2 * This file is part of FFmpeg.
4 * FFmpeg is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * FFmpeg is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with FFmpeg; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include <va/va_enc_hevc.h>
24 #include "libavutil/avassert.h"
25 #include "libavutil/common.h"
26 #include "libavutil/opt.h"
27 #include "libavutil/mastering_display_metadata.h"
36 #include "vaapi_encode.h"
39 SEI_MASTERING_DISPLAY = 0x08,
40 SEI_CONTENT_LIGHT_LEVEL = 0x10,
43 typedef struct VAAPIEncodeH265Context {
44 VAAPIEncodeContext common;
54 unsigned int ctu_width;
55 unsigned int ctu_height;
62 int64_t last_idr_frame;
75 H265RawSlice raw_slice;
77 H265RawSEIMasteringDisplayColourVolume sei_mastering_display;
78 H265RawSEIContentLightLevelInfo sei_content_light_level;
80 CodedBitstreamContext *cbc;
81 CodedBitstreamFragment current_access_unit;
84 } VAAPIEncodeH265Context;
87 static int vaapi_encode_h265_write_access_unit(AVCodecContext *avctx,
88 char *data, size_t *data_len,
89 CodedBitstreamFragment *au)
91 VAAPIEncodeH265Context *priv = avctx->priv_data;
94 err = ff_cbs_write_fragment_data(priv->cbc, au);
96 av_log(avctx, AV_LOG_ERROR, "Failed to write packed header.\n");
100 if (*data_len < 8 * au->data_size - au->data_bit_padding) {
101 av_log(avctx, AV_LOG_ERROR, "Access unit too large: "
102 "%zu < %zu.\n", *data_len,
103 8 * au->data_size - au->data_bit_padding);
104 return AVERROR(ENOSPC);
107 memcpy(data, au->data, au->data_size);
108 *data_len = 8 * au->data_size - au->data_bit_padding;
113 static int vaapi_encode_h265_add_nal(AVCodecContext *avctx,
114 CodedBitstreamFragment *au,
117 VAAPIEncodeH265Context *priv = avctx->priv_data;
118 H265RawNALUnitHeader *header = nal_unit;
121 err = ff_cbs_insert_unit_content(priv->cbc, au, -1,
122 header->nal_unit_type, nal_unit, NULL);
124 av_log(avctx, AV_LOG_ERROR, "Failed to add NAL unit: "
125 "type = %d.\n", header->nal_unit_type);
132 static int vaapi_encode_h265_write_sequence_header(AVCodecContext *avctx,
133 char *data, size_t *data_len)
135 VAAPIEncodeH265Context *priv = avctx->priv_data;
136 CodedBitstreamFragment *au = &priv->current_access_unit;
139 if (priv->aud_needed) {
140 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_aud);
143 priv->aud_needed = 0;
146 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_vps);
150 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_sps);
154 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_pps);
158 err = vaapi_encode_h265_write_access_unit(avctx, data, data_len, au);
160 ff_cbs_fragment_uninit(priv->cbc, au);
164 static int vaapi_encode_h265_write_slice_header(AVCodecContext *avctx,
165 VAAPIEncodePicture *pic,
166 VAAPIEncodeSlice *slice,
167 char *data, size_t *data_len)
169 VAAPIEncodeH265Context *priv = avctx->priv_data;
170 CodedBitstreamFragment *au = &priv->current_access_unit;
173 if (priv->aud_needed) {
174 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_aud);
177 priv->aud_needed = 0;
180 err = vaapi_encode_h265_add_nal(avctx, au, &priv->raw_slice);
184 err = vaapi_encode_h265_write_access_unit(avctx, data, data_len, au);
186 ff_cbs_fragment_uninit(priv->cbc, au);
190 static int vaapi_encode_h265_write_extra_header(AVCodecContext *avctx,
191 VAAPIEncodePicture *pic,
192 int index, int *type,
193 char *data, size_t *data_len)
195 VAAPIEncodeH265Context *priv = avctx->priv_data;
196 CodedBitstreamFragment *au = &priv->current_access_unit;
199 if (priv->sei_needed) {
200 H265RawSEI *sei = &priv->raw_sei;
202 if (priv->aud_needed) {
203 err = vaapi_encode_h265_add_nal(avctx, au, &priv->aud);
206 priv->aud_needed = 0;
209 *sei = (H265RawSEI) {
211 .nal_unit_type = HEVC_NAL_SEI_PREFIX,
213 .nuh_temporal_id_plus1 = 1,
219 if (priv->sei_needed & SEI_MASTERING_DISPLAY) {
220 sei->payload[i].payload_type = HEVC_SEI_TYPE_MASTERING_DISPLAY_INFO;
221 sei->payload[i].payload.mastering_display = priv->sei_mastering_display;
225 if (priv->sei_needed & SEI_CONTENT_LIGHT_LEVEL) {
226 sei->payload[i].payload_type = HEVC_SEI_TYPE_CONTENT_LIGHT_LEVEL_INFO;
227 sei->payload[i].payload.content_light_level = priv->sei_content_light_level;
231 sei->payload_count = i;
232 av_assert0(sei->payload_count > 0);
234 err = vaapi_encode_h265_add_nal(avctx, au, sei);
237 priv->sei_needed = 0;
239 err = vaapi_encode_h265_write_access_unit(avctx, data, data_len, au);
243 ff_cbs_fragment_uninit(priv->cbc, au);
245 *type = VAEncPackedHeaderRawData;
252 ff_cbs_fragment_uninit(priv->cbc, au);
256 static int vaapi_encode_h265_init_sequence_params(AVCodecContext *avctx)
258 VAAPIEncodeContext *ctx = avctx->priv_data;
259 VAAPIEncodeH265Context *priv = avctx->priv_data;
260 H265RawVPS *vps = &priv->raw_vps;
261 H265RawSPS *sps = &priv->raw_sps;
262 H265RawPPS *pps = &priv->raw_pps;
263 H265RawVUI *vui = &sps->vui;
264 VAEncSequenceParameterBufferHEVC *vseq = ctx->codec_sequence_params;
265 VAEncPictureParameterBufferHEVC *vpic = ctx->codec_picture_params;
268 memset(&priv->current_access_unit, 0,
269 sizeof(priv->current_access_unit));
271 memset(vps, 0, sizeof(*vps));
272 memset(sps, 0, sizeof(*sps));
273 memset(pps, 0, sizeof(*pps));
278 vps->nal_unit_header = (H265RawNALUnitHeader) {
279 .nal_unit_type = HEVC_NAL_VPS,
281 .nuh_temporal_id_plus1 = 1,
284 vps->vps_video_parameter_set_id = 0;
286 vps->vps_base_layer_internal_flag = 1;
287 vps->vps_base_layer_available_flag = 1;
288 vps->vps_max_layers_minus1 = 0;
289 vps->vps_max_sub_layers_minus1 = 0;
290 vps->vps_temporal_id_nesting_flag = 1;
292 vps->profile_tier_level = (H265RawProfileTierLevel) {
293 .general_profile_space = 0,
294 .general_profile_idc = avctx->profile,
295 .general_tier_flag = 0,
297 .general_progressive_source_flag = 1,
298 .general_interlaced_source_flag = 0,
299 .general_non_packed_constraint_flag = 1,
300 .general_frame_only_constraint_flag = 1,
302 .general_level_idc = avctx->level,
304 vps->profile_tier_level.general_profile_compatibility_flag[avctx->profile & 31] = 1;
306 vps->vps_sub_layer_ordering_info_present_flag = 0;
307 vps->vps_max_dec_pic_buffering_minus1[0] = (ctx->b_per_p > 0) + 1;
308 vps->vps_max_num_reorder_pics[0] = (ctx->b_per_p > 0);
309 vps->vps_max_latency_increase_plus1[0] = 0;
311 vps->vps_max_layer_id = 0;
312 vps->vps_num_layer_sets_minus1 = 0;
313 vps->layer_id_included_flag[0][0] = 1;
315 vps->vps_timing_info_present_flag = 1;
316 if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {
317 vps->vps_num_units_in_tick = avctx->framerate.den;
318 vps->vps_time_scale = avctx->framerate.num;
319 vps->vps_poc_proportional_to_timing_flag = 1;
320 vps->vps_num_ticks_poc_diff_one_minus1 = 0;
322 vps->vps_num_units_in_tick = avctx->time_base.num;
323 vps->vps_time_scale = avctx->time_base.den;
324 vps->vps_poc_proportional_to_timing_flag = 0;
326 vps->vps_num_hrd_parameters = 0;
331 sps->nal_unit_header = (H265RawNALUnitHeader) {
332 .nal_unit_type = HEVC_NAL_SPS,
334 .nuh_temporal_id_plus1 = 1,
337 sps->sps_video_parameter_set_id = vps->vps_video_parameter_set_id;
339 sps->sps_max_sub_layers_minus1 = vps->vps_max_sub_layers_minus1;
340 sps->sps_temporal_id_nesting_flag = vps->vps_temporal_id_nesting_flag;
342 sps->profile_tier_level = vps->profile_tier_level;
344 sps->sps_seq_parameter_set_id = 0;
346 sps->chroma_format_idc = 1; // YUV 4:2:0.
347 sps->separate_colour_plane_flag = 0;
349 sps->pic_width_in_luma_samples = ctx->surface_width;
350 sps->pic_height_in_luma_samples = ctx->surface_height;
352 if (avctx->width != ctx->surface_width ||
353 avctx->height != ctx->surface_height) {
354 sps->conformance_window_flag = 1;
355 sps->conf_win_left_offset = 0;
356 sps->conf_win_right_offset =
357 (ctx->surface_width - avctx->width) / 2;
358 sps->conf_win_top_offset = 0;
359 sps->conf_win_bottom_offset =
360 (ctx->surface_height - avctx->height) / 2;
362 sps->conformance_window_flag = 0;
365 sps->bit_depth_luma_minus8 =
366 avctx->profile == FF_PROFILE_HEVC_MAIN_10 ? 2 : 0;
367 sps->bit_depth_chroma_minus8 = sps->bit_depth_luma_minus8;
369 sps->log2_max_pic_order_cnt_lsb_minus4 = 8;
371 sps->sps_sub_layer_ordering_info_present_flag =
372 vps->vps_sub_layer_ordering_info_present_flag;
373 for (i = 0; i <= sps->sps_max_sub_layers_minus1; i++) {
374 sps->sps_max_dec_pic_buffering_minus1[i] =
375 vps->vps_max_dec_pic_buffering_minus1[i];
376 sps->sps_max_num_reorder_pics[i] =
377 vps->vps_max_num_reorder_pics[i];
378 sps->sps_max_latency_increase_plus1[i] =
379 vps->vps_max_latency_increase_plus1[i];
382 // These have to come from the capabilities of the encoder. We have no
383 // way to query them, so just hardcode parameters which work on the Intel
385 // CTB size from 8x8 to 32x32.
386 sps->log2_min_luma_coding_block_size_minus3 = 0;
387 sps->log2_diff_max_min_luma_coding_block_size = 2;
388 // Transform size from 4x4 to 32x32.
389 sps->log2_min_luma_transform_block_size_minus2 = 0;
390 sps->log2_diff_max_min_luma_transform_block_size = 3;
391 // Full transform hierarchy allowed (2-5).
392 sps->max_transform_hierarchy_depth_inter = 3;
393 sps->max_transform_hierarchy_depth_intra = 3;
395 sps->amp_enabled_flag = 1;
396 // SAO and temporal MVP do not work.
397 sps->sample_adaptive_offset_enabled_flag = 0;
398 sps->sps_temporal_mvp_enabled_flag = 0;
400 sps->pcm_enabled_flag = 0;
402 // STRPSs should ideally be here rather than defined individually in
403 // each slice, but the structure isn't completely fixed so for now
405 sps->num_short_term_ref_pic_sets = 0;
406 sps->long_term_ref_pics_present_flag = 0;
408 sps->vui_parameters_present_flag = 1;
410 if (avctx->sample_aspect_ratio.num != 0 &&
411 avctx->sample_aspect_ratio.den != 0) {
412 static const AVRational sar_idc[] = {
414 { 1, 1 }, { 12, 11 }, { 10, 11 }, { 16, 11 },
415 { 40, 33 }, { 24, 11 }, { 20, 11 }, { 32, 11 },
416 { 80, 33 }, { 18, 11 }, { 15, 11 }, { 64, 33 },
417 { 160, 99 }, { 4, 3 }, { 3, 2 }, { 2, 1 },
420 for (i = 0; i < FF_ARRAY_ELEMS(sar_idc); i++) {
421 if (avctx->sample_aspect_ratio.num == sar_idc[i].num &&
422 avctx->sample_aspect_ratio.den == sar_idc[i].den) {
423 vui->aspect_ratio_idc = i;
427 if (i >= FF_ARRAY_ELEMS(sar_idc)) {
428 vui->aspect_ratio_idc = 255;
429 vui->sar_width = avctx->sample_aspect_ratio.num;
430 vui->sar_height = avctx->sample_aspect_ratio.den;
432 vui->aspect_ratio_info_present_flag = 1;
435 if (avctx->color_range != AVCOL_RANGE_UNSPECIFIED ||
436 avctx->color_primaries != AVCOL_PRI_UNSPECIFIED ||
437 avctx->color_trc != AVCOL_TRC_UNSPECIFIED ||
438 avctx->colorspace != AVCOL_SPC_UNSPECIFIED) {
439 vui->video_signal_type_present_flag = 1;
440 vui->video_format = 5; // Unspecified.
441 vui->video_full_range_flag =
442 avctx->color_range == AVCOL_RANGE_JPEG;
444 if (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED ||
445 avctx->color_trc != AVCOL_TRC_UNSPECIFIED ||
446 avctx->colorspace != AVCOL_SPC_UNSPECIFIED) {
447 vui->colour_description_present_flag = 1;
448 vui->colour_primaries = avctx->color_primaries;
449 vui->transfer_characteristics = avctx->color_trc;
450 vui->matrix_coefficients = avctx->colorspace;
453 vui->video_format = 5;
454 vui->video_full_range_flag = 0;
455 vui->colour_primaries = avctx->color_primaries;
456 vui->transfer_characteristics = avctx->color_trc;
457 vui->matrix_coefficients = avctx->colorspace;
460 if (avctx->chroma_sample_location != AVCHROMA_LOC_UNSPECIFIED) {
461 vui->chroma_loc_info_present_flag = 1;
462 vui->chroma_sample_loc_type_top_field =
463 vui->chroma_sample_loc_type_bottom_field =
464 avctx->chroma_sample_location - 1;
467 vui->vui_timing_info_present_flag = 1;
468 vui->vui_num_units_in_tick = vps->vps_num_units_in_tick;
469 vui->vui_time_scale = vps->vps_time_scale;
470 vui->vui_poc_proportional_to_timing_flag = vps->vps_poc_proportional_to_timing_flag;
471 vui->vui_num_ticks_poc_diff_one_minus1 = vps->vps_num_ticks_poc_diff_one_minus1;
472 vui->vui_hrd_parameters_present_flag = 0;
474 vui->bitstream_restriction_flag = 1;
475 vui->motion_vectors_over_pic_boundaries_flag = 1;
476 vui->restricted_ref_pic_lists_flag = 1;
477 vui->max_bytes_per_pic_denom = 0;
478 vui->max_bits_per_min_cu_denom = 0;
479 vui->log2_max_mv_length_horizontal = 15;
480 vui->log2_max_mv_length_vertical = 15;
485 pps->nal_unit_header = (H265RawNALUnitHeader) {
486 .nal_unit_type = HEVC_NAL_PPS,
488 .nuh_temporal_id_plus1 = 1,
491 pps->pps_pic_parameter_set_id = 0;
492 pps->pps_seq_parameter_set_id = sps->sps_seq_parameter_set_id;
494 pps->num_ref_idx_l0_default_active_minus1 = 0;
495 pps->num_ref_idx_l1_default_active_minus1 = 0;
497 pps->init_qp_minus26 = priv->fixed_qp_idr - 26;
499 pps->cu_qp_delta_enabled_flag = (ctx->va_rc_mode != VA_RC_CQP);
500 pps->diff_cu_qp_delta_depth = 0;
502 pps->pps_loop_filter_across_slices_enabled_flag = 1;
505 // Fill VAAPI parameter buffers.
507 *vseq = (VAEncSequenceParameterBufferHEVC) {
508 .general_profile_idc = vps->profile_tier_level.general_profile_idc,
509 .general_level_idc = vps->profile_tier_level.general_level_idc,
510 .general_tier_flag = vps->profile_tier_level.general_tier_flag,
512 .intra_period = avctx->gop_size,
513 .intra_idr_period = avctx->gop_size,
514 .ip_period = ctx->b_per_p + 1,
515 .bits_per_second = avctx->bit_rate,
517 .pic_width_in_luma_samples = sps->pic_width_in_luma_samples,
518 .pic_height_in_luma_samples = sps->pic_height_in_luma_samples,
521 .chroma_format_idc = sps->chroma_format_idc,
522 .separate_colour_plane_flag = sps->separate_colour_plane_flag,
523 .bit_depth_luma_minus8 = sps->bit_depth_luma_minus8,
524 .bit_depth_chroma_minus8 = sps->bit_depth_chroma_minus8,
525 .scaling_list_enabled_flag = sps->scaling_list_enabled_flag,
526 .strong_intra_smoothing_enabled_flag =
527 sps->strong_intra_smoothing_enabled_flag,
528 .amp_enabled_flag = sps->amp_enabled_flag,
529 .sample_adaptive_offset_enabled_flag =
530 sps->sample_adaptive_offset_enabled_flag,
531 .pcm_enabled_flag = sps->pcm_enabled_flag,
532 .pcm_loop_filter_disabled_flag = sps->pcm_loop_filter_disabled_flag,
533 .sps_temporal_mvp_enabled_flag = sps->sps_temporal_mvp_enabled_flag,
536 .log2_min_luma_coding_block_size_minus3 =
537 sps->log2_min_luma_coding_block_size_minus3,
538 .log2_diff_max_min_luma_coding_block_size =
539 sps->log2_diff_max_min_luma_coding_block_size,
540 .log2_min_transform_block_size_minus2 =
541 sps->log2_min_luma_transform_block_size_minus2,
542 .log2_diff_max_min_transform_block_size =
543 sps->log2_diff_max_min_luma_transform_block_size,
544 .max_transform_hierarchy_depth_inter =
545 sps->max_transform_hierarchy_depth_inter,
546 .max_transform_hierarchy_depth_intra =
547 sps->max_transform_hierarchy_depth_intra,
549 .pcm_sample_bit_depth_luma_minus1 =
550 sps->pcm_sample_bit_depth_luma_minus1,
551 .pcm_sample_bit_depth_chroma_minus1 =
552 sps->pcm_sample_bit_depth_chroma_minus1,
553 .log2_min_pcm_luma_coding_block_size_minus3 =
554 sps->log2_min_pcm_luma_coding_block_size_minus3,
555 .log2_max_pcm_luma_coding_block_size_minus3 =
556 sps->log2_min_pcm_luma_coding_block_size_minus3 +
557 sps->log2_diff_max_min_pcm_luma_coding_block_size,
559 .vui_parameters_present_flag = 0,
562 *vpic = (VAEncPictureParameterBufferHEVC) {
563 .decoded_curr_pic = {
564 .picture_id = VA_INVALID_ID,
565 .flags = VA_PICTURE_HEVC_INVALID,
568 .coded_buf = VA_INVALID_ID,
570 .collocated_ref_pic_index = 0xff,
574 .pic_init_qp = pps->init_qp_minus26 + 26,
575 .diff_cu_qp_delta_depth = pps->diff_cu_qp_delta_depth,
576 .pps_cb_qp_offset = pps->pps_cb_qp_offset,
577 .pps_cr_qp_offset = pps->pps_cr_qp_offset,
579 .num_tile_columns_minus1 = pps->num_tile_columns_minus1,
580 .num_tile_rows_minus1 = pps->num_tile_rows_minus1,
582 .log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level_minus2,
583 .ctu_max_bitsize_allowed = 0,
585 .num_ref_idx_l0_default_active_minus1 =
586 pps->num_ref_idx_l0_default_active_minus1,
587 .num_ref_idx_l1_default_active_minus1 =
588 pps->num_ref_idx_l1_default_active_minus1,
590 .slice_pic_parameter_set_id = pps->pps_pic_parameter_set_id,
593 .sign_data_hiding_enabled_flag = pps->sign_data_hiding_enabled_flag,
594 .constrained_intra_pred_flag = pps->constrained_intra_pred_flag,
595 .transform_skip_enabled_flag = pps->transform_skip_enabled_flag,
596 .cu_qp_delta_enabled_flag = pps->cu_qp_delta_enabled_flag,
597 .weighted_pred_flag = pps->weighted_pred_flag,
598 .weighted_bipred_flag = pps->weighted_bipred_flag,
599 .transquant_bypass_enabled_flag = pps->transquant_bypass_enabled_flag,
600 .tiles_enabled_flag = pps->tiles_enabled_flag,
601 .entropy_coding_sync_enabled_flag = pps->entropy_coding_sync_enabled_flag,
602 .loop_filter_across_tiles_enabled_flag =
603 pps->loop_filter_across_tiles_enabled_flag,
604 .scaling_list_data_present_flag = (sps->sps_scaling_list_data_present_flag |
605 pps->pps_scaling_list_data_present_flag),
606 .screen_content_flag = 0,
607 .enable_gpu_weighted_prediction = 0,
608 .no_output_of_prior_pics_flag = 0,
615 static int vaapi_encode_h265_init_picture_params(AVCodecContext *avctx,
616 VAAPIEncodePicture *pic)
618 VAAPIEncodeH265Context *priv = avctx->priv_data;
619 VAEncPictureParameterBufferHEVC *vpic = pic->codec_picture_params;
622 if (pic->type == PICTURE_TYPE_IDR) {
623 av_assert0(pic->display_order == pic->encode_order);
625 priv->last_idr_frame = pic->display_order;
627 priv->slice_nal_unit = HEVC_NAL_IDR_W_RADL;
628 priv->slice_type = HEVC_SLICE_I;
631 av_assert0(pic->encode_order > priv->last_idr_frame);
633 if (pic->type == PICTURE_TYPE_I) {
634 priv->slice_nal_unit = HEVC_NAL_CRA_NUT;
635 priv->slice_type = HEVC_SLICE_I;
637 } else if (pic->type == PICTURE_TYPE_P) {
638 av_assert0(pic->refs[0]);
639 priv->slice_nal_unit = HEVC_NAL_TRAIL_R;
640 priv->slice_type = HEVC_SLICE_P;
643 av_assert0(pic->refs[0] && pic->refs[1]);
644 if (pic->refs[1]->type == PICTURE_TYPE_I)
645 priv->slice_nal_unit = HEVC_NAL_RASL_N;
647 priv->slice_nal_unit = HEVC_NAL_TRAIL_N;
648 priv->slice_type = HEVC_SLICE_B;
652 priv->pic_order_cnt = pic->display_order - priv->last_idr_frame;
655 priv->aud_needed = 1;
656 priv->raw_aud = (H265RawAUD) {
658 .nal_unit_type = HEVC_NAL_AUD,
660 .nuh_temporal_id_plus1 = 1,
662 .pic_type = priv->pic_type,
665 priv->aud_needed = 0;
668 priv->sei_needed = 0;
670 // Only look for the metadata on I/IDR frame on the output. We
671 // may force an IDR frame on the output where the medadata gets
672 // changed on the input frame.
673 if ((priv->sei & SEI_MASTERING_DISPLAY) &&
674 (pic->type == PICTURE_TYPE_I || pic->type == PICTURE_TYPE_IDR)) {
675 AVFrameSideData *sd =
676 av_frame_get_side_data(pic->input_image,
677 AV_FRAME_DATA_MASTERING_DISPLAY_METADATA);
680 AVMasteringDisplayMetadata *mdm =
681 (AVMasteringDisplayMetadata *)sd->data;
683 // SEI is needed when both the primaries and luminance are set
684 if (mdm->has_primaries && mdm->has_luminance) {
685 H265RawSEIMasteringDisplayColourVolume *mdcv =
686 &priv->sei_mastering_display;
687 const int mapping[3] = {1, 2, 0};
688 const int chroma_den = 50000;
689 const int luma_den = 10000;
691 for (i = 0; i < 3; i++) {
692 const int j = mapping[i];
693 mdcv->display_primaries_x[i] =
694 FFMIN(lrint(chroma_den *
695 av_q2d(mdm->display_primaries[j][0])),
697 mdcv->display_primaries_y[i] =
698 FFMIN(lrint(chroma_den *
699 av_q2d(mdm->display_primaries[j][1])),
703 mdcv->white_point_x =
704 FFMIN(lrint(chroma_den * av_q2d(mdm->white_point[0])),
706 mdcv->white_point_y =
707 FFMIN(lrint(chroma_den * av_q2d(mdm->white_point[1])),
710 mdcv->max_display_mastering_luminance =
711 lrint(luma_den * av_q2d(mdm->max_luminance));
712 mdcv->min_display_mastering_luminance =
713 FFMIN(lrint(luma_den * av_q2d(mdm->min_luminance)),
714 mdcv->max_display_mastering_luminance);
716 priv->sei_needed |= SEI_MASTERING_DISPLAY;
721 if ((priv->sei & SEI_CONTENT_LIGHT_LEVEL) &&
722 (pic->type == PICTURE_TYPE_I || pic->type == PICTURE_TYPE_IDR)) {
723 AVFrameSideData *sd =
724 av_frame_get_side_data(pic->input_image,
725 AV_FRAME_DATA_CONTENT_LIGHT_LEVEL);
728 AVContentLightMetadata *clm =
729 (AVContentLightMetadata *)sd->data;
730 H265RawSEIContentLightLevelInfo *clli =
731 &priv->sei_content_light_level;
733 clli->max_content_light_level = FFMIN(clm->MaxCLL, 65535);
734 clli->max_pic_average_light_level = FFMIN(clm->MaxFALL, 65535);
736 priv->sei_needed |= SEI_CONTENT_LIGHT_LEVEL;
740 vpic->decoded_curr_pic = (VAPictureHEVC) {
741 .picture_id = pic->recon_surface,
742 .pic_order_cnt = priv->pic_order_cnt,
746 for (i = 0; i < pic->nb_refs; i++) {
747 VAAPIEncodePicture *ref = pic->refs[i];
748 av_assert0(ref && ref->encode_order < pic->encode_order);
750 vpic->reference_frames[i] = (VAPictureHEVC) {
751 .picture_id = ref->recon_surface,
752 .pic_order_cnt = ref->display_order - priv->last_idr_frame,
753 .flags = (ref->display_order < pic->display_order ?
754 VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE : 0) |
755 (ref->display_order > pic->display_order ?
756 VA_PICTURE_HEVC_RPS_ST_CURR_AFTER : 0),
759 for (; i < FF_ARRAY_ELEMS(vpic->reference_frames); i++) {
760 vpic->reference_frames[i] = (VAPictureHEVC) {
761 .picture_id = VA_INVALID_ID,
762 .flags = VA_PICTURE_HEVC_INVALID,
766 vpic->coded_buf = pic->output_buffer;
768 vpic->nal_unit_type = priv->slice_nal_unit;
771 case PICTURE_TYPE_IDR:
772 vpic->pic_fields.bits.idr_pic_flag = 1;
773 vpic->pic_fields.bits.coding_type = 1;
774 vpic->pic_fields.bits.reference_pic_flag = 1;
777 vpic->pic_fields.bits.idr_pic_flag = 0;
778 vpic->pic_fields.bits.coding_type = 1;
779 vpic->pic_fields.bits.reference_pic_flag = 1;
782 vpic->pic_fields.bits.idr_pic_flag = 0;
783 vpic->pic_fields.bits.coding_type = 2;
784 vpic->pic_fields.bits.reference_pic_flag = 1;
787 vpic->pic_fields.bits.idr_pic_flag = 0;
788 vpic->pic_fields.bits.coding_type = 3;
789 vpic->pic_fields.bits.reference_pic_flag = 0;
792 av_assert0(0 && "invalid picture type");
800 static int vaapi_encode_h265_init_slice_params(AVCodecContext *avctx,
801 VAAPIEncodePicture *pic,
802 VAAPIEncodeSlice *slice)
804 VAAPIEncodeContext *ctx = avctx->priv_data;
805 VAAPIEncodeH265Context *priv = avctx->priv_data;
806 const H265RawSPS *sps = &priv->raw_sps;
807 const H265RawPPS *pps = &priv->raw_pps;
808 H265RawSliceHeader *sh = &priv->raw_slice.header;
809 VAEncPictureParameterBufferHEVC *vpic = pic->codec_picture_params;
810 VAEncSliceParameterBufferHEVC *vslice = slice->codec_slice_params;
813 sh->nal_unit_header = (H265RawNALUnitHeader) {
814 .nal_unit_type = priv->slice_nal_unit,
816 .nuh_temporal_id_plus1 = 1,
819 sh->slice_pic_parameter_set_id = pps->pps_pic_parameter_set_id;
821 // Currently we only support one slice per frame.
822 sh->first_slice_segment_in_pic_flag = 1;
823 sh->slice_segment_address = 0;
825 sh->slice_type = priv->slice_type;
827 sh->slice_pic_order_cnt_lsb = priv->pic_order_cnt &
828 (1 << (sps->log2_max_pic_order_cnt_lsb_minus4 + 4)) - 1;
830 if (pic->type != PICTURE_TYPE_IDR) {
831 H265RawSTRefPicSet *rps;
832 VAAPIEncodePicture *st;
835 sh->short_term_ref_pic_set_sps_flag = 0;
837 rps = &sh->short_term_ref_pic_set;
838 memset(rps, 0, sizeof(*rps));
840 for (st = ctx->pic_start; st; st = st->next) {
841 if (st->encode_order >= pic->encode_order) {
846 for (i = 0; i < pic->nb_refs; i++) {
847 if (pic->refs[i] == st)
851 // Usually each picture always uses all of the others in the
852 // DPB as references. The one case we have to treat here is
853 // a non-IDR IRAP picture, which may need to hold unused
854 // references across itself to be used for the decoding of
855 // following RASL pictures. This looks for such an RASL
856 // picture, and keeps the reference if there is one.
857 VAAPIEncodePicture *rp;
858 for (rp = ctx->pic_start; rp; rp = rp->next) {
859 if (rp->encode_order < pic->encode_order)
861 if (rp->type != PICTURE_TYPE_B)
863 if (rp->refs[0] == st && rp->refs[1] == pic)
869 // This only works for one instance of each (delta_poc_sN_minus1
870 // is relative to the previous frame in the list, not relative to
871 // the current frame directly).
872 if (st->display_order < pic->display_order) {
873 rps->delta_poc_s0_minus1[rps->num_negative_pics] =
874 pic->display_order - st->display_order - 1;
875 rps->used_by_curr_pic_s0_flag[rps->num_negative_pics] = used;
876 ++rps->num_negative_pics;
878 rps->delta_poc_s1_minus1[rps->num_positive_pics] =
879 st->display_order - pic->display_order - 1;
880 rps->used_by_curr_pic_s1_flag[rps->num_positive_pics] = used;
881 ++rps->num_positive_pics;
885 sh->num_long_term_sps = 0;
886 sh->num_long_term_pics = 0;
888 sh->slice_temporal_mvp_enabled_flag =
889 sps->sps_temporal_mvp_enabled_flag;
890 if (sh->slice_temporal_mvp_enabled_flag) {
891 sh->collocated_from_l0_flag = sh->slice_type == HEVC_SLICE_B;
892 sh->collocated_ref_idx = 0;
895 sh->num_ref_idx_active_override_flag = 0;
896 sh->num_ref_idx_l0_active_minus1 = pps->num_ref_idx_l0_default_active_minus1;
897 sh->num_ref_idx_l1_active_minus1 = pps->num_ref_idx_l1_default_active_minus1;
900 sh->slice_sao_luma_flag = sh->slice_sao_chroma_flag =
901 sps->sample_adaptive_offset_enabled_flag;
903 if (pic->type == PICTURE_TYPE_B)
904 sh->slice_qp_delta = priv->fixed_qp_b - (pps->init_qp_minus26 + 26);
905 else if (pic->type == PICTURE_TYPE_P)
906 sh->slice_qp_delta = priv->fixed_qp_p - (pps->init_qp_minus26 + 26);
908 sh->slice_qp_delta = priv->fixed_qp_idr - (pps->init_qp_minus26 + 26);
911 *vslice = (VAEncSliceParameterBufferHEVC) {
912 .slice_segment_address = sh->slice_segment_address,
913 .num_ctu_in_slice = priv->ctu_width * priv->ctu_height,
915 .slice_type = sh->slice_type,
916 .slice_pic_parameter_set_id = sh->slice_pic_parameter_set_id,
918 .num_ref_idx_l0_active_minus1 = sh->num_ref_idx_l0_active_minus1,
919 .num_ref_idx_l1_active_minus1 = sh->num_ref_idx_l1_active_minus1,
921 .luma_log2_weight_denom = sh->luma_log2_weight_denom,
922 .delta_chroma_log2_weight_denom = sh->delta_chroma_log2_weight_denom,
924 .max_num_merge_cand = 5 - sh->five_minus_max_num_merge_cand,
926 .slice_qp_delta = sh->slice_qp_delta,
927 .slice_cb_qp_offset = sh->slice_cb_qp_offset,
928 .slice_cr_qp_offset = sh->slice_cr_qp_offset,
930 .slice_beta_offset_div2 = sh->slice_beta_offset_div2,
931 .slice_tc_offset_div2 = sh->slice_tc_offset_div2,
933 .slice_fields.bits = {
934 .last_slice_of_pic_flag = 1,
935 .dependent_slice_segment_flag = sh->dependent_slice_segment_flag,
936 .colour_plane_id = sh->colour_plane_id,
937 .slice_temporal_mvp_enabled_flag =
938 sh->slice_temporal_mvp_enabled_flag,
939 .slice_sao_luma_flag = sh->slice_sao_luma_flag,
940 .slice_sao_chroma_flag = sh->slice_sao_chroma_flag,
941 .num_ref_idx_active_override_flag =
942 sh->num_ref_idx_active_override_flag,
943 .mvd_l1_zero_flag = sh->mvd_l1_zero_flag,
944 .cabac_init_flag = sh->cabac_init_flag,
945 .slice_deblocking_filter_disabled_flag =
946 sh->slice_deblocking_filter_disabled_flag,
947 .slice_loop_filter_across_slices_enabled_flag =
948 sh->slice_loop_filter_across_slices_enabled_flag,
949 .collocated_from_l0_flag = sh->collocated_from_l0_flag,
953 for (i = 0; i < FF_ARRAY_ELEMS(vslice->ref_pic_list0); i++) {
954 vslice->ref_pic_list0[i].picture_id = VA_INVALID_ID;
955 vslice->ref_pic_list0[i].flags = VA_PICTURE_HEVC_INVALID;
956 vslice->ref_pic_list1[i].picture_id = VA_INVALID_ID;
957 vslice->ref_pic_list1[i].flags = VA_PICTURE_HEVC_INVALID;
960 av_assert0(pic->nb_refs <= 2);
961 if (pic->nb_refs >= 1) {
962 // Backward reference for P- or B-frame.
963 av_assert0(pic->type == PICTURE_TYPE_P ||
964 pic->type == PICTURE_TYPE_B);
965 vslice->ref_pic_list0[0] = vpic->reference_frames[0];
967 if (pic->nb_refs >= 2) {
968 // Forward reference for B-frame.
969 av_assert0(pic->type == PICTURE_TYPE_B);
970 vslice->ref_pic_list1[0] = vpic->reference_frames[1];
976 static av_cold int vaapi_encode_h265_configure(AVCodecContext *avctx)
978 VAAPIEncodeContext *ctx = avctx->priv_data;
979 VAAPIEncodeH265Context *priv = avctx->priv_data;
982 err = ff_cbs_init(&priv->cbc, AV_CODEC_ID_HEVC, avctx);
986 priv->ctu_width = FFALIGN(ctx->surface_width, 32) / 32;
987 priv->ctu_height = FFALIGN(ctx->surface_height, 32) / 32;
989 av_log(avctx, AV_LOG_VERBOSE, "Input %ux%u -> Surface %ux%u -> CTU %ux%u.\n",
990 avctx->width, avctx->height, ctx->surface_width,
991 ctx->surface_height, priv->ctu_width, priv->ctu_height);
993 if (ctx->va_rc_mode == VA_RC_CQP) {
994 priv->fixed_qp_p = priv->qp;
995 if (avctx->i_quant_factor > 0.0)
996 priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor +
997 avctx->i_quant_offset) + 0.5);
999 priv->fixed_qp_idr = priv->fixed_qp_p;
1000 if (avctx->b_quant_factor > 0.0)
1001 priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor +
1002 avctx->b_quant_offset) + 0.5);
1004 priv->fixed_qp_b = priv->fixed_qp_p;
1006 av_log(avctx, AV_LOG_DEBUG, "Using fixed QP = "
1007 "%d / %d / %d for IDR- / P- / B-frames.\n",
1008 priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b);
1010 } else if (ctx->va_rc_mode == VA_RC_CBR ||
1011 ctx->va_rc_mode == VA_RC_VBR) {
1012 // These still need to be set for pic_init_qp/slice_qp_delta.
1013 priv->fixed_qp_idr = 30;
1014 priv->fixed_qp_p = 30;
1015 priv->fixed_qp_b = 30;
1017 av_log(avctx, AV_LOG_DEBUG, "Using %s-bitrate = %"PRId64" bps.\n",
1018 ctx->va_rc_mode == VA_RC_CBR ? "constant" : "variable",
1022 av_assert0(0 && "Invalid RC mode.");
1028 static const VAAPIEncodeType vaapi_encode_type_h265 = {
1029 .configure = &vaapi_encode_h265_configure,
1031 .sequence_params_size = sizeof(VAEncSequenceParameterBufferHEVC),
1032 .init_sequence_params = &vaapi_encode_h265_init_sequence_params,
1034 .picture_params_size = sizeof(VAEncPictureParameterBufferHEVC),
1035 .init_picture_params = &vaapi_encode_h265_init_picture_params,
1037 .slice_params_size = sizeof(VAEncSliceParameterBufferHEVC),
1038 .init_slice_params = &vaapi_encode_h265_init_slice_params,
1040 .sequence_header_type = VAEncPackedHeaderSequence,
1041 .write_sequence_header = &vaapi_encode_h265_write_sequence_header,
1043 .slice_header_type = VAEncPackedHeaderHEVC_Slice,
1044 .write_slice_header = &vaapi_encode_h265_write_slice_header,
1046 .write_extra_header = &vaapi_encode_h265_write_extra_header,
1049 static av_cold int vaapi_encode_h265_init(AVCodecContext *avctx)
1051 VAAPIEncodeContext *ctx = avctx->priv_data;
1052 VAAPIEncodeH265Context *priv = avctx->priv_data;
1054 ctx->codec = &vaapi_encode_type_h265;
1056 if (avctx->profile == FF_PROFILE_UNKNOWN)
1057 avctx->profile = priv->profile;
1058 if (avctx->level == FF_LEVEL_UNKNOWN)
1059 avctx->level = priv->level;
1061 switch (avctx->profile) {
1062 case FF_PROFILE_HEVC_MAIN:
1063 case FF_PROFILE_UNKNOWN:
1064 ctx->va_profile = VAProfileHEVCMain;
1065 ctx->va_rt_format = VA_RT_FORMAT_YUV420;
1067 case FF_PROFILE_HEVC_MAIN_10:
1068 #ifdef VA_RT_FORMAT_YUV420_10BPP
1069 ctx->va_profile = VAProfileHEVCMain10;
1070 ctx->va_rt_format = VA_RT_FORMAT_YUV420_10BPP;
1073 av_log(avctx, AV_LOG_ERROR, "10-bit encoding is not "
1074 "supported with this VAAPI version.\n");
1075 return AVERROR(ENOSYS);
1078 av_log(avctx, AV_LOG_ERROR, "Unknown H.265 profile %d.\n",
1080 return AVERROR(EINVAL);
1082 ctx->va_entrypoint = VAEntrypointEncSlice;
1084 if (avctx->bit_rate > 0) {
1085 if (avctx->rc_max_rate == avctx->bit_rate)
1086 ctx->va_rc_mode = VA_RC_CBR;
1088 ctx->va_rc_mode = VA_RC_VBR;
1090 ctx->va_rc_mode = VA_RC_CQP;
1092 ctx->va_packed_headers =
1093 VA_ENC_PACKED_HEADER_SEQUENCE | // VPS, SPS and PPS.
1094 VA_ENC_PACKED_HEADER_SLICE | // Slice headers.
1095 VA_ENC_PACKED_HEADER_MISC; // SEI
1097 ctx->surface_width = FFALIGN(avctx->width, 16);
1098 ctx->surface_height = FFALIGN(avctx->height, 16);
1100 return ff_vaapi_encode_init(avctx);
1103 static av_cold int vaapi_encode_h265_close(AVCodecContext *avctx)
1105 VAAPIEncodeH265Context *priv = avctx->priv_data;
1107 ff_cbs_close(&priv->cbc);
1109 return ff_vaapi_encode_close(avctx);
1112 #define OFFSET(x) offsetof(VAAPIEncodeH265Context, x)
1113 #define FLAGS (AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM)
1114 static const AVOption vaapi_encode_h265_options[] = {
1115 { "qp", "Constant QP (for P-frames; scaled by qfactor/qoffset for I/B)",
1116 OFFSET(qp), AV_OPT_TYPE_INT, { .i64 = 25 }, 0, 52, FLAGS },
1118 { "aud", "Include AUD",
1119 OFFSET(aud), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
1121 { "profile", "Set profile (general_profile_idc)",
1122 OFFSET(profile), AV_OPT_TYPE_INT,
1123 { .i64 = FF_PROFILE_HEVC_MAIN }, 0x00, 0xff, FLAGS, "profile" },
1125 #define PROFILE(name, value) name, NULL, 0, AV_OPT_TYPE_CONST, \
1126 { .i64 = value }, 0, 0, FLAGS, "profile"
1127 { PROFILE("main", FF_PROFILE_HEVC_MAIN) },
1128 { PROFILE("main10", FF_PROFILE_HEVC_MAIN_10) },
1131 { "level", "Set level (general_level_idc)",
1132 OFFSET(level), AV_OPT_TYPE_INT,
1133 { .i64 = 153 }, 0x00, 0xff, FLAGS, "level" },
1135 #define LEVEL(name, value) name, NULL, 0, AV_OPT_TYPE_CONST, \
1136 { .i64 = value }, 0, 0, FLAGS, "level"
1139 { LEVEL("2.1", 63) },
1141 { LEVEL("3.1", 93) },
1142 { LEVEL("4", 120) },
1143 { LEVEL("4.1", 123) },
1144 { LEVEL("5", 150) },
1145 { LEVEL("5.1", 153) },
1146 { LEVEL("5.2", 156) },
1147 { LEVEL("6", 180) },
1148 { LEVEL("6.1", 183) },
1149 { LEVEL("6.2", 186) },
1152 { "sei", "Set SEI to include",
1153 OFFSET(sei), AV_OPT_TYPE_FLAGS,
1154 { .i64 = SEI_MASTERING_DISPLAY | SEI_CONTENT_LIGHT_LEVEL },
1155 0, INT_MAX, FLAGS, "sei" },
1157 "Include HDR metadata for mastering display colour volume "
1158 "and content light level information",
1159 0, AV_OPT_TYPE_CONST,
1160 { .i64 = SEI_MASTERING_DISPLAY | SEI_CONTENT_LIGHT_LEVEL },
1161 INT_MIN, INT_MAX, FLAGS, "sei" },
1166 static const AVCodecDefault vaapi_encode_h265_defaults[] = {
1170 { "i_qfactor", "1" },
1171 { "i_qoffset", "0" },
1172 { "b_qfactor", "6/5" },
1173 { "b_qoffset", "0" },
1177 static const AVClass vaapi_encode_h265_class = {
1178 .class_name = "h265_vaapi",
1179 .item_name = av_default_item_name,
1180 .option = vaapi_encode_h265_options,
1181 .version = LIBAVUTIL_VERSION_INT,
1184 AVCodec ff_hevc_vaapi_encoder = {
1185 .name = "hevc_vaapi",
1186 .long_name = NULL_IF_CONFIG_SMALL("H.265/HEVC (VAAPI)"),
1187 .type = AVMEDIA_TYPE_VIDEO,
1188 .id = AV_CODEC_ID_HEVC,
1189 .priv_data_size = sizeof(VAAPIEncodeH265Context),
1190 .init = &vaapi_encode_h265_init,
1191 .encode2 = &ff_vaapi_encode2,
1192 .close = &vaapi_encode_h265_close,
1193 .priv_class = &vaapi_encode_h265_class,
1194 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE,
1195 .defaults = vaapi_encode_h265_defaults,
1196 .pix_fmts = (const enum AVPixelFormat[]) {
1200 .wrapper_name = "vaapi",