avcodec/aacdec_fixed: use 64bit to avoid overflow in rounding in apply_dependent_coup...
[ffmpeg.git] / libavcodec / dxva2_hevc.c
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_data.h"
27 #include "hevcdec.h"
28
29 #define MAX_SLICES 256
30
31 struct hevc_dxva2_picture_context {
32     DXVA_PicParams_HEVC   pp;
33     DXVA_Qmatrix_HEVC     qm;
34     unsigned              slice_count;
35     DXVA_Slice_HEVC_Short slice_short[MAX_SLICES];
36     const uint8_t         *bitstream;
37     unsigned              bitstream_size;
38 };
39
40 static void fill_picture_entry(DXVA_PicEntry_HEVC *pic,
41                                unsigned index, unsigned flag)
42 {
43     av_assert0((index & 0x7f) == index && (flag & 0x01) == flag);
44     pic->bPicEntry = index | (flag << 7);
45 }
46
47 static int get_refpic_index(const DXVA_PicParams_HEVC *pp, int surface_index)
48 {
49     int i;
50     for (i = 0; i < FF_ARRAY_ELEMS(pp->RefPicList); i++) {
51         if ((pp->RefPicList[i].bPicEntry & 0x7f) == surface_index)
52           return i;
53     }
54     return 0xff;
55 }
56
57 static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const HEVCContext *h,
58                                     DXVA_PicParams_HEVC *pp)
59 {
60     const HEVCFrame *current_picture = h->ref;
61     const HEVCSPS *sps = h->ps.sps;
62     const HEVCPPS *pps = h->ps.pps;
63     int i, j;
64
65     memset(pp, 0, sizeof(*pp));
66
67     pp->PicWidthInMinCbsY  = sps->min_cb_width;
68     pp->PicHeightInMinCbsY = sps->min_cb_height;
69
70     pp->wFormatAndSequenceInfoFlags = (sps->chroma_format_idc             <<  0) |
71                                       (sps->separate_colour_plane_flag    <<  2) |
72                                       ((sps->bit_depth - 8)               <<  3) |
73                                       ((sps->bit_depth - 8)               <<  6) |
74                                       ((sps->log2_max_poc_lsb - 4)        <<  9) |
75                                       (0                                  << 13) |
76                                       (0                                  << 14) |
77                                       (0                                  << 15);
78
79     fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, current_picture->frame), 0);
80
81     pp->sps_max_dec_pic_buffering_minus1         = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1;
82     pp->log2_min_luma_coding_block_size_minus3   = sps->log2_min_cb_size - 3;
83     pp->log2_diff_max_min_luma_coding_block_size = sps->log2_diff_max_min_coding_block_size;
84     pp->log2_min_transform_block_size_minus2     = sps->log2_min_tb_size - 2;
85     pp->log2_diff_max_min_transform_block_size   = sps->log2_max_trafo_size  - sps->log2_min_tb_size;
86     pp->max_transform_hierarchy_depth_inter      = sps->max_transform_hierarchy_depth_inter;
87     pp->max_transform_hierarchy_depth_intra      = sps->max_transform_hierarchy_depth_intra;
88     pp->num_short_term_ref_pic_sets              = sps->nb_st_rps;
89     pp->num_long_term_ref_pics_sps               = sps->num_long_term_ref_pics_sps;
90
91     pp->num_ref_idx_l0_default_active_minus1     = pps->num_ref_idx_l0_default_active - 1;
92     pp->num_ref_idx_l1_default_active_minus1     = pps->num_ref_idx_l1_default_active - 1;
93     pp->init_qp_minus26                          = pps->pic_init_qp_minus26;
94
95     if (h->sh.short_term_ref_pic_set_sps_flag == 0 && h->sh.short_term_rps) {
96         pp->ucNumDeltaPocsOfRefRpsIdx            = h->sh.short_term_rps->rps_idx_num_delta_pocs;
97         pp->wNumBitsForShortTermRPSInSlice       = h->sh.short_term_ref_pic_set_size;
98     }
99
100     pp->dwCodingParamToolFlags = (sps->scaling_list_enable_flag                  <<  0) |
101                                  (sps->amp_enabled_flag                          <<  1) |
102                                  (sps->sao_enabled                               <<  2) |
103                                  (sps->pcm_enabled_flag                          <<  3) |
104                                  ((sps->pcm_enabled_flag ? (sps->pcm.bit_depth - 1) : 0)            <<  4) |
105                                  ((sps->pcm_enabled_flag ? (sps->pcm.bit_depth_chroma - 1) : 0)     <<  8) |
106                                  ((sps->pcm_enabled_flag ? (sps->pcm.log2_min_pcm_cb_size - 3) : 0) << 12) |
107                                  ((sps->pcm_enabled_flag ? (sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size) : 0) << 14) |
108                                  (sps->pcm.loop_filter_disable_flag              << 16) |
109                                  (sps->long_term_ref_pics_present_flag           << 17) |
110                                  (sps->sps_temporal_mvp_enabled_flag             << 18) |
111                                  (sps->sps_strong_intra_smoothing_enable_flag    << 19) |
112                                  (pps->dependent_slice_segments_enabled_flag     << 20) |
113                                  (pps->output_flag_present_flag                  << 21) |
114                                  (pps->num_extra_slice_header_bits               << 22) |
115                                  (pps->sign_data_hiding_flag                     << 25) |
116                                  (pps->cabac_init_present_flag                   << 26) |
117                                  (0                                              << 27);
118
119     pp->dwCodingSettingPicturePropertyFlags = (pps->constrained_intra_pred_flag                   <<  0) |
120                                               (pps->transform_skip_enabled_flag                   <<  1) |
121                                               (pps->cu_qp_delta_enabled_flag                      <<  2) |
122                                               (pps->pic_slice_level_chroma_qp_offsets_present_flag <<  3) |
123                                               (pps->weighted_pred_flag                            <<  4) |
124                                               (pps->weighted_bipred_flag                          <<  5) |
125                                               (pps->transquant_bypass_enable_flag                 <<  6) |
126                                               (pps->tiles_enabled_flag                            <<  7) |
127                                               (pps->entropy_coding_sync_enabled_flag              <<  8) |
128                                               (pps->uniform_spacing_flag                          <<  9) |
129                                               ((pps->tiles_enabled_flag ? pps->loop_filter_across_tiles_enabled_flag : 0) << 10) |
130                                               (pps->seq_loop_filter_across_slices_enabled_flag    << 11) |
131                                               (pps->deblocking_filter_override_enabled_flag       << 12) |
132                                               (pps->disable_dbf                                   << 13) |
133                                               (pps->lists_modification_present_flag               << 14) |
134                                               (pps->slice_header_extension_present_flag           << 15) |
135                                               (IS_IRAP(h)                                         << 16) |
136                                               (IS_IDR(h)                                          << 17) |
137                                               /* IntraPicFlag */
138                                               (IS_IRAP(h)                                         << 18) |
139                                               (0                                                  << 19);
140     pp->pps_cb_qp_offset            = pps->cb_qp_offset;
141     pp->pps_cr_qp_offset            = pps->cr_qp_offset;
142     if (pps->tiles_enabled_flag) {
143         pp->num_tile_columns_minus1 = pps->num_tile_columns - 1;
144         pp->num_tile_rows_minus1    = pps->num_tile_rows - 1;
145
146         if (!pps->uniform_spacing_flag) {
147             for (i = 0; i < pps->num_tile_columns; i++)
148                 pp->column_width_minus1[i] = pps->column_width[i] - 1;
149
150             for (i = 0; i < pps->num_tile_rows; i++)
151                 pp->row_height_minus1[i] = pps->row_height[i] - 1;
152         }
153     }
154
155     pp->diff_cu_qp_delta_depth           = pps->diff_cu_qp_delta_depth;
156     pp->pps_beta_offset_div2             = pps->beta_offset / 2;
157     pp->pps_tc_offset_div2               = pps->tc_offset / 2;
158     pp->log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level - 2;
159     pp->CurrPicOrderCntVal               = h->poc;
160
161     // fill RefPicList from the DPB
162     for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->RefPicList); i++) {
163         const HEVCFrame *frame = NULL;
164         while (!frame && j < FF_ARRAY_ELEMS(h->DPB)) {
165             if (&h->DPB[j] != current_picture && (h->DPB[j].flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF)))
166                 frame = &h->DPB[j];
167             j++;
168         }
169
170         if (frame) {
171             fill_picture_entry(&pp->RefPicList[i], ff_dxva2_get_surface_index(avctx, ctx, frame->frame), !!(frame->flags & HEVC_FRAME_FLAG_LONG_REF));
172             pp->PicOrderCntValList[i] = frame->poc;
173         } else {
174             pp->RefPicList[i].bPicEntry = 0xff;
175             pp->PicOrderCntValList[i]   = 0;
176         }
177     }
178
179     #define DO_REF_LIST(ref_idx, ref_list) { \
180         const RefPicList *rpl = &h->rps[ref_idx]; \
181         for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->ref_list); i++) { \
182             const HEVCFrame *frame = NULL; \
183             while (!frame && j < rpl->nb_refs) \
184                 frame = rpl->ref[j++]; \
185             if (frame) \
186                 pp->ref_list[i] = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, frame->frame)); \
187             else \
188                 pp->ref_list[i] = 0xff; \
189         } \
190     }
191
192     // Fill short term and long term lists
193     DO_REF_LIST(ST_CURR_BEF, RefPicSetStCurrBefore);
194     DO_REF_LIST(ST_CURR_AFT, RefPicSetStCurrAfter);
195     DO_REF_LIST(LT_CURR, RefPicSetLtCurr);
196
197     pp->StatusReportFeedbackNumber = 1 + DXVA_CONTEXT_REPORT_ID(avctx, ctx)++;
198 }
199
200 static void fill_scaling_lists(AVDXVAContext *ctx, const HEVCContext *h, DXVA_Qmatrix_HEVC *qm)
201 {
202     unsigned i, j, pos;
203     const ScalingList *sl = h->ps.pps->scaling_list_data_present_flag ?
204                             &h->ps.pps->scaling_list : &h->ps.sps->scaling_list;
205
206     memset(qm, 0, sizeof(*qm));
207     for (i = 0; i < 6; i++) {
208         for (j = 0; j < 16; j++) {
209             pos = 4 * ff_hevc_diag_scan4x4_y[j] + ff_hevc_diag_scan4x4_x[j];
210             qm->ucScalingLists0[i][j] = sl->sl[0][i][pos];
211         }
212
213         for (j = 0; j < 64; j++) {
214             pos = 8 * ff_hevc_diag_scan8x8_y[j] + ff_hevc_diag_scan8x8_x[j];
215             qm->ucScalingLists1[i][j] = sl->sl[1][i][pos];
216             qm->ucScalingLists2[i][j] = sl->sl[2][i][pos];
217
218             if (i < 2)
219                 qm->ucScalingLists3[i][j] = sl->sl[3][i * 3][pos];
220         }
221
222         qm->ucScalingListDCCoefSizeID2[i] = sl->sl_dc[0][i];
223         if (i < 2)
224             qm->ucScalingListDCCoefSizeID3[i] = sl->sl_dc[1][i * 3];
225     }
226 }
227
228 static void fill_slice_short(DXVA_Slice_HEVC_Short *slice,
229                              unsigned position, unsigned size)
230 {
231     memset(slice, 0, sizeof(*slice));
232     slice->BSNALunitDataLocation = position;
233     slice->SliceBytesInBuffer    = size;
234     slice->wBadSliceChopping     = 0;
235 }
236
237 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
238                                              DECODER_BUFFER_DESC *bs,
239                                              DECODER_BUFFER_DESC *sc)
240 {
241     const HEVCContext *h = avctx->priv_data;
242     AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
243     const HEVCFrame *current_picture = h->ref;
244     struct hevc_dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
245     DXVA_Slice_HEVC_Short *slice = NULL;
246     void     *dxva_data_ptr;
247     uint8_t  *dxva_data, *current, *end;
248     unsigned dxva_size;
249     void     *slice_data;
250     unsigned slice_size;
251     unsigned padding;
252     unsigned i;
253     unsigned type;
254
255     /* Create an annex B bitstream buffer with only slice NAL and finalize slice */
256 #if CONFIG_D3D11VA
257     if (ff_dxva2_is_d3d11(avctx)) {
258         type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
259         if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
260                                                        D3D11VA_CONTEXT(ctx)->decoder,
261                                                        type,
262                                                        &dxva_size, &dxva_data_ptr)))
263             return -1;
264     }
265 #endif
266 #if CONFIG_DXVA2
267     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
268         type = DXVA2_BitStreamDateBufferType;
269         if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
270                                                   type,
271                                                   &dxva_data_ptr, &dxva_size)))
272             return -1;
273     }
274 #endif
275
276     dxva_data = dxva_data_ptr;
277     current = dxva_data;
278     end = dxva_data + dxva_size;
279
280     for (i = 0; i < ctx_pic->slice_count; i++) {
281         static const uint8_t start_code[] = { 0, 0, 1 };
282         static const unsigned start_code_size = sizeof(start_code);
283         unsigned position, size;
284
285         slice = &ctx_pic->slice_short[i];
286
287         position = slice->BSNALunitDataLocation;
288         size     = slice->SliceBytesInBuffer;
289         if (start_code_size + size > end - current) {
290             av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
291             break;
292         }
293
294         slice->BSNALunitDataLocation = current - dxva_data;
295         slice->SliceBytesInBuffer    = start_code_size + size;
296
297         memcpy(current, start_code, start_code_size);
298         current += start_code_size;
299
300         memcpy(current, &ctx_pic->bitstream[position], size);
301         current += size;
302     }
303     padding = FFMIN(128 - ((current - dxva_data) & 127), end - current);
304     if (slice && padding > 0) {
305         memset(current, 0, padding);
306         current += padding;
307
308         slice->SliceBytesInBuffer += padding;
309     }
310 #if CONFIG_D3D11VA
311     if (ff_dxva2_is_d3d11(avctx))
312         if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
313             return -1;
314 #endif
315 #if CONFIG_DXVA2
316     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
317         if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
318             return -1;
319 #endif
320     if (i < ctx_pic->slice_count)
321         return -1;
322
323 #if CONFIG_D3D11VA
324     if (ff_dxva2_is_d3d11(avctx)) {
325         D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
326         memset(dsc11, 0, sizeof(*dsc11));
327         dsc11->BufferType           = type;
328         dsc11->DataSize             = current - dxva_data;
329         dsc11->NumMBsInBuffer       = 0;
330
331         type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
332     }
333 #endif
334 #if CONFIG_DXVA2
335     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
336         DXVA2_DecodeBufferDesc *dsc2 = bs;
337         memset(dsc2, 0, sizeof(*dsc2));
338         dsc2->CompressedBufferType = type;
339         dsc2->DataSize             = current - dxva_data;
340         dsc2->NumMBsInBuffer       = 0;
341
342         type = DXVA2_SliceControlBufferType;
343     }
344 #endif
345
346     slice_data = ctx_pic->slice_short;
347     slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_short);
348
349     av_assert0(((current - dxva_data) & 127) == 0);
350     return ff_dxva2_commit_buffer(avctx, ctx, sc,
351                                   type,
352                                   slice_data, slice_size, 0);
353 }
354
355
356 static int dxva2_hevc_start_frame(AVCodecContext *avctx,
357                                   av_unused const uint8_t *buffer,
358                                   av_unused uint32_t size)
359 {
360     const HEVCContext *h = avctx->priv_data;
361     AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
362     struct hevc_dxva2_picture_context *ctx_pic = h->ref->hwaccel_picture_private;
363
364     if (!DXVA_CONTEXT_VALID(avctx, ctx))
365         return -1;
366     av_assert0(ctx_pic);
367
368     /* Fill up DXVA_PicParams_HEVC */
369     fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp);
370
371     /* Fill up DXVA_Qmatrix_HEVC */
372     fill_scaling_lists(ctx, h, &ctx_pic->qm);
373
374     ctx_pic->slice_count    = 0;
375     ctx_pic->bitstream_size = 0;
376     ctx_pic->bitstream      = NULL;
377     return 0;
378 }
379
380 static int dxva2_hevc_decode_slice(AVCodecContext *avctx,
381                                    const uint8_t *buffer,
382                                    uint32_t size)
383 {
384     const HEVCContext *h = avctx->priv_data;
385     const HEVCFrame *current_picture = h->ref;
386     struct hevc_dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
387     unsigned position;
388
389     if (ctx_pic->slice_count >= MAX_SLICES)
390         return -1;
391
392     if (!ctx_pic->bitstream)
393         ctx_pic->bitstream = buffer;
394     ctx_pic->bitstream_size += size;
395
396     position = buffer - ctx_pic->bitstream;
397     fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count], position, size);
398     ctx_pic->slice_count++;
399
400     return 0;
401 }
402
403 static int dxva2_hevc_end_frame(AVCodecContext *avctx)
404 {
405     HEVCContext *h = avctx->priv_data;
406     struct hevc_dxva2_picture_context *ctx_pic = h->ref->hwaccel_picture_private;
407     int scale = ctx_pic->pp.dwCodingParamToolFlags & 1;
408     int ret;
409
410     if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
411         return -1;
412
413     ret = ff_dxva2_common_end_frame(avctx, h->ref->frame,
414                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
415                                     scale ? &ctx_pic->qm : NULL, scale ? sizeof(ctx_pic->qm) : 0,
416                                     commit_bitstream_and_slice_buffer);
417     return ret;
418 }
419
420 #if CONFIG_HEVC_DXVA2_HWACCEL
421 const AVHWAccel ff_hevc_dxva2_hwaccel = {
422     .name           = "hevc_dxva2",
423     .type           = AVMEDIA_TYPE_VIDEO,
424     .id             = AV_CODEC_ID_HEVC,
425     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
426     .init           = ff_dxva2_decode_init,
427     .uninit         = ff_dxva2_decode_uninit,
428     .start_frame    = dxva2_hevc_start_frame,
429     .decode_slice   = dxva2_hevc_decode_slice,
430     .end_frame      = dxva2_hevc_end_frame,
431     .frame_params   = ff_dxva2_common_frame_params,
432     .frame_priv_data_size = sizeof(struct hevc_dxva2_picture_context),
433     .priv_data_size = sizeof(FFDXVASharedContext),
434 };
435 #endif
436
437 #if CONFIG_HEVC_D3D11VA_HWACCEL
438 const AVHWAccel ff_hevc_d3d11va_hwaccel = {
439     .name           = "hevc_d3d11va",
440     .type           = AVMEDIA_TYPE_VIDEO,
441     .id             = AV_CODEC_ID_HEVC,
442     .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
443     .init           = ff_dxva2_decode_init,
444     .uninit         = ff_dxva2_decode_uninit,
445     .start_frame    = dxva2_hevc_start_frame,
446     .decode_slice   = dxva2_hevc_decode_slice,
447     .end_frame      = dxva2_hevc_end_frame,
448     .frame_params   = ff_dxva2_common_frame_params,
449     .frame_priv_data_size = sizeof(struct hevc_dxva2_picture_context),
450     .priv_data_size = sizeof(FFDXVASharedContext),
451 };
452 #endif
453
454 #if CONFIG_HEVC_D3D11VA2_HWACCEL
455 const AVHWAccel ff_hevc_d3d11va2_hwaccel = {
456     .name           = "hevc_d3d11va2",
457     .type           = AVMEDIA_TYPE_VIDEO,
458     .id             = AV_CODEC_ID_HEVC,
459     .pix_fmt        = AV_PIX_FMT_D3D11,
460     .init           = ff_dxva2_decode_init,
461     .uninit         = ff_dxva2_decode_uninit,
462     .start_frame    = dxva2_hevc_start_frame,
463     .decode_slice   = dxva2_hevc_decode_slice,
464     .end_frame      = dxva2_hevc_end_frame,
465     .frame_params   = ff_dxva2_common_frame_params,
466     .frame_priv_data_size = sizeof(struct hevc_dxva2_picture_context),
467     .priv_data_size = sizeof(FFDXVASharedContext),
468 };
469 #endif