1 /*
2 * HEVC HW decode acceleration through VA API
3 *
4 * Copyright (C) 2015 Timo Rothenpieler <timo@rothenpieler.org>
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 <va/va.h>
24 #include <va/va_dec_hevc.h>
25
32
34 #if VA_CHECK_VERSION(1, 2, 0)
35 VAPictureParameterBufferHEVCExtension
pic_param;
37 #else
40 #endif
43
46
48 {
49 va_pic->picture_id = VA_INVALID_ID;
50 va_pic->flags = VA_PICTURE_HEVC_INVALID;
51 va_pic->pic_order_cnt = 0;
52 }
53
55 {
57 va_pic->pic_order_cnt = pic->
poc;
58 va_pic->flags = rps_type;
59
61 va_pic->flags |= VA_PICTURE_HEVC_LONG_TERM_REFERENCE;
62
64 va_pic->flags |= VA_PICTURE_HEVC_FIELD_PIC;
65
67 va_pic->flags |= VA_PICTURE_HEVC_BOTTOM_FIELD;
68 }
69 }
70
72 {
75
78 return VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE;
79 }
80
83 return VA_PICTURE_HEVC_RPS_ST_CURR_AFTER;
84 }
85
88 return VA_PICTURE_HEVC_RPS_LT_CURR;
89 }
90
91 return 0;
92 }
93
95 {
98
101
105 j++;
106 }
107
109
113 }
114 }
115 }
116
120 {
125
127 int pic_param_size, err,
i;
128
129 VAPictureParameterBufferHEVC *pic_param = (VAPictureParameterBufferHEVC *)&pic->
pic_param;
130
132
133 *pic_param = (VAPictureParameterBufferHEVC) {
134 .pic_width_in_luma_samples =
sps->width,
135 .pic_height_in_luma_samples =
sps->height,
136 .log2_min_luma_coding_block_size_minus3 =
sps->log2_min_cb_size - 3,
137 .sps_max_dec_pic_buffering_minus1 =
sps->temporal_layer[
sps->max_sub_layers - 1].max_dec_pic_buffering - 1,
138 .log2_diff_max_min_luma_coding_block_size =
sps->log2_diff_max_min_coding_block_size,
139 .log2_min_transform_block_size_minus2 =
sps->log2_min_tb_size - 2,
140 .log2_diff_max_min_transform_block_size =
sps->log2_max_trafo_size -
sps->log2_min_tb_size,
141 .max_transform_hierarchy_depth_inter =
sps->max_transform_hierarchy_depth_inter,
142 .max_transform_hierarchy_depth_intra =
sps->max_transform_hierarchy_depth_intra,
143 .num_short_term_ref_pic_sets =
sps->nb_st_rps,
144 .num_long_term_ref_pic_sps =
sps->num_long_term_ref_pics_sps,
145 .num_ref_idx_l0_default_active_minus1 =
pps->num_ref_idx_l0_default_active - 1,
146 .num_ref_idx_l1_default_active_minus1 =
pps->num_ref_idx_l1_default_active - 1,
147 .init_qp_minus26 =
pps->pic_init_qp_minus26,
148 .pps_cb_qp_offset =
pps->cb_qp_offset,
149 .pps_cr_qp_offset =
pps->cr_qp_offset,
150 .pcm_sample_bit_depth_luma_minus1 =
sps->pcm.bit_depth - 1,
151 .pcm_sample_bit_depth_chroma_minus1 =
sps->pcm.bit_depth_chroma - 1,
152 .log2_min_pcm_luma_coding_block_size_minus3 =
sps->pcm.log2_min_pcm_cb_size - 3,
153 .log2_diff_max_min_pcm_luma_coding_block_size =
sps->pcm.log2_max_pcm_cb_size -
sps->pcm.log2_min_pcm_cb_size,
154 .diff_cu_qp_delta_depth =
pps->diff_cu_qp_delta_depth,
155 .pps_beta_offset_div2 =
pps->beta_offset / 2,
156 .pps_tc_offset_div2 =
pps->tc_offset / 2,
157 .log2_parallel_merge_level_minus2 =
pps->log2_parallel_merge_level - 2,
158 .bit_depth_luma_minus8 =
sps->bit_depth - 8,
159 .bit_depth_chroma_minus8 =
sps->bit_depth - 8,
160 .log2_max_pic_order_cnt_lsb_minus4 =
sps->log2_max_poc_lsb - 4,
161 .num_extra_slice_header_bits =
pps->num_extra_slice_header_bits,
162 .pic_fields.bits = {
163 .chroma_format_idc =
sps->chroma_format_idc,
164 .tiles_enabled_flag =
pps->tiles_enabled_flag,
165 .separate_colour_plane_flag =
sps->separate_colour_plane_flag,
166 .pcm_enabled_flag =
sps->pcm_enabled_flag,
167 .scaling_list_enabled_flag =
sps->scaling_list_enable_flag,
168 .transform_skip_enabled_flag =
pps->transform_skip_enabled_flag,
169 .amp_enabled_flag =
sps->amp_enabled_flag,
170 .strong_intra_smoothing_enabled_flag =
sps->sps_strong_intra_smoothing_enable_flag,
171 .sign_data_hiding_enabled_flag =
pps->sign_data_hiding_flag,
172 .constrained_intra_pred_flag =
pps->constrained_intra_pred_flag,
173 .cu_qp_delta_enabled_flag =
pps->cu_qp_delta_enabled_flag,
174 .weighted_pred_flag =
pps->weighted_pred_flag,
175 .weighted_bipred_flag =
pps->weighted_bipred_flag,
176 .transquant_bypass_enabled_flag =
pps->transquant_bypass_enable_flag,
177 .entropy_coding_sync_enabled_flag =
pps->entropy_coding_sync_enabled_flag,
178 .pps_loop_filter_across_slices_enabled_flag =
pps->seq_loop_filter_across_slices_enabled_flag,
179 .loop_filter_across_tiles_enabled_flag =
pps->loop_filter_across_tiles_enabled_flag,
180 .pcm_loop_filter_disabled_flag =
sps->pcm.loop_filter_disable_flag,
181 },
182 .slice_parsing_fields.bits = {
183 .lists_modification_present_flag =
pps->lists_modification_present_flag,
184 .long_term_ref_pics_present_flag =
sps->long_term_ref_pics_present_flag,
185 .sps_temporal_mvp_enabled_flag =
sps->sps_temporal_mvp_enabled_flag,
186 .cabac_init_present_flag =
pps->cabac_init_present_flag,
187 .output_flag_present_flag =
pps->output_flag_present_flag,
188 .dependent_slice_segments_enabled_flag =
pps->dependent_slice_segments_enabled_flag,
189 .pps_slice_chroma_qp_offsets_present_flag =
pps->pic_slice_level_chroma_qp_offsets_present_flag,
190 .sample_adaptive_offset_enabled_flag =
sps->sao_enabled,
191 .deblocking_filter_override_enabled_flag =
pps->deblocking_filter_override_enabled_flag,
192 .pps_disable_deblocking_filter_flag =
pps->disable_dbf,
193 .slice_segment_header_extension_present_flag =
pps->slice_header_extension_present_flag,
197 },
198 };
199
202
203 if (
pps->tiles_enabled_flag) {
204 pic_param->num_tile_columns_minus1 =
pps->num_tile_columns - 1;
205 pic_param->num_tile_rows_minus1 =
pps->num_tile_rows - 1;
206
207 for (
i = 0;
i <
pps->num_tile_columns;
i++)
208 pic_param->column_width_minus1[
i] =
pps->column_width[
i] - 1;
209
211 pic_param->row_height_minus1[
i] =
pps->row_height[
i] - 1;
212 }
213
214 if (
h->sh.short_term_ref_pic_set_sps_flag == 0 &&
h->sh.short_term_rps) {
215 pic_param->st_rps_bits =
h->sh.short_term_ref_pic_set_size;
216 } else {
217 pic_param->st_rps_bits = 0;
218 }
219
220 #if VA_CHECK_VERSION(1, 2, 0)
222 pic->
pic_param.rext = (VAPictureParameterBufferHEVCRext) {
223 .range_extension_pic_fields.bits = {
224 .transform_skip_rotation_enabled_flag =
sps->transform_skip_rotation_enabled_flag,
225 .transform_skip_context_enabled_flag =
sps->transform_skip_context_enabled_flag,
226 .implicit_rdpcm_enabled_flag =
sps->implicit_rdpcm_enabled_flag,
227 .explicit_rdpcm_enabled_flag =
sps->explicit_rdpcm_enabled_flag,
228 .extended_precision_processing_flag =
sps->extended_precision_processing_flag,
229 .intra_smoothing_disabled_flag =
sps->intra_smoothing_disabled_flag,
230 .high_precision_offsets_enabled_flag =
sps->high_precision_offsets_enabled_flag,
231 .persistent_rice_adaptation_enabled_flag =
sps->persistent_rice_adaptation_enabled_flag,
232 .cabac_bypass_alignment_enabled_flag =
sps->cabac_bypass_alignment_enabled_flag,
233 .cross_component_prediction_enabled_flag =
pps->cross_component_prediction_enabled_flag,
234 .chroma_qp_offset_list_enabled_flag =
pps->chroma_qp_offset_list_enabled_flag,
235 },
236 .diff_cu_chroma_qp_offset_depth =
pps->diff_cu_chroma_qp_offset_depth,
237 .chroma_qp_offset_list_len_minus1 =
pps->chroma_qp_offset_list_len_minus1,
238 .log2_sao_offset_scale_luma =
pps->log2_sao_offset_scale_luma,
239 .log2_sao_offset_scale_chroma =
pps->log2_sao_offset_scale_chroma,
240 .log2_max_transform_skip_block_size_minus2 =
pps->log2_max_transform_skip_block_size - 2,
241 };
242
243 for (
i = 0;
i < 6;
i++)
244 pic->
pic_param.rext.cb_qp_offset_list[
i] =
pps->cb_qp_offset_list[
i];
246 pic->
pic_param.rext.cr_qp_offset_list[
i] =
pps->cr_qp_offset_list[
i];
247 }
248 #endif
250 sizeof(pic->
pic_param) :
sizeof(VAPictureParameterBufferHEVC);
251
253 VAPictureParameterBufferType,
255 if (err < 0)
257
258 if (
pps->scaling_list_data_present_flag)
260 else if (
sps->scaling_list_enable_flag)
262
264 VAIQMatrixBufferHEVC iq_matrix;
265 int j;
266
267 for (
i = 0;
i < 6;
i++) {
268 for (j = 0; j < 16; j++)
270 for (j = 0; j < 64; j++) {
275 }
279 }
280
282 VAIQMatrixBufferType,
283 &iq_matrix, sizeof(iq_matrix));
284 if (err < 0)
286 }
287
288 return 0;
289
292 return err;
293 }
294
296 {
299 VASliceParameterBufferHEVC *last_slice_param = (VASliceParameterBufferHEVC *)&pic->
last_slice_param;
301
304
306 last_slice_param->LongSliceFlags.fields.LastSliceOfPic = 1;
312 }
313
314
318
319 return 0;
323 }
324
328 VASliceParameterBufferHEVC *slice_param)
329 {
331 #if VA_CHECK_VERSION(1, 2, 0)
333 #else
334 int is_rext = 0;
337 "or above, which can support REXT related setting correctly.\n");
338 #endif
339
340 memset(slice_param->delta_luma_weight_l0, 0, sizeof(slice_param->delta_luma_weight_l0));
341 memset(slice_param->delta_luma_weight_l1, 0, sizeof(slice_param->delta_luma_weight_l1));
342 memset(slice_param->luma_offset_l0, 0, sizeof(slice_param->luma_offset_l0));
343 memset(slice_param->luma_offset_l1, 0, sizeof(slice_param->luma_offset_l1));
344 memset(slice_param->delta_chroma_weight_l0, 0, sizeof(slice_param->delta_chroma_weight_l0));
345 memset(slice_param->delta_chroma_weight_l1, 0, sizeof(slice_param->delta_chroma_weight_l1));
346 memset(slice_param->ChromaOffsetL0, 0, sizeof(slice_param->ChromaOffsetL0));
347 memset(slice_param->ChromaOffsetL1, 0, sizeof(slice_param->ChromaOffsetL1));
348
349 slice_param->delta_chroma_log2_weight_denom = 0;
350 slice_param->luma_log2_weight_denom = 0;
351
355 return;
356
358
359 if (
h->ps.sps->chroma_format_idc) {
361 }
362
367 if (!is_rext) {
371 }
372 }
373
379 if (!is_rext) {
383 }
384 }
385 }
386 }
387
389 {
391 VAPictureParameterBufferHEVC *pp = (VAPictureParameterBufferHEVC *)&pic->
pic_param;
393
395 return 0xff;
396
398 VASurfaceID pid = pp->ReferenceFrames[
i].picture_id;
399 int poc = pp->ReferenceFrames[
i].pic_order_cnt;
402 }
403
404 return 0xff;
405 }
406
410 {
414 VASliceParameterBufferHEVC *last_slice_param = (VASliceParameterBufferHEVC *)&pic->
last_slice_param;
415
418
421
422 int err,
i, list_idx;
423
430 if (err) {
432 return err;
433 }
434 }
435
436 *last_slice_param = (VASliceParameterBufferHEVC) {
437 .slice_data_size =
size,
438 .slice_data_offset = 0,
439 .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
440 /* Add 1 to the bits count here to account for the byte_alignment bit, which
441 * always is at least one bit and not accounted for otherwise. */
442 .slice_data_byte_offset = (
get_bits_count(&
h->HEVClc->gb) + 1 + 7) / 8,
448 .slice_tc_offset_div2 = sh->
tc_offset / 2,
453
454 .LongSliceFlags.fields = {
466 },
467 };
468
469 memset(last_slice_param->RefPicList, 0xFF, sizeof(last_slice_param->RefPicList));
470
471 for (list_idx = 0; list_idx < nb_list; list_idx++) {
473
476 }
477
479
480 #if VA_CHECK_VERSION(1, 2, 0)
482 pic->last_slice_param.rext = (VASliceParameterBufferHEVCRext) {
483 .slice_ext_flags.bits = {
484 .cu_chroma_qp_offset_enabled_flag = sh->cu_chroma_qp_offset_enabled_flag,
485 },
486 };
487 for (
i = 0;
i < 15 &&
i < sh->nb_refs[
L0];
i++) {
488 pic->last_slice_param.rext.luma_offset_l0[
i] = sh->luma_offset_l0[
i];
489 pic->last_slice_param.rext.ChromaOffsetL0[
i][0] = sh->chroma_offset_l0[
i][0];
490 pic->last_slice_param.rext.ChromaOffsetL0[
i][1] = sh->chroma_offset_l0[
i][1];
491 }
492
493 for (
i = 0;
i < 15 &&
i < sh->nb_refs[
L0];
i++) {
494 pic->last_slice_param.rext.luma_offset_l0[
i] = sh->luma_offset_l0[
i];
495 pic->last_slice_param.rext.ChromaOffsetL0[
i][0] = sh->chroma_offset_l0[
i][0];
496 pic->last_slice_param.rext.ChromaOffsetL0[
i][1] = sh->chroma_offset_l0[
i][1];
497 }
498
500 for (
i = 0;
i < 15 &&
i < sh->nb_refs[
L1];
i++) {
501 pic->last_slice_param.rext.luma_offset_l1[
i] = sh->luma_offset_l1[
i];
502 pic->last_slice_param.rext.ChromaOffsetL1[
i][0] = sh->chroma_offset_l1[
i][0];
503 pic->last_slice_param.rext.ChromaOffsetL1[
i][1] = sh->chroma_offset_l1[
i][1];
504 }
505 }
506 }
507 #endif
508
509 pic->last_buffer =
buffer;
510 pic->last_size =
size;
511
512 return 0;
513 }
514
516 {
520
523
524 #define copy_field(name) h265_raw_ptl->general_ ## name = general_ptl->name
541 #undef copy_field
542
543 return 0;
544 }
545
546 /*
547 * Find exact va_profile for HEVC Range Extension
548 */
550 {
557
558 /* convert PTLCommon to H265RawProfileTierLevel */
560
564 goto end;
565 } else {
567 }
568
569 #if VA_CHECK_VERSION(1, 2, 0)
570 if (!strcmp(
profile->name,
"Main 12") ||
571 !strcmp(
profile->name,
"Main 12 Intra"))
572 return VAProfileHEVCMain12;
573 else if (!strcmp(
profile->name,
"Main 4:2:2 10") ||
574 !strcmp(
profile->name,
"Main 4:2:2 10 Intra"))
575 return VAProfileHEVCMain422_10;
576 else if (!strcmp(
profile->name,
"Main 4:2:2 12") ||
577 !strcmp(
profile->name,
"Main 4:2:2 12 Intra"))
578 return VAProfileHEVCMain422_12;
579 else if (!strcmp(
profile->name,
"Main 4:4:4") ||
580 !strcmp(
profile->name,
"Main 4:4:4 Intra"))
581 return VAProfileHEVCMain444;
582 else if (!strcmp(
profile->name,
"Main 4:4:4 10") ||
583 !strcmp(
profile->name,
"Main 4:4:4 10 Intra"))
584 return VAProfileHEVCMain444_10;
585 else if (!strcmp(
profile->name,
"Main 4:4:4 12") ||
586 !strcmp(
profile->name,
"Main 4:4:4 12 Intra"))
587 return VAProfileHEVCMain444_12;
588 #else
590 "not supported with this VA version.\n",
profile->name);
591 #endif
592
593 end:
595 // Default to selecting Main profile if profile mismatch is allowed
596 return VAProfileHEVCMain;
597 } else
598 return VAProfileNone;
599 }
600
602 .
name =
"hevc_vaapi",
615 };