FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
dxva2_hevc.c
Go to the documentation of this file.
1 /*
2  * DXVA2 HEVC HW acceleration.
3  *
4  * copyright (c) 2014 - 2015 Hendrik Leppkes
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include "libavutil/avassert.h"
24 
25 #include "dxva2_internal.h"
26 #include "hevc.h"
27 
28 #define MAX_SLICES 256
29 
31  DXVA_PicParams_HEVC pp;
32  DXVA_Qmatrix_HEVC qm;
33  unsigned slice_count;
34  DXVA_Slice_HEVC_Short slice_short[MAX_SLICES];
36  unsigned bitstream_size;
37 };
38 
39 static void fill_picture_entry(DXVA_PicEntry_HEVC *pic,
40  unsigned index, unsigned flag)
41 {
42  av_assert0((index & 0x7f) == index && (flag & 0x01) == flag);
43  pic->bPicEntry = index | (flag << 7);
44 }
45 
46 static int get_refpic_index(const DXVA_PicParams_HEVC *pp, int surface_index)
47 {
48  int i;
49  for (i = 0; i < FF_ARRAY_ELEMS(pp->RefPicList); i++) {
50  if ((pp->RefPicList[i].bPicEntry & 0x7f) == surface_index)
51  return i;
52  }
53  return 0xff;
54 }
55 
56 static void fill_picture_parameters(struct dxva_context *ctx, const HEVCContext *h,
57  DXVA_PicParams_HEVC *pp)
58 {
59  const HEVCFrame *current_picture = h->ref;
60  int i, j, k;
61 
62  memset(pp, 0, sizeof(*pp));
63 
64  pp->PicWidthInMinCbsY = h->sps->min_cb_width;
65  pp->PicHeightInMinCbsY = h->sps->min_cb_height;
66 
67  pp->wFormatAndSequenceInfoFlags = (h->sps->chroma_format_idc << 0) |
68  (h->sps->separate_colour_plane_flag << 2) |
69  ((h->sps->bit_depth - 8) << 3) |
70  ((h->sps->bit_depth - 8) << 6) |
71  ((h->sps->log2_max_poc_lsb - 4) << 9) |
72  (0 << 13) |
73  (0 << 14) |
74  (0 << 15);
75 
76  fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(ctx, current_picture->frame), 0);
77 
78  pp->sps_max_dec_pic_buffering_minus1 = h->sps->temporal_layer[h->sps->max_sub_layers - 1].max_dec_pic_buffering - 1;
79  pp->log2_min_luma_coding_block_size_minus3 = h->sps->log2_min_cb_size - 3;
80  pp->log2_diff_max_min_luma_coding_block_size = h->sps->log2_diff_max_min_coding_block_size;
81  pp->log2_min_transform_block_size_minus2 = h->sps->log2_min_tb_size - 2;
82  pp->log2_diff_max_min_transform_block_size = h->sps->log2_max_trafo_size - h->sps->log2_min_tb_size;
83  pp->max_transform_hierarchy_depth_inter = h->sps->max_transform_hierarchy_depth_inter;
84  pp->max_transform_hierarchy_depth_intra = h->sps->max_transform_hierarchy_depth_intra;
85  pp->num_short_term_ref_pic_sets = h->sps->nb_st_rps;
86  pp->num_long_term_ref_pics_sps = h->sps->num_long_term_ref_pics_sps;
87 
88  pp->num_ref_idx_l0_default_active_minus1 = h->pps->num_ref_idx_l0_default_active - 1;
89  pp->num_ref_idx_l1_default_active_minus1 = h->pps->num_ref_idx_l1_default_active - 1;
90  pp->init_qp_minus26 = h->pps->pic_init_qp_minus26;
91 
93  pp->ucNumDeltaPocsOfRefRpsIdx = h->sh.short_term_rps->num_delta_pocs;
94  pp->wNumBitsForShortTermRPSInSlice = h->sh.short_term_ref_pic_set_size;
95  }
96 
97  pp->dwCodingParamToolFlags = (h->sps->scaling_list_enable_flag << 0) |
98  (h->sps->amp_enabled_flag << 1) |
99  (h->sps->sao_enabled << 2) |
100  (h->sps->pcm_enabled_flag << 3) |
101  ((h->sps->pcm_enabled_flag ? (h->sps->pcm.bit_depth - 1) : 0) << 4) |
102  ((h->sps->pcm_enabled_flag ? (h->sps->pcm.bit_depth_chroma - 1) : 0) << 8) |
103  ((h->sps->pcm_enabled_flag ? (h->sps->pcm.log2_min_pcm_cb_size - 3) : 0) << 12) |
104  ((h->sps->pcm_enabled_flag ? (h->sps->pcm.log2_max_pcm_cb_size - h->sps->pcm.log2_min_pcm_cb_size) : 0) << 14) |
105  (h->sps->pcm.loop_filter_disable_flag << 16) |
107  (h->sps->sps_temporal_mvp_enabled_flag << 18) |
110  (h->pps->output_flag_present_flag << 21) |
111  (h->pps->num_extra_slice_header_bits << 22) |
112  (h->pps->sign_data_hiding_flag << 25) |
113  (h->pps->cabac_init_present_flag << 26) |
114  (0 << 27);
115 
116  pp->dwCodingSettingPicturePropertyFlags = (h->pps->constrained_intra_pred_flag << 0) |
117  (h->pps->transform_skip_enabled_flag << 1) |
118  (h->pps->cu_qp_delta_enabled_flag << 2) |
120  (h->pps->weighted_pred_flag << 4) |
121  (h->pps->weighted_bipred_flag << 5) |
123  (h->pps->tiles_enabled_flag << 7) |
125  (h->pps->uniform_spacing_flag << 9) |
129  (h->pps->disable_dbf << 13) |
132  (IS_IRAP(h) << 16) |
133  (IS_IDR(h) << 17) |
134  /* IntraPicFlag */
135  (IS_IRAP(h) << 18) |
136  (0 << 19);
137  pp->pps_cb_qp_offset = h->pps->cb_qp_offset;
138  pp->pps_cr_qp_offset = h->pps->cr_qp_offset;
139  if (h->pps->tiles_enabled_flag) {
140  pp->num_tile_columns_minus1 = h->pps->num_tile_columns - 1;
141  pp->num_tile_rows_minus1 = h->pps->num_tile_rows - 1;
142 
143  if (!h->pps->uniform_spacing_flag) {
144  for (i = 0; i < h->pps->num_tile_columns; i++)
145  pp->column_width_minus1[i] = h->pps->column_width[i] - 1;
146 
147  for (i = 0; i < h->pps->num_tile_rows; i++)
148  pp->row_height_minus1[i] = h->pps->row_height[i] - 1;
149  }
150  }
151 
152  pp->diff_cu_qp_delta_depth = h->pps->diff_cu_qp_delta_depth;
153  pp->pps_beta_offset_div2 = h->pps->beta_offset / 2;
154  pp->pps_tc_offset_div2 = h->pps->tc_offset / 2;
155  pp->log2_parallel_merge_level_minus2 = h->pps->log2_parallel_merge_level - 2;
156  pp->CurrPicOrderCntVal = h->poc;
157 
158  // empty the lists
159  memset(&pp->RefPicList, 0xff, sizeof(pp->RefPicList));
160  memset(&pp->RefPicSetStCurrBefore, 0xff, sizeof(pp->RefPicSetStCurrBefore));
161  memset(&pp->RefPicSetStCurrAfter, 0xff, sizeof(pp->RefPicSetStCurrAfter));
162  memset(&pp->RefPicSetLtCurr, 0xff, sizeof(pp->RefPicSetLtCurr));
163 
164  // fill RefPicList from the DPB
165  for (i = 0, j = 0; i < FF_ARRAY_ELEMS(h->DPB); i++) {
166  const HEVCFrame *frame = &h->DPB[i];
167  if (frame != current_picture && (frame->flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF))) {
168  fill_picture_entry(&pp->RefPicList[j], ff_dxva2_get_surface_index(ctx, frame->frame), !!(frame->flags & HEVC_FRAME_FLAG_LONG_REF));
169  pp->PicOrderCntValList[j] = frame->poc;
170  j++;
171  }
172  }
173 
174  #define DO_REF_LIST(ref_idx, ref_list) { \
175  const RefPicList *rpl = &h->rps[ref_idx]; \
176  av_assert0(rpl->nb_refs <= FF_ARRAY_ELEMS(pp->ref_list)); \
177  for (j = 0, k = 0; j < rpl->nb_refs; j++) { \
178  if (rpl->ref[j]) { \
179  pp->ref_list[k] = get_refpic_index(pp, ff_dxva2_get_surface_index(ctx, rpl->ref[j]->frame)); \
180  k++; \
181  } \
182  } \
183  }
184 
185  // Fill short term and long term lists
186  DO_REF_LIST(ST_CURR_BEF, RefPicSetStCurrBefore);
187  DO_REF_LIST(ST_CURR_AFT, RefPicSetStCurrAfter);
188  DO_REF_LIST(LT_CURR, RefPicSetLtCurr);
189 
190  pp->StatusReportFeedbackNumber = 1 + ctx->report_id++;
191 }
192 
193 static void fill_scaling_lists(struct dxva_context *ctx, const HEVCContext *h, DXVA_Qmatrix_HEVC *qm)
194 {
195  unsigned i, j, pos;
197  &h->pps->scaling_list : &h->sps->scaling_list;
198 
199  memset(qm, 0, sizeof(*qm));
200  for (i = 0; i < 6; i++) {
201  for (j = 0; j < 16; j++) {
203  qm->ucScalingLists0[i][j] = sl->sl[0][i][pos];
204  }
205 
206  for (j = 0; j < 64; j++) {
208  qm->ucScalingLists1[i][j] = sl->sl[1][i][pos];
209  qm->ucScalingLists2[i][j] = sl->sl[2][i][pos];
210 
211  if (i < 2)
212  qm->ucScalingLists3[i][j] = sl->sl[3][i][pos];
213  }
214 
215  qm->ucScalingListDCCoefSizeID2[i] = sl->sl_dc[0][i];
216  if (i < 2)
217  qm->ucScalingListDCCoefSizeID3[i] = sl->sl_dc[1][i];
218  }
219 }
220 
221 static void fill_slice_short(DXVA_Slice_HEVC_Short *slice,
222  unsigned position, unsigned size)
223 {
224  memset(slice, 0, sizeof(*slice));
225  slice->BSNALunitDataLocation = position;
226  slice->SliceBytesInBuffer = size;
227  slice->wBadSliceChopping = 0;
228 }
229 
231  DXVA2_DecodeBufferDesc *bs,
232  DXVA2_DecodeBufferDesc *sc)
233 {
234  const HEVCContext *h = avctx->priv_data;
235  struct dxva_context *ctx = avctx->hwaccel_context;
236  const HEVCFrame *current_picture = h->ref;
237  struct hevc_dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
238  DXVA_Slice_HEVC_Short *slice = NULL;
239  void *dxva_data_ptr;
240  uint8_t *dxva_data, *current, *end;
241  unsigned dxva_size;
242  void *slice_data;
243  unsigned slice_size;
244  unsigned padding;
245  unsigned i;
246 
247  /* Create an annex B bitstream buffer with only slice NAL and finalize slice */
248  if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder,
249  DXVA2_BitStreamDateBufferType,
250  &dxva_data_ptr, &dxva_size)))
251  return -1;
252 
253  dxva_data = dxva_data_ptr;
254  current = dxva_data;
255  end = dxva_data + dxva_size;
256 
257  for (i = 0; i < ctx_pic->slice_count; i++) {
258  static const uint8_t start_code[] = { 0, 0, 1 };
259  static const unsigned start_code_size = sizeof(start_code);
260  unsigned position, size;
261 
262  slice = &ctx_pic->slice_short[i];
263 
264  position = slice->BSNALunitDataLocation;
265  size = slice->SliceBytesInBuffer;
266  if (start_code_size + size > end - current) {
267  av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
268  break;
269  }
270 
271  slice->BSNALunitDataLocation = current - dxva_data;
272  slice->SliceBytesInBuffer = start_code_size + size;
273 
274  memcpy(current, start_code, start_code_size);
275  current += start_code_size;
276 
277  memcpy(current, &ctx_pic->bitstream[position], size);
278  current += size;
279  }
280  padding = FFMIN(128 - ((current - dxva_data) & 127), end - current);
281  if (slice && padding > 0) {
282  memset(current, 0, padding);
283  current += padding;
284 
285  slice->SliceBytesInBuffer += padding;
286  }
287  if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder,
288  DXVA2_BitStreamDateBufferType)))
289  return -1;
290  if (i < ctx_pic->slice_count)
291  return -1;
292 
293  memset(bs, 0, sizeof(*bs));
294  bs->CompressedBufferType = DXVA2_BitStreamDateBufferType;
295  bs->DataSize = current - dxva_data;
296  bs->NumMBsInBuffer = 0;
297 
298  slice_data = ctx_pic->slice_short;
299  slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_short);
300 
301  av_assert0((bs->DataSize & 127) == 0);
302  return ff_dxva2_commit_buffer(avctx, ctx, sc,
303  DXVA2_SliceControlBufferType,
304  slice_data, slice_size, 0);
305 }
306 
307 
309  av_unused const uint8_t *buffer,
310  av_unused uint32_t size)
311 {
312  const HEVCContext *h = avctx->priv_data;
313  struct dxva_context *ctx = avctx->hwaccel_context;
315 
316  if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0)
317  return -1;
318  av_assert0(ctx_pic);
319 
320  /* Fill up DXVA_PicParams_HEVC */
321  fill_picture_parameters(ctx, h, &ctx_pic->pp);
322 
323  /* Fill up DXVA_Qmatrix_HEVC */
324  fill_scaling_lists(ctx, h, &ctx_pic->qm);
325 
326  ctx_pic->slice_count = 0;
327  ctx_pic->bitstream_size = 0;
328  ctx_pic->bitstream = NULL;
329  return 0;
330 }
331 
333  const uint8_t *buffer,
334  uint32_t size)
335 {
336  const HEVCContext *h = avctx->priv_data;
337  const HEVCFrame *current_picture = h->ref;
338  struct hevc_dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
339  unsigned position;
340 
341  if (ctx_pic->slice_count >= MAX_SLICES)
342  return -1;
343 
344  if (!ctx_pic->bitstream)
345  ctx_pic->bitstream = buffer;
346  ctx_pic->bitstream_size += size;
347 
348  position = buffer - ctx_pic->bitstream;
349  fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count], position, size);
350  ctx_pic->slice_count++;
351 
352  return 0;
353 }
354 
356 {
357  HEVCContext *h = avctx->priv_data;
359  int scale = ctx_pic->pp.dwCodingParamToolFlags & 1;
360  int ret;
361 
362  if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
363  return -1;
364 
365  ret = ff_dxva2_common_end_frame(avctx, h->ref->frame,
366  &ctx_pic->pp, sizeof(ctx_pic->pp),
367  scale ? &ctx_pic->qm : NULL, scale ? sizeof(ctx_pic->qm) : 0,
369  return ret;
370 }
371 
373  .name = "hevc_dxva2",
374  .type = AVMEDIA_TYPE_VIDEO,
375  .id = AV_CODEC_ID_HEVC,
376  .pix_fmt = AV_PIX_FMT_DXVA2_VLD,
377  .start_frame = dxva2_hevc_start_frame,
378  .decode_slice = dxva2_hevc_decode_slice,
379  .end_frame = dxva2_hevc_end_frame,
380  .frame_priv_data_size = sizeof(struct hevc_dxva2_picture_context),
381 };