_FORMAT_R16_SNORM, 0.0f, SCALE_FACTOR_SNORM }
  73 };
  74
  75 static const unsigned num_mc_format_configs =
  76    sizeof(mc_format_config) / sizeof(struct format_config);
  77
  78 static bool
  79 init_zscan_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buffer)
  80 {
  81    enum pipe_format formats[3];
  82
  83    struct pipe_sampler_view **source;
  84    struct pipe_surface **destination;
  85
  86    unsigned i;
  87
  88    assert(dec && buffer);
  89
  90    formats[0] = formats[1] = formats[2] = dec->zscan_source_format;
  91    buffer->zscan_source = vl_video_buffer_create_ex
  92    (
  93       dec->base.context,
  94       dec->blocks_per_line * BLOCK_WIDTH * BLOCK_HEIGHT,
  95       align(dec->num_blocks, dec->blocks_per_line) / dec->blocks_per_line,
  96       1, PIPE_VIDEO_CHROMA_FORMAT_444, formats, PIPE_USAGE_STATIC
  97    );
  98
  99    if (!buffer->zscan_source)
 100       goto error_source;
 101
 102    source = buffer->zscan_source->get_sampler_view_planes(buffer->zscan_source);
 103    if (!source)
 104       goto error_sampler;
 105
 106    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 107       destination = dec->idct_source->get_surfaces(dec->idct_source);
 108    else
 109       destination = dec->mc_source->get_surfaces(dec->mc_source);
 110
 111    if (!destination)
 112       goto error_surface;
 113
 114    for (i = 0; i < VL_MAX_PLANES; ++i)
 115       if (!vl_zscan_init_buffer(i == 0 ? &dec->zscan_y : &dec->zscan_c,
 116                                 &buffer->zscan[i], source[i], destination[i]))
 117          goto error_plane;
 118
 119    return true;
 120
 121 error_plane:
 122    for (; i > 0; --i)
 123       vl_zscan_cleanup_buffer(&buffer->zscan[i - 1]);
 124
 125 error_surface:
 126 error_sampler:
 127    buffer->zscan_source->destroy(buffer->zscan_source);
 128
 129 error_source:
 130    return false;
 131 }
 132
 133 static void
 134 cleanup_zscan_buffer(struct vl_mpeg12_buffer *buffer)
 135 {
 136    unsigned i;
 137
 138    assert(buffer);
 139
 140    for (i = 0; i < VL_MAX_PLANES; ++i)
 141       vl_zscan_cleanup_buffer(&buffer->zscan[i]);
 142    buffer->zscan_source->destroy(buffer->zscan_source);
 143 }
 144
 145 static bool
 146 init_idct_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buffer)
 147 {
 148    struct pipe_sampler_view **idct_source_sv, **mc_source_sv;
 149
 150    unsigned i;
 151
 152    assert(dec && buffer);
 153
 154    idct_source_sv = dec->idct_source->get_sampler_view_planes(dec->idct_source);
 155    if (!idct_source_sv)
 156       goto error_source_sv;
 157
 158    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 159    if (!mc_source_sv)
 160       goto error_mc_source_sv;
 161
 162    for (i = 0; i < 3; ++i)
 163       if (!vl_idct_init_buffer(i == 0 ? &dec->idct_y : &dec->idct_c,
 164                                &buffer->idct[i], idct_source_sv[i],
 165                                mc_source_sv[i]))
 166          goto error_plane;
 167
 168    return true;
 169
 170 error_plane:
 171    for (; i > 0; --i)
 172       vl_idct_cleanup_buffer(&buffer->idct[i - 1]);
 173
 174 error_mc_source_sv:
 175 error_source_sv:
 176    return false;
 177 }
 178
 179 static void
 180 cleanup_idct_buffer(struct vl_mpeg12_buffer *buf)
 181 {
 182    unsigned i;
 183
 184    assert(buf);
 185
 186    for (i = 0; i < 3; ++i)
 187       vl_idct_cleanup_buffer(&buf->idct[0]);
 188 }
 189
 190 static bool
 191 init_mc_buffer(struct vl_mpeg12_decoder *dec, struct vl_mpeg12_buffer *buf)
 192 {
 193    assert(dec && buf);
 194
 195    if(!vl_mc_init_buffer(&dec->mc_y, &buf->mc[0]))
 196       goto error_mc_y;
 197
 198    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[1]))
 199       goto error_mc_cb;
 200
 201    if(!vl_mc_init_buffer(&dec->mc_c, &buf->mc[2]))
 202       goto error_mc_cr;
 203
 204    return true;
 205
 206 error_mc_cr:
 207    vl_mc_cleanup_buffer(&buf->mc[1]);
 208
 209 error_mc_cb:
 210    vl_mc_cleanup_buffer(&buf->mc[0]);
 211
 212 error_mc_y:
 213    return false;
 214 }
 215
 216 static void
 217 cleanup_mc_buffer(struct vl_mpeg12_buffer *buf)
 218 {
 219    unsigned i;
 220
 221    assert(buf);
 222
 223    for (i = 0; i < VL_MAX_PLANES; ++i)
 224       vl_mc_cleanup_buffer(&buf->mc[i]);
 225 }
 226
 227 static void
 228 vl_mpeg12_destroy(struct pipe_video_decoder *decoder)
 229 {
 230    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 231
 232    assert(decoder);
 233
 234    /* Asserted in softpipe_delete_fs_state() for some reason */
 235    dec->base.context->bind_vs_state(dec->base.context, NULL);
 236    dec->base.context->bind_fs_state(dec->base.context, NULL);
 237
 238    dec->base.context->delete_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 239    dec->base.context->delete_sampler_state(dec->base.context, dec->sampler_ycbcr);
 240
 241    vl_mc_cleanup(&dec->mc_y);
 242    vl_mc_cleanup(&dec->mc_c);
 243    dec->mc_source->destroy(dec->mc_source);
 244
 245    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 246       vl_idct_cleanup(&dec->idct_y);
 247       vl_idct_cleanup(&dec->idct_c);
 248       dec->idct_source->destroy(dec->idct_source);
 249    }
 250
 251    vl_zscan_cleanup(&dec->zscan_y);
 252    vl_zscan_cleanup(&dec->zscan_c);
 253
 254    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 255    dec->base.context->delete_vertex_elements_state(dec->base.context, dec->ves_mv);
 256
 257    pipe_resource_reference(&dec->quads.buffer, NULL);
 258    pipe_resource_reference(&dec->pos.buffer, NULL);
 259    pipe_resource_reference(&dec->block_num.buffer, NULL);
 260
 261    pipe_sampler_view_reference(&dec->zscan_linear, NULL);
 262    pipe_sampler_view_reference(&dec->zscan_normal, NULL);
 263    pipe_sampler_view_reference(&dec->zscan_alternate, NULL);
 264
 265    FREE(dec);
 266 }
 267
 268 static void *
 269 vl_mpeg12_create_buffer(struct pipe_video_decoder *decoder)
 270 {
 271    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 272    struct vl_mpeg12_buffer *buffer;
 273
 274    assert(dec);
 275
 276    buffer = CALLOC_STRUCT(vl_mpeg12_buffer);
 277    if (buffer == NULL)
 278       return NULL;
 279
 280    if (!vl_vb_init(&buffer->vertex_stream, dec->base.context,
 281                    dec->base.width / MACROBLOCK_WIDTH,
 282                    dec->base.height / MACROBLOCK_HEIGHT))
 283       goto error_vertex_buffer;
 284
 285    if (!init_mc_buffer(dec, buffer))
 286       goto error_mc;
 287
 288    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 289       if (!init_idct_buffer(dec, buffer))
 290          goto error_idct;
 291
 292    if (!init_zscan_buffer(dec, buffer))
 293       goto error_zscan;
 294
 295    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM)
 296       vl_mpg12_bs_init(&buffer->bs,
 297                        dec->base.width / MACROBLOCK_WIDTH,
 298                        dec->base.height / MACROBLOCK_HEIGHT);
 299
 300    return buffer;
 301
 302 error_zscan:
 303    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 304       cleanup_idct_buffer(buffer);
 305
 306 error_idct:
 307    cleanup_mc_buffer(buffer);
 308
 309 error_mc:
 310    vl_vb_cleanup(&buffer->vertex_stream);
 311
 312 error_vertex_buffer:
 313    FREE(buffer);
 314    return NULL;
 315 }
 316
 317 static void
 318 vl_mpeg12_destroy_buffer(struct pipe_video_decoder *decoder, void *buffer)
 319 {
 320    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder*)decoder;
 321    struct vl_mpeg12_buffer *buf = buffer;
 322
 323    assert(dec && buf);
 324
 325    cleanup_zscan_buffer(buf);
 326
 327    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 328       cleanup_idct_buffer(buf);
 329
 330    cleanup_mc_buffer(buf);
 331
 332    vl_vb_cleanup(&buf->vertex_stream);
 333
 334    FREE(buf);
 335 }
 336
 337 static void
 338 vl_mpeg12_set_decode_buffer(struct pipe_video_decoder *decoder, void *buffer)
 339 {
 340    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 341
 342    assert(dec && buffer);
 343
 344    dec->current_buffer = buffer;
 345 }
 346
 347 static void
 348 vl_mpeg12_set_picture_parameters(struct pipe_video_decoder *decoder,
 349                                  struct pipe_picture_desc *picture)
 350 {
 351    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 352    struct pipe_mpeg12_picture_desc *pic = (struct pipe_mpeg12_picture_desc *)picture;
 353
 354    assert(dec && pic);
 355
 356    dec->picture_desc = *pic;
 357 }
 358
 359 static void
 360 vl_mpeg12_set_quant_matrix(struct pipe_video_decoder *decoder,
 361                            const uint8_t intra_matrix[64],
 362                            const uint8_t non_intra_matrix[64])
 363 {
 364    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 365
 366    assert(dec);
 367
 368    memcpy(dec->intra_matrix, intra_matrix, 64);
 369    memcpy(dec->non_intra_matrix, non_intra_matrix, 64);
 370 }
 371
 372 static void
 373 vl_mpeg12_set_decode_target(struct pipe_video_decoder *decoder,
 374                             struct pipe_video_buffer *target)
 375 {
 376    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 377    struct pipe_surface **surfaces;
 378    unsigned i;
 379
 380    assert(dec);
 381
 382    surfaces = target->get_surfaces(target);
 383    for (i = 0; i < VL_MAX_PLANES; ++i)
 384       pipe_surface_reference(&dec->target_surfaces[i], surfaces[i]);
 385 }
 386
 387 static void
 388 vl_mpeg12_set_reference_frames(struct pipe_video_decoder *decoder,
 389                                struct pipe_video_buffer **ref_frames,
 390                                unsigned num_ref_frames)
 391 {
 392    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 393    struct pipe_sampler_view **sv;
 394    unsigned i,j;
 395
 396    assert(dec);
 397    assert(num_ref_frames <= VL_MAX_REF_FRAMES);
 398
 399    for (i = 0; i < num_ref_frames; ++i) {
 400       sv = ref_frames[i]->get_sampler_view_planes(ref_frames[i]);
 401       for (j = 0; j < VL_MAX_PLANES; ++j)
 402          pipe_sampler_view_reference(&dec->ref_frames[i][j], sv[j]);
 403    }
 404
 405    for (; i < VL_MAX_REF_FRAMES; ++i)
 406       for (j = 0; j < VL_MAX_PLANES; ++j)
 407          pipe_sampler_view_reference(&dec->ref_frames[i][j], NULL);
 408 }
 409
 410 static void
 411 vl_mpeg12_begin_frame(struct pipe_video_decoder *decoder)
 412 {
 413    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 414
 415    struct vl_mpeg12_buffer *buf;
 416    struct pipe_sampler_view **sampler_views;
 417    unsigned i;
 418
 419    assert(dec);
 420
 421    buf = dec->current_buffer;
 422    assert(buf);
 423
 424    for (i = 0; i < VL_MAX_PLANES; ++i) {
 425       vl_zscan_upload_quant(&buf->zscan[i], dec->intra_matrix, true);
 426       vl_zscan_upload_quant(&buf->zscan[i], dec->non_intra_matrix, false);
 427    }
 428
 429    vl_vb_map(&buf->vertex_stream, dec->base.context);
 430
 431    sampler_views = buf->zscan_source->get_sampler_view_planes(buf->zscan_source);
 432
 433    assert(sampler_views);
 434
 435    for (i = 0; i < VL_MAX_PLANES; ++i) {
 436       struct pipe_resource *tex = sampler_views[i]->texture;
 437       struct pipe_box rect =
 438       {
 439          0, 0, 0,
 440          tex->width0,
 441          tex->height0,
 442          1
 443       };
 444
 445       buf->tex_transfer[i] = dec->base.context->get_transfer
 446       (
 447          dec->base.context, tex,
 448          0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 449          &rect
 450       );
 451
 452       buf->texels[i] = dec->base.context->transfer_map(dec->base.context, buf->tex_transfer[i]);
 453    }
 454
 455    if (dec->base.entrypoint == PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 456       struct pipe_ycbcr_block *ycbcr_stream[VL_MAX_PLANES];
 457       struct pipe_motionvector *mv_stream[VL_MAX_REF_FRAMES];
 458
 459       for (i = 0; i < VL_MAX_PLANES; ++i)
 460          ycbcr_stream[i] = vl_vb_get_ycbcr_stream(&buf->vertex_stream, i);
 461
 462       for (i = 0; i < VL_MAX_REF_FRAMES; ++i)
 463          mv_stream[i] = vl_vb_get_mv_stream(&buf->vertex_stream, i);
 464
 465       vl_mpg12_bs_set_buffers(&buf->bs, ycbcr_stream, buf->texels, mv_stream);
 466    } else {
 467
 468       for (i = 0; i < VL_MAX_PLANES; ++i)
 469          vl_zscan_set_layout(&buf->zscan[i], dec->zscan_linear);
 470    }
 471 }
 472
 473 static struct pipe_ycbcr_block *
 474 vl_mpeg12_get_ycbcr_stream(struct pipe_video_decoder *decoder, int component)
 475 {
 476    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 477
 478    assert(dec && dec->current_buffer);
 479    assert(component < VL_MAX_PLANES);
 480
 481    return vl_vb_get_ycbcr_stream(&dec->current_buffer->vertex_stream, component);
 482 }
 483
 484 static short *
 485 vl_mpeg12_get_ycbcr_buffer(struct pipe_video_decoder *decoder, int component)
 486 {
 487    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 488
 489    assert(dec && dec->current_buffer);
 490    assert(component < VL_MAX_PLANES);
 491
 492    return dec->current_buffer->texels[component];
 493 }
 494
 495 static unsigned
 496 vl_mpeg12_get_mv_stream_stride(struct pipe_video_decoder *decoder)
 497 {
 498    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 499
 500    assert(dec && dec->current_buffer);
 501
 502    return vl_vb_get_mv_stream_stride(&dec->current_buffer->vertex_stream);
 503 }
 504
 505 static struct pipe_motionvector *
 506 vl_mpeg12_get_mv_stream(struct pipe_video_decoder *decoder, int ref_frame)
 507 {
 508    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 509
 510    assert(dec && dec->current_buffer);
 511
 512    return vl_vb_get_mv_stream(&dec->current_buffer->vertex_stream, ref_frame);
 513 }
 514
 515 static void
 516 vl_mpeg12_decode_bitstream(struct pipe_video_decoder *decoder,
 517                            unsigned num_bytes, const void *data,
 518                            unsigned num_ycbcr_blocks[3])
 519 {
 520    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 521    struct vl_mpeg12_buffer *buf;
 522
 523    unsigned i;
 524
 525    assert(dec && dec->current_buffer);
 526
 527    buf = dec->current_buffer;
 528    assert(buf);
 529
 530    for (i = 0; i < VL_MAX_PLANES; ++i)
 531       vl_zscan_set_layout(&buf->zscan[i], dec->picture_desc.alternate_scan ?
 532                           dec->zscan_alternate : dec->zscan_normal);
 533
 534    vl_mpg12_bs_decode(&buf->bs, num_bytes, data, &dec->picture_desc, num_ycbcr_blocks);
 535 }
 536
 537 static void
 538 vl_mpeg12_end_frame(struct pipe_video_decoder *decoder, unsigned num_ycbcr_blocks[3])
 539 {
 540    struct vl_mpeg12_decoder *dec = (struct vl_mpeg12_decoder *)decoder;
 541    struct pipe_sampler_view **mc_source_sv;
 542    struct pipe_vertex_buffer vb[3];
 543    struct vl_mpeg12_buffer *buf;
 544
 545    unsigned i, j, component;
 546    unsigned nr_components;
 547
 548    assert(dec && dec->current_buffer);
 549
 550    buf = dec->current_buffer;
 551
 552    vl_vb_unmap(&buf->vertex_stream, dec->base.context);
 553
 554    for (i = 0; i < VL_MAX_PLANES; ++i) {
 555       dec->base.context->transfer_unmap(dec->base.context, buf->tex_transfer[i]);
 556       dec->base.context->transfer_destroy(dec->base.context, buf->tex_transfer[i]);
 557    }
 558
 559    vb[0] = dec->quads;
 560    vb[1] = dec->pos;
 561
 562    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_mv);
 563    for (i = 0; i < VL_MAX_PLANES; ++i) {
 564       if (!dec->target_surfaces[i]) continue;
 565
 566       vl_mc_set_surface(&buf->mc[i], dec->target_surfaces[i]);
 567
 568       for (j = 0; j < VL_MAX_REF_FRAMES; ++j) {
 569          if (!dec->ref_frames[j][i]) continue;
 570
 571          vb[2] = vl_vb_get_mv(&buf->vertex_stream, j);;
 572          dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 573
 574          vl_mc_render_ref(&buf->mc[i], dec->ref_frames[j][i]);
 575       }
 576    }
 577
 578    vb[2] = dec->block_num;
 579
 580    dec->base.context->bind_vertex_elements_state(dec->base.context, dec->ves_ycbcr);
 581    for (i = 0; i < VL_MAX_PLANES; ++i) {
 582       if (!num_ycbcr_blocks[i]) continue;
 583
 584       vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, i);
 585       dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 586
 587       vl_zscan_render(&buf->zscan[i] , num_ycbcr_blocks[i]);
 588
 589       if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 590          vl_idct_flush(&buf->idct[i], num_ycbcr_blocks[i]);
 591    }
 592
 593    mc_source_sv = dec->mc_source->get_sampler_view_planes(dec->mc_source);
 594    for (i = 0, component = 0; i < VL_MAX_PLANES; ++i) {
 595       if (!dec->target_surfaces[i]) continue;
 596
 597       nr_components = util_format_get_nr_components(dec->target_surfaces[i]->texture->format);
 598       for (j = 0; j < nr_components; ++j, ++component) {
 599          if (!num_ycbcr_blocks[i]) continue;
 600
 601          vb[1] = vl_vb_get_ycbcr(&buf->vertex_stream, component);
 602          dec->base.context->set_vertex_buffers(dec->base.context, 3, vb);
 603
 604          if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
 605             vl_idct_prepare_stage2(&buf->idct[component]);
 606          else {
 607             dec->base.context->set_fragment_sampler_views(dec->base.context, 1, &mc_source_sv[component]);
 608             dec->base.context->bind_fragment_sampler_states(dec->base.context, 1, &dec->sampler_ycbcr);
 609          }
 610          vl_mc_render_ycbcr(&buf->mc[i], j, num_ycbcr_blocks[component]);
 611       }
 612    }
 613 }
 614
 615 static void
 616 vl_mpeg12_flush(struct pipe_video_decoder *decoder)
 617 {
 618    assert(decoder);
 619
 620    //Noop, for shaders it is much faster to flush everything in end_frame
 621 }
 622
 623 static bool
 624 init_pipe_state(struct vl_mpeg12_decoder *dec)
 625 {
 626    struct pipe_depth_stencil_alpha_state dsa;
 627    struct pipe_sampler_state sampler;
 628    unsigned i;
 629
 630    assert(dec);
 631
 632    memset(&dsa, 0, sizeof dsa);
 633    dsa.depth.enabled = 0;
 634    dsa.depth.writemask = 0;
 635    dsa.depth.func = PIPE_FUNC_ALWAYS;
 636    for (i = 0; i < 2; ++i) {
 637       dsa.stencil[i].enabled = 0;
 638       dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
 639       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
 640       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
 641       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
 642       dsa.stencil[i].valuemask = 0;
 643       dsa.stencil[i].writemask = 0;
 644    }
 645    dsa.alpha.enabled = 0;
 646    dsa.alpha.func = PIPE_FUNC_ALWAYS;
 647    dsa.alpha.ref_value = 0;
 648    dec->dsa = dec->base.context->create_depth_stencil_alpha_state(dec->base.context, &dsa);
 649    dec->base.context->bind_depth_stencil_alpha_state(dec->base.context, dec->dsa);
 650
 651    memset(&sampler, 0, sizeof(sampler));
 652    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 653    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 654    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 655    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 656    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 657    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 658    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 659    sampler.compare_func = PIPE_FUNC_ALWAYS;
 660    sampler.normalized_coords = 1;
 661    dec->sampler_ycbcr = dec->base.context->create_sampler_state(dec->base.context, &sampler);
 662    if (!dec->sampler_ycbcr)
 663       return false;
 664
 665    return true;
 666 }
 667
 668 static const struct format_config*
 669 find_format_config(struct vl_mpeg12_decoder *dec, const struct format_config configs[], unsigned num_configs)
 670 {
 671    struct pipe_screen *screen;
 672    unsigned i;
 673
 674    assert(dec);
 675
 676    screen = dec->base.context->screen;
 677
 678    for (i = 0; i < num_configs; ++i) {
 679       if (!screen->is_format_supported(screen, configs[i].zscan_source_format, PIPE_TEXTURE_2D,
 680                                        1, PIPE_BIND_SAMPLER_VIEW))
 681          continue;
 682
 683       if (configs[i].idct_source_format != PIPE_FORMAT_NONE) {
 684          if (!screen->is_format_supported(screen, configs[i].idct_source_format, PIPE_TEXTURE_2D,
 685                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 686             continue;
 687
 688          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_3D,
 689                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 690             continue;
 691       } else {
 692          if (!screen->is_format_supported(screen, configs[i].mc_source_format, PIPE_TEXTURE_2D,
 693                                           1, PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET))
 694             continue;
 695       }
 696       return &configs[i];
 697    }
 698
 699    return NULL;
 700 }
 701
 702 static bool
 703 init_zscan(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 704 {
 705    unsigned num_channels;
 706
 707    assert(dec);
 708
 709    dec->zscan_source_format = format_config->zscan_source_format;
 710    dec->zscan_linear = vl_zscan_layout(dec->base.context, vl_zscan_linear, dec->blocks_per_line);
 711    dec->zscan_normal = vl_zscan_layout(dec->base.context, vl_zscan_normal, dec->blocks_per_line);
 712    dec->zscan_alternate = vl_zscan_layout(dec->base.context, vl_zscan_alternate, dec->blocks_per_line);
 713
 714    num_channels = dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT ? 4 : 1;
 715
 716    if (!vl_zscan_init(&dec->zscan_y, dec->base.context, dec->base.width, dec->base.height,
 717                       dec->blocks_per_line, dec->num_blocks, num_channels))
 718       return false;
 719
 720    if (!vl_zscan_init(&dec->zscan_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 721                       dec->blocks_per_line, dec->num_blocks, num_channels))
 722       return false;
 723
 724    return true;
 725 }
 726
 727 static bool
 728 init_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 729 {
 730    unsigned nr_of_idct_render_targets, max_inst;
 731    enum pipe_format formats[3];
 732
 733    struct pipe_sampler_view *matrix = NULL;
 734
 735    nr_of_idct_render_targets = dec->base.context->screen->get_param
 736    (
 737       dec->base.context->screen, PIPE_CAP_MAX_RENDER_TARGETS
 738    );
 739
 740    max_inst = dec->base.context->screen->get_shader_param
 741    (
 742       dec->base.context->screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_INSTRUCTIONS
 743    );
 744
 745    // Just assume we need 32 inst per render target, not 100% true, but should work in most cases
 746    if (nr_of_idct_render_targets >= 4 && max_inst >= 32*4)
 747       // more than 4 render targets usually doesn't makes any seens
 748       nr_of_idct_render_targets = 4;
 749    else
 750       nr_of_idct_render_targets = 1;
 751
 752    formats[0] = formats[1] = formats[2] = format_config->idct_source_format;
 753    dec->idct_source = vl_video_buffer_create_ex
 754    (
 755       dec->base.context, dec->base.width / 4, dec->base.height, 1,
 756       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 757    );
 758
 759    if (!dec->idct_source)
 760       goto error_idct_source;
 761
 762    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 763    dec->mc_source = vl_video_buffer_create_ex
 764    (
 765       dec->base.context, dec->base.width / nr_of_idct_render_targets,
 766       dec->base.height / 4, nr_of_idct_render_targets,
 767       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 768    );
 769
 770    if (!dec->mc_source)
 771       goto error_mc_source;
 772
 773    if (!(matrix = vl_idct_upload_matrix(dec->base.context, format_config->idct_scale)))
 774       goto error_matrix;
 775
 776    if (!vl_idct_init(&dec->idct_y, dec->base.context, dec->base.width, dec->base.height,
 777                      nr_of_idct_render_targets, matrix, matrix))
 778       goto error_y;
 779
 780    if(!vl_idct_init(&dec->idct_c, dec->base.context, dec->chroma_width, dec->chroma_height,
 781                     nr_of_idct_render_targets, matrix, matrix))
 782       goto error_c;
 783
 784    pipe_sampler_view_reference(&matrix, NULL);
 785
 786    return true;
 787
 788 error_c:
 789    vl_idct_cleanup(&dec->idct_y);
 790
 791 error_y:
 792    pipe_sampler_view_reference(&matrix, NULL);
 793
 794 error_matrix:
 795    dec->mc_source->destroy(dec->mc_source);
 796
 797 error_mc_source:
 798    dec->idct_source->destroy(dec->idct_source);
 799
 800 error_idct_source:
 801    return false;
 802 }
 803
 804 static bool
 805 init_mc_source_widthout_idct(struct vl_mpeg12_decoder *dec, const struct format_config* format_config)
 806 {
 807    enum pipe_format formats[3];
 808
 809    formats[0] = formats[1] = formats[2] = format_config->mc_source_format;
 810    dec->mc_source = vl_video_buffer_create_ex
 811    (
 812       dec->base.context, dec->base.width, dec->base.height, 1,
 813       dec->base.chroma_format, formats, PIPE_USAGE_STATIC
 814    );
 815
 816    return dec->mc_source != NULL;
 817 }
 818
 819 static void
 820 mc_vert_shader_callback(void *priv, struct vl_mc *mc,
 821                         struct ureg_program *shader,
 822                         unsigned first_output,
 823                         struct ureg_dst tex)
 824 {
 825    struct vl_mpeg12_decoder *dec = priv;
 826    struct ureg_dst o_vtex;
 827
 828    assert(priv && mc);
 829    assert(shader);
 830
 831    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 832       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
 833       vl_idct_stage2_vert_shader(idct, shader, first_output, tex);
 834    } else {
 835       o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output);
 836       ureg_MOV(shader, ureg_writemask(o_vtex, TGSI_WRITEMASK_XY), ureg_src(tex));
 837    }
 838 }
 839
 840 static void
 841 mc_frag_shader_callback(void *priv, struct vl_mc *mc,
 842                         struct ureg_program *shader,
 843                         unsigned first_input,
 844                         struct ureg_dst dst)
 845 {
 846    struct vl_mpeg12_decoder *dec = priv;
 847    struct ureg_src src, sampler;
 848
 849    assert(priv && mc);
 850    assert(shader);
 851
 852    if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 853       struct vl_idct *idct = mc == &dec->mc_y ? &dec->idct_y : &dec->idct_c;
 854       vl_idct_stage2_frag_shader(idct, shader, first_input, dst);
 855    } else {
 856       src = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input, TGSI_INTERPOLATE_LINEAR);
 857       sampler = ureg_DECL_sampler(shader, 0);
 858       ureg_TEX(shader, dst, TGSI_TEXTURE_2D, src, sampler);
 859    }
 860 }
 861
 862 struct pipe_video_decoder *
 863 vl_create_mpeg12_decoder(struct pipe_context *context,
 864                          enum pipe_video_profile profile,
 865                          enum pipe_video_entrypoint entrypoint,
 866                          enum pipe_video_chroma_format chroma_format,
 867                          unsigned width, unsigned height)
 868 {
 869    const unsigned block_size_pixels = BLOCK_WIDTH * BLOCK_HEIGHT;
 870    const struct format_config *format_config;
 871    struct vl_mpeg12_decoder *dec;
 872
 873    assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
 874
 875    dec = CALLOC_STRUCT(vl_mpeg12_decoder);
 876
 877    if (!dec)
 878       return NULL;
 879
 880    dec->base.context = context;
 881    dec->base.profile = profile;
 882    dec->base.entrypoint = entrypoint;
 883    dec->base.chroma_format = chroma_format;
 884    dec->base.width = width;
 885    dec->base.height = height;
 886
 887    dec->base.destroy = vl_mpeg12_destroy;
 888    dec->base.create_buffer = vl_mpeg12_create_buffer;
 889    dec->base.destroy_buffer = vl_mpeg12_destroy_buffer;
 890    dec->base.set_decode_buffer = vl_mpeg12_set_decode_buffer;
 891    dec->base.set_picture_parameters = vl_mpeg12_set_picture_parameters;
 892    dec->base.set_quant_matrix = vl_mpeg12_set_quant_matrix;
 893    dec->base.set_decode_target = vl_mpeg12_set_decode_target;
 894    dec->base.set_reference_frames = vl_mpeg12_set_reference_frames;
 895    dec->base.begin_frame = vl_mpeg12_begin_frame;
 896    dec->base.get_ycbcr_stream = vl_mpeg12_get_ycbcr_stream;
 897    dec->base.get_ycbcr_buffer = vl_mpeg12_get_ycbcr_buffer;
 898    dec->base.get_mv_stream_stride = vl_mpeg12_get_mv_stream_stride;
 899    dec->base.get_mv_stream = vl_mpeg12_get_mv_stream;
 900    dec->base.decode_bitstream = vl_mpeg12_decode_bitstream;
 901    dec->base.end_frame = vl_mpeg12_end_frame;
 902    dec->base.flush = vl_mpeg12_flush;
 903
 904    dec->blocks_per_line = MAX2(util_next_power_of_two(dec->base.width) / block_size_pixels, 4);
 905    dec->num_blocks = (dec->base.width * dec->base.height) / block_size_pixels;
 906
 907    dec->quads = vl_vb_upload_quads(dec->base.context);
 908    dec->pos = vl_vb_upload_pos(
 909       dec->base.context,
 910       dec->base.width / MACROBLOCK_WIDTH,
 911       dec->base.height / MACROBLOCK_HEIGHT
 912    );
 913    dec->block_num = vl_vb_upload_block_num(dec->base.context, dec->num_blocks);
 914
 915    dec->ves_ycbcr = vl_vb_get_ves_ycbcr(dec->base.context);
 916    dec->ves_mv = vl_vb_get_ves_mv(dec->base.context);
 917
 918    /* TODO: Implement 422, 444 */
 919    assert(dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 920
 921    if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
 922       dec->chroma_width = dec->base.width / 2;
 923       dec->chroma_height = dec->base.height / 2;
 924    } else if (dec->base.chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422) {
 925       dec->chroma_width = dec->base.width;
 926       dec->chroma_height = dec->base.height / 2;
 927    } else {
 928       dec->chroma_width = dec->base.width;
 929       dec->chroma_height = dec->base.height;
 930    }
 931
 932    switch (entrypoint) {
 933    case PIPE_VIDEO_ENTRYPOINT_BITSTREAM:
 934       format_config = find_format_config(dec, bitstream_format_config, num_bitstream_format_configs);
 935       break;
 936
 937    case PIPE_VIDEO_ENTRYPOINT_IDCT:
 938       format_config = find_format_config(dec, idct_format_config, num_idct_format_configs);
 939       break;
 940
 941    case PIPE_VIDEO_ENTRYPOINT_MC:
 942       format_config = find_format_config(dec, mc_format_config, num_mc_format_configs);
 943       break;
 944
 945    default:
 946       assert(0);
 947       return NULL;
 948    }
 949
 950    if (!format_config)
 951       return NULL;
 952
 953    if (!init_zscan(dec, format_config))
 954       goto error_zscan;
 955
 956    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 957       if (!init_idct(dec, format_config))
 958          goto error_sources;
 959    } else {
 960       if (!init_mc_source_widthout_idct(dec, format_config))
 961          goto error_sources;
 962    }
 963
 964    if (!vl_mc_init(&dec->mc_y, dec->base.context, dec->base.width, dec->base.height,
 965                    MACROBLOCK_HEIGHT, format_config->mc_scale,
 966                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
 967       goto error_mc_y;
 968
 969    // TODO
 970    if (!vl_mc_init(&dec->mc_c, dec->base.context, dec->base.width, dec->base.height,
 971                    BLOCK_HEIGHT, format_config->mc_scale,
 972                    mc_vert_shader_callback, mc_frag_shader_callback, dec))
 973       goto error_mc_c;
 974
 975    if (!init_pipe_state(dec))
 976       goto error_pipe_state;
 977
 978    return &dec->base;
 979
 980 error_pipe_state:
 981    vl_mc_cleanup(&dec->mc_c);
 982
 983 error_mc_c:
 984    vl_mc_cleanup(&dec->mc_y);
 985
 986 error_mc_y:
 987    if (entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
 988       vl_idct_cleanup(&dec->idct_y);
 989       vl_idct_cleanup(&dec->idct_c);
 990       dec->idct_source->destroy(dec->idct_source);
 991    }
 992    dec->mc_source->destroy(dec->mc_source);
 993
 994 error_sources:
 995    vl_zscan_cleanup(&dec->zscan_y);
 996    vl_zscan_cleanup(&dec->zscan_c);
 997
 998 error_zscan:
 999    FREE(dec);
1000    return NULL;
1001 }