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/pixdesc.h"
27 #include "libavutil/opt.h"
28 #include "libavutil/mastering_display_metadata.h"
33 #include "h265_profile_level.h"
38 #include "vaapi_encode.h"
41 SEI_MASTERING_DISPLAY = 0x08,
42 SEI_CONTENT_LIGHT_LEVEL = 0x10,
45 typedef struct VAAPIEncodeH265Context {
46 VAAPIEncodeContext common;
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 H265RawProfileTierLevel *ptl = &vps->profile_tier_level;
264 H265RawVUI *vui = &sps->vui;
265 VAEncSequenceParameterBufferHEVC *vseq = ctx->codec_sequence_params;
266 VAEncPictureParameterBufferHEVC *vpic = ctx->codec_picture_params;
267 const AVPixFmtDescriptor *desc;
268 int chroma_format, bit_depth;
271 memset(&priv->current_access_unit, 0,
272 sizeof(priv->current_access_unit));
274 memset(vps, 0, sizeof(*vps));
275 memset(sps, 0, sizeof(*sps));
276 memset(pps, 0, sizeof(*pps));
279 desc = av_pix_fmt_desc_get(priv->common.input_frames->sw_format);
281 if (desc->nb_components == 1) {
284 if (desc->log2_chroma_w == 1 && desc->log2_chroma_h == 1) {
286 } else if (desc->log2_chroma_w == 1 && desc->log2_chroma_h == 0) {
288 } else if (desc->log2_chroma_w == 0 && desc->log2_chroma_h == 0) {
291 av_log(avctx, AV_LOG_ERROR, "Chroma format of input pixel format "
292 "%s is not supported.\n", desc->name);
293 return AVERROR(EINVAL);
296 bit_depth = desc->comp[0].depth;
301 vps->nal_unit_header = (H265RawNALUnitHeader) {
302 .nal_unit_type = HEVC_NAL_VPS,
304 .nuh_temporal_id_plus1 = 1,
307 vps->vps_video_parameter_set_id = 0;
309 vps->vps_base_layer_internal_flag = 1;
310 vps->vps_base_layer_available_flag = 1;
311 vps->vps_max_layers_minus1 = 0;
312 vps->vps_max_sub_layers_minus1 = 0;
313 vps->vps_temporal_id_nesting_flag = 1;
315 ptl->general_profile_space = 0;
316 ptl->general_profile_idc = avctx->profile;
317 ptl->general_tier_flag = priv->tier;
319 if (chroma_format == 1) {
320 ptl->general_profile_compatibility_flag[1] = bit_depth == 8;
321 ptl->general_profile_compatibility_flag[2] = bit_depth <= 10;
323 ptl->general_profile_compatibility_flag[4] = 1;
325 ptl->general_progressive_source_flag = 1;
326 ptl->general_interlaced_source_flag = 0;
327 ptl->general_non_packed_constraint_flag = 1;
328 ptl->general_frame_only_constraint_flag = 1;
330 ptl->general_max_12bit_constraint_flag = bit_depth <= 12;
331 ptl->general_max_10bit_constraint_flag = bit_depth <= 10;
332 ptl->general_max_8bit_constraint_flag = bit_depth == 8;
334 ptl->general_max_422chroma_constraint_flag = chroma_format <= 2;
335 ptl->general_max_420chroma_constraint_flag = chroma_format <= 1;
336 ptl->general_max_monochrome_constraint_flag = chroma_format == 0;
338 ptl->general_intra_constraint_flag = ctx->gop_size == 1;
340 ptl->general_lower_bit_rate_constraint_flag = 1;
342 if (avctx->level != FF_LEVEL_UNKNOWN) {
343 ptl->general_level_idc = avctx->level;
345 const H265LevelDescriptor *level;
347 level = ff_h265_guess_level(ptl, avctx->bit_rate,
348 ctx->surface_width, ctx->surface_height,
349 ctx->nb_slices, 1, 1,
350 (ctx->b_per_p > 0) + 1);
352 av_log(avctx, AV_LOG_VERBOSE, "Using level %s.\n", level->name);
353 ptl->general_level_idc = level->level_idc;
355 av_log(avctx, AV_LOG_VERBOSE, "Stream will not conform to "
356 "any normal level; using level 8.5.\n");
357 ptl->general_level_idc = 255;
358 // The tier flag must be set in level 8.5.
359 ptl->general_tier_flag = 1;
363 vps->vps_sub_layer_ordering_info_present_flag = 0;
364 vps->vps_max_dec_pic_buffering_minus1[0] = (ctx->b_per_p > 0) + 1;
365 vps->vps_max_num_reorder_pics[0] = (ctx->b_per_p > 0);
366 vps->vps_max_latency_increase_plus1[0] = 0;
368 vps->vps_max_layer_id = 0;
369 vps->vps_num_layer_sets_minus1 = 0;
370 vps->layer_id_included_flag[0][0] = 1;
372 vps->vps_timing_info_present_flag = 1;
373 if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {
374 vps->vps_num_units_in_tick = avctx->framerate.den;
375 vps->vps_time_scale = avctx->framerate.num;
376 vps->vps_poc_proportional_to_timing_flag = 1;
377 vps->vps_num_ticks_poc_diff_one_minus1 = 0;
379 vps->vps_num_units_in_tick = avctx->time_base.num;
380 vps->vps_time_scale = avctx->time_base.den;
381 vps->vps_poc_proportional_to_timing_flag = 0;
383 vps->vps_num_hrd_parameters = 0;
388 sps->nal_unit_header = (H265RawNALUnitHeader) {
389 .nal_unit_type = HEVC_NAL_SPS,
391 .nuh_temporal_id_plus1 = 1,
394 sps->sps_video_parameter_set_id = vps->vps_video_parameter_set_id;
396 sps->sps_max_sub_layers_minus1 = vps->vps_max_sub_layers_minus1;
397 sps->sps_temporal_id_nesting_flag = vps->vps_temporal_id_nesting_flag;
399 sps->profile_tier_level = vps->profile_tier_level;
401 sps->sps_seq_parameter_set_id = 0;
403 sps->chroma_format_idc = chroma_format;
404 sps->separate_colour_plane_flag = 0;
406 sps->pic_width_in_luma_samples = ctx->surface_width;
407 sps->pic_height_in_luma_samples = ctx->surface_height;
409 if (avctx->width != ctx->surface_width ||
410 avctx->height != ctx->surface_height) {
411 sps->conformance_window_flag = 1;
412 sps->conf_win_left_offset = 0;
413 sps->conf_win_right_offset =
414 (ctx->surface_width - avctx->width) / 2;
415 sps->conf_win_top_offset = 0;
416 sps->conf_win_bottom_offset =
417 (ctx->surface_height - avctx->height) / 2;
419 sps->conformance_window_flag = 0;
422 sps->bit_depth_luma_minus8 = bit_depth - 8;
423 sps->bit_depth_chroma_minus8 = bit_depth - 8;
425 sps->log2_max_pic_order_cnt_lsb_minus4 = 8;
427 sps->sps_sub_layer_ordering_info_present_flag =
428 vps->vps_sub_layer_ordering_info_present_flag;
429 for (i = 0; i <= sps->sps_max_sub_layers_minus1; i++) {
430 sps->sps_max_dec_pic_buffering_minus1[i] =
431 vps->vps_max_dec_pic_buffering_minus1[i];
432 sps->sps_max_num_reorder_pics[i] =
433 vps->vps_max_num_reorder_pics[i];
434 sps->sps_max_latency_increase_plus1[i] =
435 vps->vps_max_latency_increase_plus1[i];
438 // These have to come from the capabilities of the encoder. We have no
439 // way to query them, so just hardcode parameters which work on the Intel
441 // CTB size from 8x8 to 32x32.
442 sps->log2_min_luma_coding_block_size_minus3 = 0;
443 sps->log2_diff_max_min_luma_coding_block_size = 2;
444 // Transform size from 4x4 to 32x32.
445 sps->log2_min_luma_transform_block_size_minus2 = 0;
446 sps->log2_diff_max_min_luma_transform_block_size = 3;
447 // Full transform hierarchy allowed (2-5).
448 sps->max_transform_hierarchy_depth_inter = 3;
449 sps->max_transform_hierarchy_depth_intra = 3;
451 sps->amp_enabled_flag = 1;
452 // SAO and temporal MVP do not work.
453 sps->sample_adaptive_offset_enabled_flag = 0;
454 sps->sps_temporal_mvp_enabled_flag = 0;
456 sps->pcm_enabled_flag = 0;
458 // STRPSs should ideally be here rather than defined individually in
459 // each slice, but the structure isn't completely fixed so for now
461 sps->num_short_term_ref_pic_sets = 0;
462 sps->long_term_ref_pics_present_flag = 0;
464 sps->vui_parameters_present_flag = 1;
466 if (avctx->sample_aspect_ratio.num != 0 &&
467 avctx->sample_aspect_ratio.den != 0) {
468 static const AVRational sar_idc[] = {
470 { 1, 1 }, { 12, 11 }, { 10, 11 }, { 16, 11 },
471 { 40, 33 }, { 24, 11 }, { 20, 11 }, { 32, 11 },
472 { 80, 33 }, { 18, 11 }, { 15, 11 }, { 64, 33 },
473 { 160, 99 }, { 4, 3 }, { 3, 2 }, { 2, 1 },
476 av_reduce(&num, &den, avctx->sample_aspect_ratio.num,
477 avctx->sample_aspect_ratio.den, 65535);
478 for (i = 0; i < FF_ARRAY_ELEMS(sar_idc); i++) {
479 if (num == sar_idc[i].num &&
480 den == sar_idc[i].den) {
481 vui->aspect_ratio_idc = i;
485 if (i >= FF_ARRAY_ELEMS(sar_idc)) {
486 vui->aspect_ratio_idc = 255;
487 vui->sar_width = num;
488 vui->sar_height = den;
490 vui->aspect_ratio_info_present_flag = 1;
493 if (avctx->color_range != AVCOL_RANGE_UNSPECIFIED ||
494 avctx->color_primaries != AVCOL_PRI_UNSPECIFIED ||
495 avctx->color_trc != AVCOL_TRC_UNSPECIFIED ||
496 avctx->colorspace != AVCOL_SPC_UNSPECIFIED) {
497 vui->video_signal_type_present_flag = 1;
498 vui->video_format = 5; // Unspecified.
499 vui->video_full_range_flag =
500 avctx->color_range == AVCOL_RANGE_JPEG;
502 if (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED ||
503 avctx->color_trc != AVCOL_TRC_UNSPECIFIED ||
504 avctx->colorspace != AVCOL_SPC_UNSPECIFIED) {
505 vui->colour_description_present_flag = 1;
506 vui->colour_primaries = avctx->color_primaries;
507 vui->transfer_characteristics = avctx->color_trc;
508 vui->matrix_coefficients = avctx->colorspace;
511 vui->video_format = 5;
512 vui->video_full_range_flag = 0;
513 vui->colour_primaries = avctx->color_primaries;
514 vui->transfer_characteristics = avctx->color_trc;
515 vui->matrix_coefficients = avctx->colorspace;
518 if (avctx->chroma_sample_location != AVCHROMA_LOC_UNSPECIFIED) {
519 vui->chroma_loc_info_present_flag = 1;
520 vui->chroma_sample_loc_type_top_field =
521 vui->chroma_sample_loc_type_bottom_field =
522 avctx->chroma_sample_location - 1;
525 vui->vui_timing_info_present_flag = 1;
526 vui->vui_num_units_in_tick = vps->vps_num_units_in_tick;
527 vui->vui_time_scale = vps->vps_time_scale;
528 vui->vui_poc_proportional_to_timing_flag = vps->vps_poc_proportional_to_timing_flag;
529 vui->vui_num_ticks_poc_diff_one_minus1 = vps->vps_num_ticks_poc_diff_one_minus1;
530 vui->vui_hrd_parameters_present_flag = 0;
532 vui->bitstream_restriction_flag = 1;
533 vui->motion_vectors_over_pic_boundaries_flag = 1;
534 vui->restricted_ref_pic_lists_flag = 1;
535 vui->max_bytes_per_pic_denom = 0;
536 vui->max_bits_per_min_cu_denom = 0;
537 vui->log2_max_mv_length_horizontal = 15;
538 vui->log2_max_mv_length_vertical = 15;
543 pps->nal_unit_header = (H265RawNALUnitHeader) {
544 .nal_unit_type = HEVC_NAL_PPS,
546 .nuh_temporal_id_plus1 = 1,
549 pps->pps_pic_parameter_set_id = 0;
550 pps->pps_seq_parameter_set_id = sps->sps_seq_parameter_set_id;
552 pps->num_ref_idx_l0_default_active_minus1 = 0;
553 pps->num_ref_idx_l1_default_active_minus1 = 0;
555 pps->init_qp_minus26 = priv->fixed_qp_idr - 26;
557 pps->cu_qp_delta_enabled_flag = (ctx->va_rc_mode != VA_RC_CQP);
558 pps->diff_cu_qp_delta_depth = 0;
560 pps->pps_loop_filter_across_slices_enabled_flag = 1;
563 // Fill VAAPI parameter buffers.
565 *vseq = (VAEncSequenceParameterBufferHEVC) {
566 .general_profile_idc = vps->profile_tier_level.general_profile_idc,
567 .general_level_idc = vps->profile_tier_level.general_level_idc,
568 .general_tier_flag = vps->profile_tier_level.general_tier_flag,
570 .intra_period = ctx->gop_size,
571 .intra_idr_period = ctx->gop_size,
572 .ip_period = ctx->b_per_p + 1,
573 .bits_per_second = ctx->va_bit_rate,
575 .pic_width_in_luma_samples = sps->pic_width_in_luma_samples,
576 .pic_height_in_luma_samples = sps->pic_height_in_luma_samples,
579 .chroma_format_idc = sps->chroma_format_idc,
580 .separate_colour_plane_flag = sps->separate_colour_plane_flag,
581 .bit_depth_luma_minus8 = sps->bit_depth_luma_minus8,
582 .bit_depth_chroma_minus8 = sps->bit_depth_chroma_minus8,
583 .scaling_list_enabled_flag = sps->scaling_list_enabled_flag,
584 .strong_intra_smoothing_enabled_flag =
585 sps->strong_intra_smoothing_enabled_flag,
586 .amp_enabled_flag = sps->amp_enabled_flag,
587 .sample_adaptive_offset_enabled_flag =
588 sps->sample_adaptive_offset_enabled_flag,
589 .pcm_enabled_flag = sps->pcm_enabled_flag,
590 .pcm_loop_filter_disabled_flag = sps->pcm_loop_filter_disabled_flag,
591 .sps_temporal_mvp_enabled_flag = sps->sps_temporal_mvp_enabled_flag,
594 .log2_min_luma_coding_block_size_minus3 =
595 sps->log2_min_luma_coding_block_size_minus3,
596 .log2_diff_max_min_luma_coding_block_size =
597 sps->log2_diff_max_min_luma_coding_block_size,
598 .log2_min_transform_block_size_minus2 =
599 sps->log2_min_luma_transform_block_size_minus2,
600 .log2_diff_max_min_transform_block_size =
601 sps->log2_diff_max_min_luma_transform_block_size,
602 .max_transform_hierarchy_depth_inter =
603 sps->max_transform_hierarchy_depth_inter,
604 .max_transform_hierarchy_depth_intra =
605 sps->max_transform_hierarchy_depth_intra,
607 .pcm_sample_bit_depth_luma_minus1 =
608 sps->pcm_sample_bit_depth_luma_minus1,
609 .pcm_sample_bit_depth_chroma_minus1 =
610 sps->pcm_sample_bit_depth_chroma_minus1,
611 .log2_min_pcm_luma_coding_block_size_minus3 =
612 sps->log2_min_pcm_luma_coding_block_size_minus3,
613 .log2_max_pcm_luma_coding_block_size_minus3 =
614 sps->log2_min_pcm_luma_coding_block_size_minus3 +
615 sps->log2_diff_max_min_pcm_luma_coding_block_size,
617 .vui_parameters_present_flag = 0,
620 *vpic = (VAEncPictureParameterBufferHEVC) {
621 .decoded_curr_pic = {
622 .picture_id = VA_INVALID_ID,
623 .flags = VA_PICTURE_HEVC_INVALID,
626 .coded_buf = VA_INVALID_ID,
628 .collocated_ref_pic_index = 0xff,
632 .pic_init_qp = pps->init_qp_minus26 + 26,
633 .diff_cu_qp_delta_depth = pps->diff_cu_qp_delta_depth,
634 .pps_cb_qp_offset = pps->pps_cb_qp_offset,
635 .pps_cr_qp_offset = pps->pps_cr_qp_offset,
637 .num_tile_columns_minus1 = pps->num_tile_columns_minus1,
638 .num_tile_rows_minus1 = pps->num_tile_rows_minus1,
640 .log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level_minus2,
641 .ctu_max_bitsize_allowed = 0,
643 .num_ref_idx_l0_default_active_minus1 =
644 pps->num_ref_idx_l0_default_active_minus1,
645 .num_ref_idx_l1_default_active_minus1 =
646 pps->num_ref_idx_l1_default_active_minus1,
648 .slice_pic_parameter_set_id = pps->pps_pic_parameter_set_id,
651 .sign_data_hiding_enabled_flag = pps->sign_data_hiding_enabled_flag,
652 .constrained_intra_pred_flag = pps->constrained_intra_pred_flag,
653 .transform_skip_enabled_flag = pps->transform_skip_enabled_flag,
654 .cu_qp_delta_enabled_flag = pps->cu_qp_delta_enabled_flag,
655 .weighted_pred_flag = pps->weighted_pred_flag,
656 .weighted_bipred_flag = pps->weighted_bipred_flag,
657 .transquant_bypass_enabled_flag = pps->transquant_bypass_enabled_flag,
658 .tiles_enabled_flag = pps->tiles_enabled_flag,
659 .entropy_coding_sync_enabled_flag = pps->entropy_coding_sync_enabled_flag,
660 .loop_filter_across_tiles_enabled_flag =
661 pps->loop_filter_across_tiles_enabled_flag,
662 .scaling_list_data_present_flag = (sps->sps_scaling_list_data_present_flag |
663 pps->pps_scaling_list_data_present_flag),
664 .screen_content_flag = 0,
665 .enable_gpu_weighted_prediction = 0,
666 .no_output_of_prior_pics_flag = 0,
673 static int vaapi_encode_h265_init_picture_params(AVCodecContext *avctx,
674 VAAPIEncodePicture *pic)
676 VAAPIEncodeH265Context *priv = avctx->priv_data;
677 VAEncPictureParameterBufferHEVC *vpic = pic->codec_picture_params;
680 if (pic->type == PICTURE_TYPE_IDR) {
681 av_assert0(pic->display_order == pic->encode_order);
683 priv->last_idr_frame = pic->display_order;
685 priv->slice_nal_unit = HEVC_NAL_IDR_W_RADL;
686 priv->slice_type = HEVC_SLICE_I;
689 av_assert0(pic->encode_order > priv->last_idr_frame);
691 if (pic->type == PICTURE_TYPE_I) {
692 priv->slice_nal_unit = HEVC_NAL_CRA_NUT;
693 priv->slice_type = HEVC_SLICE_I;
695 } else if (pic->type == PICTURE_TYPE_P) {
696 av_assert0(pic->refs[0]);
697 priv->slice_nal_unit = HEVC_NAL_TRAIL_R;
698 priv->slice_type = HEVC_SLICE_P;
701 av_assert0(pic->refs[0] && pic->refs[1]);
702 if (pic->refs[1]->type == PICTURE_TYPE_I)
703 priv->slice_nal_unit = HEVC_NAL_RASL_N;
705 priv->slice_nal_unit = HEVC_NAL_TRAIL_N;
706 priv->slice_type = HEVC_SLICE_B;
710 priv->pic_order_cnt = pic->display_order - priv->last_idr_frame;
713 priv->aud_needed = 1;
714 priv->raw_aud = (H265RawAUD) {
716 .nal_unit_type = HEVC_NAL_AUD,
718 .nuh_temporal_id_plus1 = 1,
720 .pic_type = priv->pic_type,
723 priv->aud_needed = 0;
726 priv->sei_needed = 0;
728 // Only look for the metadata on I/IDR frame on the output. We
729 // may force an IDR frame on the output where the medadata gets
730 // changed on the input frame.
731 if ((priv->sei & SEI_MASTERING_DISPLAY) &&
732 (pic->type == PICTURE_TYPE_I || pic->type == PICTURE_TYPE_IDR)) {
733 AVFrameSideData *sd =
734 av_frame_get_side_data(pic->input_image,
735 AV_FRAME_DATA_MASTERING_DISPLAY_METADATA);
738 AVMasteringDisplayMetadata *mdm =
739 (AVMasteringDisplayMetadata *)sd->data;
741 // SEI is needed when both the primaries and luminance are set
742 if (mdm->has_primaries && mdm->has_luminance) {
743 H265RawSEIMasteringDisplayColourVolume *mdcv =
744 &priv->sei_mastering_display;
745 const int mapping[3] = {1, 2, 0};
746 const int chroma_den = 50000;
747 const int luma_den = 10000;
749 for (i = 0; i < 3; i++) {
750 const int j = mapping[i];
751 mdcv->display_primaries_x[i] =
752 FFMIN(lrint(chroma_den *
753 av_q2d(mdm->display_primaries[j][0])),
755 mdcv->display_primaries_y[i] =
756 FFMIN(lrint(chroma_den *
757 av_q2d(mdm->display_primaries[j][1])),
761 mdcv->white_point_x =
762 FFMIN(lrint(chroma_den * av_q2d(mdm->white_point[0])),
764 mdcv->white_point_y =
765 FFMIN(lrint(chroma_den * av_q2d(mdm->white_point[1])),
768 mdcv->max_display_mastering_luminance =
769 lrint(luma_den * av_q2d(mdm->max_luminance));
770 mdcv->min_display_mastering_luminance =
771 FFMIN(lrint(luma_den * av_q2d(mdm->min_luminance)),
772 mdcv->max_display_mastering_luminance);
774 priv->sei_needed |= SEI_MASTERING_DISPLAY;
779 if ((priv->sei & SEI_CONTENT_LIGHT_LEVEL) &&
780 (pic->type == PICTURE_TYPE_I || pic->type == PICTURE_TYPE_IDR)) {
781 AVFrameSideData *sd =
782 av_frame_get_side_data(pic->input_image,
783 AV_FRAME_DATA_CONTENT_LIGHT_LEVEL);
786 AVContentLightMetadata *clm =
787 (AVContentLightMetadata *)sd->data;
788 H265RawSEIContentLightLevelInfo *clli =
789 &priv->sei_content_light_level;
791 clli->max_content_light_level = FFMIN(clm->MaxCLL, 65535);
792 clli->max_pic_average_light_level = FFMIN(clm->MaxFALL, 65535);
794 priv->sei_needed |= SEI_CONTENT_LIGHT_LEVEL;
798 vpic->decoded_curr_pic = (VAPictureHEVC) {
799 .picture_id = pic->recon_surface,
800 .pic_order_cnt = priv->pic_order_cnt,
804 for (i = 0; i < pic->nb_refs; i++) {
805 VAAPIEncodePicture *ref = pic->refs[i];
806 av_assert0(ref && ref->encode_order < pic->encode_order);
808 vpic->reference_frames[i] = (VAPictureHEVC) {
809 .picture_id = ref->recon_surface,
810 .pic_order_cnt = ref->display_order - priv->last_idr_frame,
811 .flags = (ref->display_order < pic->display_order ?
812 VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE : 0) |
813 (ref->display_order > pic->display_order ?
814 VA_PICTURE_HEVC_RPS_ST_CURR_AFTER : 0),
817 for (; i < FF_ARRAY_ELEMS(vpic->reference_frames); i++) {
818 vpic->reference_frames[i] = (VAPictureHEVC) {
819 .picture_id = VA_INVALID_ID,
820 .flags = VA_PICTURE_HEVC_INVALID,
824 vpic->coded_buf = pic->output_buffer;
826 vpic->nal_unit_type = priv->slice_nal_unit;
829 case PICTURE_TYPE_IDR:
830 vpic->pic_fields.bits.idr_pic_flag = 1;
831 vpic->pic_fields.bits.coding_type = 1;
832 vpic->pic_fields.bits.reference_pic_flag = 1;
835 vpic->pic_fields.bits.idr_pic_flag = 0;
836 vpic->pic_fields.bits.coding_type = 1;
837 vpic->pic_fields.bits.reference_pic_flag = 1;
840 vpic->pic_fields.bits.idr_pic_flag = 0;
841 vpic->pic_fields.bits.coding_type = 2;
842 vpic->pic_fields.bits.reference_pic_flag = 1;
845 vpic->pic_fields.bits.idr_pic_flag = 0;
846 vpic->pic_fields.bits.coding_type = 3;
847 vpic->pic_fields.bits.reference_pic_flag = 0;
850 av_assert0(0 && "invalid picture type");
856 static int vaapi_encode_h265_init_slice_params(AVCodecContext *avctx,
857 VAAPIEncodePicture *pic,
858 VAAPIEncodeSlice *slice)
860 VAAPIEncodeContext *ctx = avctx->priv_data;
861 VAAPIEncodeH265Context *priv = avctx->priv_data;
862 const H265RawSPS *sps = &priv->raw_sps;
863 const H265RawPPS *pps = &priv->raw_pps;
864 H265RawSliceHeader *sh = &priv->raw_slice.header;
865 VAEncPictureParameterBufferHEVC *vpic = pic->codec_picture_params;
866 VAEncSliceParameterBufferHEVC *vslice = slice->codec_slice_params;
869 sh->nal_unit_header = (H265RawNALUnitHeader) {
870 .nal_unit_type = priv->slice_nal_unit,
872 .nuh_temporal_id_plus1 = 1,
875 sh->slice_pic_parameter_set_id = pps->pps_pic_parameter_set_id;
877 sh->first_slice_segment_in_pic_flag = slice->index == 0;
878 sh->slice_segment_address = slice->block_start;
880 sh->slice_type = priv->slice_type;
882 sh->slice_pic_order_cnt_lsb = priv->pic_order_cnt &
883 (1 << (sps->log2_max_pic_order_cnt_lsb_minus4 + 4)) - 1;
885 if (pic->type != PICTURE_TYPE_IDR) {
886 H265RawSTRefPicSet *rps;
887 VAAPIEncodePicture *st;
890 sh->short_term_ref_pic_set_sps_flag = 0;
892 rps = &sh->short_term_ref_pic_set;
893 memset(rps, 0, sizeof(*rps));
895 for (st = ctx->pic_start; st; st = st->next) {
896 if (st->encode_order >= pic->encode_order) {
901 for (i = 0; i < pic->nb_refs; i++) {
902 if (pic->refs[i] == st)
906 // Usually each picture always uses all of the others in the
907 // DPB as references. The one case we have to treat here is
908 // a non-IDR IRAP picture, which may need to hold unused
909 // references across itself to be used for the decoding of
910 // following RASL pictures. This looks for such an RASL
911 // picture, and keeps the reference if there is one.
912 VAAPIEncodePicture *rp;
913 for (rp = ctx->pic_start; rp; rp = rp->next) {
914 if (rp->encode_order < pic->encode_order)
916 if (rp->type != PICTURE_TYPE_B)
918 if (rp->refs[0] == st && rp->refs[1] == pic)
924 // This only works for one instance of each (delta_poc_sN_minus1
925 // is relative to the previous frame in the list, not relative to
926 // the current frame directly).
927 if (st->display_order < pic->display_order) {
928 rps->delta_poc_s0_minus1[rps->num_negative_pics] =
929 pic->display_order - st->display_order - 1;
930 rps->used_by_curr_pic_s0_flag[rps->num_negative_pics] = used;
931 ++rps->num_negative_pics;
933 rps->delta_poc_s1_minus1[rps->num_positive_pics] =
934 st->display_order - pic->display_order - 1;
935 rps->used_by_curr_pic_s1_flag[rps->num_positive_pics] = used;
936 ++rps->num_positive_pics;
940 sh->num_long_term_sps = 0;
941 sh->num_long_term_pics = 0;
943 sh->slice_temporal_mvp_enabled_flag =
944 sps->sps_temporal_mvp_enabled_flag;
945 if (sh->slice_temporal_mvp_enabled_flag) {
946 sh->collocated_from_l0_flag = sh->slice_type == HEVC_SLICE_B;
947 sh->collocated_ref_idx = 0;
950 sh->num_ref_idx_active_override_flag = 0;
951 sh->num_ref_idx_l0_active_minus1 = pps->num_ref_idx_l0_default_active_minus1;
952 sh->num_ref_idx_l1_active_minus1 = pps->num_ref_idx_l1_default_active_minus1;
955 sh->slice_sao_luma_flag = sh->slice_sao_chroma_flag =
956 sps->sample_adaptive_offset_enabled_flag;
958 if (pic->type == PICTURE_TYPE_B)
959 sh->slice_qp_delta = priv->fixed_qp_b - (pps->init_qp_minus26 + 26);
960 else if (pic->type == PICTURE_TYPE_P)
961 sh->slice_qp_delta = priv->fixed_qp_p - (pps->init_qp_minus26 + 26);
963 sh->slice_qp_delta = priv->fixed_qp_idr - (pps->init_qp_minus26 + 26);
966 *vslice = (VAEncSliceParameterBufferHEVC) {
967 .slice_segment_address = sh->slice_segment_address,
968 .num_ctu_in_slice = slice->block_size,
970 .slice_type = sh->slice_type,
971 .slice_pic_parameter_set_id = sh->slice_pic_parameter_set_id,
973 .num_ref_idx_l0_active_minus1 = sh->num_ref_idx_l0_active_minus1,
974 .num_ref_idx_l1_active_minus1 = sh->num_ref_idx_l1_active_minus1,
976 .luma_log2_weight_denom = sh->luma_log2_weight_denom,
977 .delta_chroma_log2_weight_denom = sh->delta_chroma_log2_weight_denom,
979 .max_num_merge_cand = 5 - sh->five_minus_max_num_merge_cand,
981 .slice_qp_delta = sh->slice_qp_delta,
982 .slice_cb_qp_offset = sh->slice_cb_qp_offset,
983 .slice_cr_qp_offset = sh->slice_cr_qp_offset,
985 .slice_beta_offset_div2 = sh->slice_beta_offset_div2,
986 .slice_tc_offset_div2 = sh->slice_tc_offset_div2,
988 .slice_fields.bits = {
989 .last_slice_of_pic_flag = slice->index == pic->nb_slices - 1,
990 .dependent_slice_segment_flag = sh->dependent_slice_segment_flag,
991 .colour_plane_id = sh->colour_plane_id,
992 .slice_temporal_mvp_enabled_flag =
993 sh->slice_temporal_mvp_enabled_flag,
994 .slice_sao_luma_flag = sh->slice_sao_luma_flag,
995 .slice_sao_chroma_flag = sh->slice_sao_chroma_flag,
996 .num_ref_idx_active_override_flag =
997 sh->num_ref_idx_active_override_flag,
998 .mvd_l1_zero_flag = sh->mvd_l1_zero_flag,
999 .cabac_init_flag = sh->cabac_init_flag,
1000 .slice_deblocking_filter_disabled_flag =
1001 sh->slice_deblocking_filter_disabled_flag,
1002 .slice_loop_filter_across_slices_enabled_flag =
1003 sh->slice_loop_filter_across_slices_enabled_flag,
1004 .collocated_from_l0_flag = sh->collocated_from_l0_flag,
1008 for (i = 0; i < FF_ARRAY_ELEMS(vslice->ref_pic_list0); i++) {
1009 vslice->ref_pic_list0[i].picture_id = VA_INVALID_ID;
1010 vslice->ref_pic_list0[i].flags = VA_PICTURE_HEVC_INVALID;
1011 vslice->ref_pic_list1[i].picture_id = VA_INVALID_ID;
1012 vslice->ref_pic_list1[i].flags = VA_PICTURE_HEVC_INVALID;
1015 av_assert0(pic->nb_refs <= 2);
1016 if (pic->nb_refs >= 1) {
1017 // Backward reference for P- or B-frame.
1018 av_assert0(pic->type == PICTURE_TYPE_P ||
1019 pic->type == PICTURE_TYPE_B);
1020 vslice->ref_pic_list0[0] = vpic->reference_frames[0];
1022 if (pic->nb_refs >= 2) {
1023 // Forward reference for B-frame.
1024 av_assert0(pic->type == PICTURE_TYPE_B);
1025 vslice->ref_pic_list1[0] = vpic->reference_frames[1];
1031 static av_cold int vaapi_encode_h265_configure(AVCodecContext *avctx)
1033 VAAPIEncodeContext *ctx = avctx->priv_data;
1034 VAAPIEncodeH265Context *priv = avctx->priv_data;
1037 err = ff_cbs_init(&priv->cbc, AV_CODEC_ID_HEVC, avctx);
1041 if (ctx->va_rc_mode == VA_RC_CQP) {
1042 priv->fixed_qp_p = priv->qp;
1043 if (avctx->i_quant_factor > 0.0)
1044 priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor +
1045 avctx->i_quant_offset) + 0.5);
1047 priv->fixed_qp_idr = priv->fixed_qp_p;
1048 if (avctx->b_quant_factor > 0.0)
1049 priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor +
1050 avctx->b_quant_offset) + 0.5);
1052 priv->fixed_qp_b = priv->fixed_qp_p;
1054 av_log(avctx, AV_LOG_DEBUG, "Using fixed QP = "
1055 "%d / %d / %d for IDR- / P- / B-frames.\n",
1056 priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b);
1058 } else if (ctx->va_rc_mode == VA_RC_CBR ||
1059 ctx->va_rc_mode == VA_RC_VBR) {
1060 // These still need to be set for pic_init_qp/slice_qp_delta.
1061 priv->fixed_qp_idr = 30;
1062 priv->fixed_qp_p = 30;
1063 priv->fixed_qp_b = 30;
1066 av_assert0(0 && "Invalid RC mode.");
1072 static const VAAPIEncodeProfile vaapi_encode_h265_profiles[] = {
1073 { FF_PROFILE_HEVC_MAIN, 8, 3, 1, 1, VAProfileHEVCMain },
1074 { FF_PROFILE_HEVC_REXT, 8, 3, 1, 1, VAProfileHEVCMain },
1075 #if VA_CHECK_VERSION(0, 37, 0)
1076 { FF_PROFILE_HEVC_MAIN_10, 10, 3, 1, 1, VAProfileHEVCMain10 },
1077 { FF_PROFILE_HEVC_REXT, 10, 3, 1, 1, VAProfileHEVCMain10 },
1079 { FF_PROFILE_UNKNOWN }
1082 static const VAAPIEncodeType vaapi_encode_type_h265 = {
1083 .profiles = vaapi_encode_h265_profiles,
1085 .flags = FLAG_SLICE_CONTROL,
1087 .configure = &vaapi_encode_h265_configure,
1089 .sequence_params_size = sizeof(VAEncSequenceParameterBufferHEVC),
1090 .init_sequence_params = &vaapi_encode_h265_init_sequence_params,
1092 .picture_params_size = sizeof(VAEncPictureParameterBufferHEVC),
1093 .init_picture_params = &vaapi_encode_h265_init_picture_params,
1095 .slice_params_size = sizeof(VAEncSliceParameterBufferHEVC),
1096 .init_slice_params = &vaapi_encode_h265_init_slice_params,
1098 .sequence_header_type = VAEncPackedHeaderSequence,
1099 .write_sequence_header = &vaapi_encode_h265_write_sequence_header,
1101 .slice_header_type = VAEncPackedHeaderHEVC_Slice,
1102 .write_slice_header = &vaapi_encode_h265_write_slice_header,
1104 .write_extra_header = &vaapi_encode_h265_write_extra_header,
1107 static av_cold int vaapi_encode_h265_init(AVCodecContext *avctx)
1109 VAAPIEncodeContext *ctx = avctx->priv_data;
1110 VAAPIEncodeH265Context *priv = avctx->priv_data;
1112 ctx->codec = &vaapi_encode_type_h265;
1114 if (avctx->profile == FF_PROFILE_UNKNOWN)
1115 avctx->profile = priv->profile;
1116 if (avctx->level == FF_LEVEL_UNKNOWN)
1117 avctx->level = priv->level;
1119 if (avctx->level != FF_LEVEL_UNKNOWN && avctx->level & ~0xff) {
1120 av_log(avctx, AV_LOG_ERROR, "Invalid level %d: must fit "
1121 "in 8-bit unsigned integer.\n", avctx->level);
1122 return AVERROR(EINVAL);
1125 ctx->desired_packed_headers =
1126 VA_ENC_PACKED_HEADER_SEQUENCE | // VPS, SPS and PPS.
1127 VA_ENC_PACKED_HEADER_SLICE | // Slice headers.
1128 VA_ENC_PACKED_HEADER_MISC; // SEI
1130 ctx->surface_width = FFALIGN(avctx->width, 16);
1131 ctx->surface_height = FFALIGN(avctx->height, 16);
1133 // CTU size is currently hard-coded to 32.
1134 ctx->slice_block_width = ctx->slice_block_height = 32;
1136 return ff_vaapi_encode_init(avctx);
1139 static av_cold int vaapi_encode_h265_close(AVCodecContext *avctx)
1141 VAAPIEncodeH265Context *priv = avctx->priv_data;
1143 ff_cbs_close(&priv->cbc);
1145 return ff_vaapi_encode_close(avctx);
1148 #define OFFSET(x) offsetof(VAAPIEncodeH265Context, x)
1149 #define FLAGS (AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM)
1150 static const AVOption vaapi_encode_h265_options[] = {
1151 VAAPI_ENCODE_COMMON_OPTIONS,
1153 { "qp", "Constant QP (for P-frames; scaled by qfactor/qoffset for I/B)",
1154 OFFSET(qp), AV_OPT_TYPE_INT, { .i64 = 25 }, 0, 52, FLAGS },
1156 { "aud", "Include AUD",
1157 OFFSET(aud), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
1159 { "profile", "Set profile (general_profile_idc)",
1160 OFFSET(profile), AV_OPT_TYPE_INT,
1161 { .i64 = FF_PROFILE_UNKNOWN }, FF_PROFILE_UNKNOWN, 0xff, FLAGS, "profile" },
1163 #define PROFILE(name, value) name, NULL, 0, AV_OPT_TYPE_CONST, \
1164 { .i64 = value }, 0, 0, FLAGS, "profile"
1165 { PROFILE("main", FF_PROFILE_HEVC_MAIN) },
1166 { PROFILE("main10", FF_PROFILE_HEVC_MAIN_10) },
1167 { PROFILE("rext", FF_PROFILE_HEVC_REXT) },
1170 { "tier", "Set tier (general_tier_flag)",
1171 OFFSET(tier), AV_OPT_TYPE_INT,
1172 { .i64 = 0 }, 0, 1, FLAGS, "tier" },
1173 { "main", NULL, 0, AV_OPT_TYPE_CONST,
1174 { .i64 = 0 }, 0, 0, FLAGS, "tier" },
1175 { "high", NULL, 0, AV_OPT_TYPE_CONST,
1176 { .i64 = 1 }, 0, 0, FLAGS, "tier" },
1178 { "level", "Set level (general_level_idc)",
1179 OFFSET(level), AV_OPT_TYPE_INT,
1180 { .i64 = FF_LEVEL_UNKNOWN }, FF_LEVEL_UNKNOWN, 0xff, FLAGS, "level" },
1182 #define LEVEL(name, value) name, NULL, 0, AV_OPT_TYPE_CONST, \
1183 { .i64 = value }, 0, 0, FLAGS, "level"
1186 { LEVEL("2.1", 63) },
1188 { LEVEL("3.1", 93) },
1189 { LEVEL("4", 120) },
1190 { LEVEL("4.1", 123) },
1191 { LEVEL("5", 150) },
1192 { LEVEL("5.1", 153) },
1193 { LEVEL("5.2", 156) },
1194 { LEVEL("6", 180) },
1195 { LEVEL("6.1", 183) },
1196 { LEVEL("6.2", 186) },
1199 { "sei", "Set SEI to include",
1200 OFFSET(sei), AV_OPT_TYPE_FLAGS,
1201 { .i64 = SEI_MASTERING_DISPLAY | SEI_CONTENT_LIGHT_LEVEL },
1202 0, INT_MAX, FLAGS, "sei" },
1204 "Include HDR metadata for mastering display colour volume "
1205 "and content light level information",
1206 0, AV_OPT_TYPE_CONST,
1207 { .i64 = SEI_MASTERING_DISPLAY | SEI_CONTENT_LIGHT_LEVEL },
1208 INT_MIN, INT_MAX, FLAGS, "sei" },
1213 static const AVCodecDefault vaapi_encode_h265_defaults[] = {
1217 { "i_qfactor", "1" },
1218 { "i_qoffset", "0" },
1219 { "b_qfactor", "6/5" },
1220 { "b_qoffset", "0" },
1226 static const AVClass vaapi_encode_h265_class = {
1227 .class_name = "h265_vaapi",
1228 .item_name = av_default_item_name,
1229 .option = vaapi_encode_h265_options,
1230 .version = LIBAVUTIL_VERSION_INT,
1233 AVCodec ff_hevc_vaapi_encoder = {
1234 .name = "hevc_vaapi",
1235 .long_name = NULL_IF_CONFIG_SMALL("H.265/HEVC (VAAPI)"),
1236 .type = AVMEDIA_TYPE_VIDEO,
1237 .id = AV_CODEC_ID_HEVC,
1238 .priv_data_size = sizeof(VAAPIEncodeH265Context),
1239 .init = &vaapi_encode_h265_init,
1240 .encode2 = &ff_vaapi_encode2,
1241 .close = &vaapi_encode_h265_close,
1242 .priv_class = &vaapi_encode_h265_class,
1243 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE,
1244 .defaults = vaapi_encode_h265_defaults,
1245 .pix_fmts = (const enum AVPixelFormat[]) {
1249 .wrapper_name = "vaapi",