FFmpeg coverage

Line	Branch	Exec	Source
1			/*
2			* AV1 HW decode acceleration through VA API
3			*
4			* This file is part of FFmpeg.
5			*
6			* FFmpeg is free software; you can redistribute it and/or
7			* modify it under the terms of the GNU Lesser General Public
8			* License as published by the Free Software Foundation; either
9			* version 2.1 of the License, or (at your option) any later version.
10			*
11			* FFmpeg is distributed in the hope that it will be useful,
12			* but WITHOUT ANY WARRANTY; without even the implied warranty of
13			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14			* Lesser General Public License for more details.
15			*
16			* You should have received a copy of the GNU Lesser General Public
17			* License along with FFmpeg; if not, write to the Free Software
18			* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19			*/
20
21			#include "libavutil/frame.h"
22			#include "libavutil/mem.h"
23			#include "hwaccel_internal.h"
24			#include "vaapi_decode.h"
25			#include "internal.h"
26			#include "av1dec.h"
27			#include "thread.h"
28
29			typedef struct VAAPIAV1FrameRef {
30			AVFrame *frame;
31			int valid;
32			} VAAPIAV1FrameRef;
33
34			typedef struct VAAPIAV1DecContext {
35			VAAPIDecodeContext base;
36
37			/**
38			* For film grain case, VAAPI generate 2 output for each frame,
39			* current_frame will not apply film grain, and will be used for
40			* references for next frames. Maintain the reference list without
41			* applying film grain here. And current_display_picture will be
42			* used to apply film grain and push to downstream.
43			*/
44			VAAPIAV1FrameRef ref_tab[AV1_NUM_REF_FRAMES];
45			AVFrame *tmp_frame;
46
47			int nb_slice_params;
48			VASliceParameterBufferAV1 *slice_params;
49			} VAAPIAV1DecContext;
50
51		✗	static VASurfaceID vaapi_av1_surface_id(AV1Frame *vf)
52			{
53		✗	if (vf->f)
54		✗	return ff_vaapi_get_surface_id(vf->f);
55			else
56		✗	return VA_INVALID_SURFACE;
57			}
58
59		✗	static int8_t vaapi_av1_get_bit_depth_idx(AVCodecContext *avctx)
60			{
61		✗	AV1DecContext *s = avctx->priv_data;
62		✗	const AV1RawSequenceHeader *seq = s->raw_seq;
63		✗	int8_t bit_depth = 8;
64
65		✗	if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
66		✗	bit_depth = seq->color_config.twelve_bit ? 12 : 10;
67		✗	else if (seq->seq_profile <= 2)
68		✗	bit_depth = seq->color_config.high_bitdepth ? 10 : 8;
69			else {
70		✗	av_log(avctx, AV_LOG_ERROR,
71		✗	"Couldn't get bit depth from profile:%d.\n", seq->seq_profile);
72		✗	return -1;
73			}
74		✗	return bit_depth == 8 ? 0 : bit_depth == 10 ? 1 : 2;
75			}
76
77		✗	static av_cold int vaapi_av1_decode_uninit(AVCodecContext *avctx)
78			{
79		✗	VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
80
81		✗	av_frame_free(&ctx->tmp_frame);
82
83		✗	for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++)
84		✗	av_frame_free(&ctx->ref_tab[i].frame);
85
86		✗	av_freep(&ctx->slice_params);
87
88		✗	return ff_vaapi_decode_uninit(avctx);
89			}
90
91		✗	static av_cold int vaapi_av1_decode_init(AVCodecContext *avctx)
92			{
93		✗	VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
94
95		✗	int ret = ff_vaapi_decode_init(avctx);
96		✗	if (ret < 0)
97		✗	return ret;
98
99		✗	ctx->tmp_frame = av_frame_alloc();
100		✗	if (!ctx->tmp_frame) {
101		✗	vaapi_av1_decode_uninit(avctx);
102		✗	return AVERROR(ENOMEM);
103			}
104
105		✗	for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++) {
106		✗	ctx->ref_tab[i].frame = av_frame_alloc();
107		✗	if (!ctx->ref_tab[i].frame) {
108		✗	vaapi_av1_decode_uninit(avctx);
109		✗	return AVERROR(ENOMEM);
110			}
111		✗	ctx->ref_tab[i].valid = 0;
112			}
113
114		✗	return ret;
115			}
116
117		✗	static int vaapi_av1_start_frame(AVCodecContext *avctx,
118			av_unused const AVBufferRef *buffer_ref,
119			av_unused const uint8_t *buffer,
120			av_unused uint32_t size)
121			{
122		✗	AV1DecContext *s = avctx->priv_data;
123		✗	const AV1RawSequenceHeader *seq = s->raw_seq;
124		✗	const AV1RawFrameHeader *frame_header = s->raw_frame_header;
125		✗	const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
126		✗	VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
127		✗	VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
128			VADecPictureParameterBufferAV1 pic_param;
129			int8_t bit_depth_idx;
130		✗	int err = 0;
131		✗	int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
132		✗	uint8_t remap_lr_type[4] = {AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ};
133		✗	uint8_t segmentation_feature_signed[AV1_SEG_LVL_MAX] = {1, 1, 1, 1, 1, 0, 0, 0};
134		✗	uint8_t segmentation_feature_max[AV1_SEG_LVL_MAX] = {255, AV1_MAX_LOOP_FILTER,
135			AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, 7 , 0 , 0 };
136
137		✗	bit_depth_idx = vaapi_av1_get_bit_depth_idx(avctx);
138		✗	if (bit_depth_idx < 0)
139		✗	goto fail;
140
141		✗	if (apply_grain) {
142		✗	if (ctx->tmp_frame->buf[0])
143		✗	av_frame_unref(ctx->tmp_frame);
144		✗	err = ff_thread_get_buffer(avctx, ctx->tmp_frame, AV_GET_BUFFER_FLAG_REF);
145		✗	if (err < 0)
146		✗	goto fail;
147		✗	pic->output_surface = ff_vaapi_get_surface_id(ctx->tmp_frame);
148			} else {
149		✗	pic->output_surface = ff_vaapi_get_surface_id(s->cur_frame.f);
150			}
151
152		✗	memset(&pic_param, 0, sizeof(VADecPictureParameterBufferAV1));
153		✗	pic_param = (VADecPictureParameterBufferAV1) {
154		✗	.profile = seq->seq_profile,
155		✗	.order_hint_bits_minus_1 = seq->order_hint_bits_minus_1,
156			.bit_depth_idx = bit_depth_idx,
157		✗	.matrix_coefficients = seq->color_config.matrix_coefficients,
158		✗	.current_frame = pic->output_surface,
159		✗	.current_display_picture = ff_vaapi_get_surface_id(s->cur_frame.f),
160		✗	.frame_width_minus1 = frame_header->frame_width_minus_1,
161		✗	.frame_height_minus1 = frame_header->frame_height_minus_1,
162		✗	.primary_ref_frame = frame_header->primary_ref_frame,
163		✗	.order_hint = frame_header->order_hint,
164		✗	.tile_cols = frame_header->tile_cols,
165		✗	.tile_rows = frame_header->tile_rows,
166		✗	.context_update_tile_id = frame_header->context_update_tile_id,
167		✗	.superres_scale_denominator = frame_header->use_superres ?
168		✗	frame_header->coded_denom + AV1_SUPERRES_DENOM_MIN :
169			AV1_SUPERRES_NUM,
170		✗	.interp_filter = frame_header->interpolation_filter,
171		✗	.filter_level[0] = frame_header->loop_filter_level[0],
172		✗	.filter_level[1] = frame_header->loop_filter_level[1],
173		✗	.filter_level_u = frame_header->loop_filter_level[2],
174		✗	.filter_level_v = frame_header->loop_filter_level[3],
175		✗	.base_qindex = frame_header->base_q_idx,
176		✗	.y_dc_delta_q = frame_header->delta_q_y_dc,
177		✗	.u_dc_delta_q = frame_header->delta_q_u_dc,
178		✗	.u_ac_delta_q = frame_header->delta_q_u_ac,
179		✗	.v_dc_delta_q = frame_header->delta_q_v_dc,
180		✗	.v_ac_delta_q = frame_header->delta_q_v_ac,
181		✗	.cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
182		✗	.cdef_bits = frame_header->cdef_bits,
183			.seq_info_fields.fields = {
184		✗	.still_picture = seq->still_picture,
185		✗	.use_128x128_superblock = seq->use_128x128_superblock,
186		✗	.enable_filter_intra = seq->enable_filter_intra,
187		✗	.enable_intra_edge_filter = seq->enable_intra_edge_filter,
188		✗	.enable_interintra_compound = seq->enable_interintra_compound,
189		✗	.enable_masked_compound = seq->enable_masked_compound,
190		✗	.enable_dual_filter = seq->enable_dual_filter,
191		✗	.enable_order_hint = seq->enable_order_hint,
192		✗	.enable_jnt_comp = seq->enable_jnt_comp,
193		✗	.enable_cdef = seq->enable_cdef,
194		✗	.mono_chrome = seq->color_config.mono_chrome,
195		✗	.color_range = seq->color_config.color_range,
196		✗	.subsampling_x = seq->color_config.subsampling_x,
197		✗	.subsampling_y = seq->color_config.subsampling_y,
198		✗	.film_grain_params_present = seq->film_grain_params_present &&
199		✗	!(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN),
200			},
201			.seg_info.segment_info_fields.bits = {
202		✗	.enabled = frame_header->segmentation_enabled,
203		✗	.update_map = frame_header->segmentation_update_map,
204		✗	.temporal_update = frame_header->segmentation_temporal_update,
205		✗	.update_data = frame_header->segmentation_update_data,
206			},
207			.film_grain_info = {
208			.film_grain_info_fields.bits = {
209			.apply_grain = apply_grain,
210		✗	.chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma,
211		✗	.grain_scaling_minus_8 = film_grain->grain_scaling_minus_8,
212		✗	.ar_coeff_lag = film_grain->ar_coeff_lag,
213		✗	.ar_coeff_shift_minus_6 = film_grain->ar_coeff_shift_minus_6,
214		✗	.grain_scale_shift = film_grain->grain_scale_shift,
215		✗	.overlap_flag = film_grain->overlap_flag,
216		✗	.clip_to_restricted_range = film_grain->clip_to_restricted_range,
217			},
218		✗	.grain_seed = film_grain->grain_seed,
219		✗	.num_y_points = film_grain->num_y_points,
220		✗	.num_cb_points = film_grain->num_cb_points,
221		✗	.num_cr_points = film_grain->num_cr_points,
222		✗	.cb_mult = film_grain->cb_mult,
223		✗	.cb_luma_mult = film_grain->cb_luma_mult,
224		✗	.cb_offset = film_grain->cb_offset,
225		✗	.cr_mult = film_grain->cr_mult,
226		✗	.cr_luma_mult = film_grain->cr_luma_mult,
227		✗	.cr_offset = film_grain->cr_offset,
228			},
229			.pic_info_fields.bits = {
230		✗	.frame_type = frame_header->frame_type,
231		✗	.show_frame = frame_header->show_frame,
232		✗	.showable_frame = frame_header->showable_frame,
233		✗	.error_resilient_mode = frame_header->error_resilient_mode,
234		✗	.disable_cdf_update = frame_header->disable_cdf_update,
235		✗	.allow_screen_content_tools = frame_header->allow_screen_content_tools,
236		✗	.force_integer_mv = s->cur_frame.force_integer_mv,
237		✗	.allow_intrabc = frame_header->allow_intrabc,
238		✗	.use_superres = frame_header->use_superres,
239		✗	.allow_high_precision_mv = frame_header->allow_high_precision_mv,
240		✗	.is_motion_mode_switchable = frame_header->is_motion_mode_switchable,
241		✗	.use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
242		✗	.disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
243		✗	.uniform_tile_spacing_flag = frame_header->uniform_tile_spacing_flag,
244		✗	.allow_warped_motion = frame_header->allow_warped_motion,
245			},
246			.loop_filter_info_fields.bits = {
247		✗	.sharpness_level = frame_header->loop_filter_sharpness,
248		✗	.mode_ref_delta_enabled = frame_header->loop_filter_delta_enabled,
249		✗	.mode_ref_delta_update = frame_header->loop_filter_delta_update,
250			},
251			.mode_control_fields.bits = {
252		✗	.delta_q_present_flag = frame_header->delta_q_present,
253		✗	.log2_delta_q_res = frame_header->delta_q_res,
254		✗	.delta_lf_present_flag = frame_header->delta_lf_present,
255		✗	.log2_delta_lf_res = frame_header->delta_lf_res,
256		✗	.delta_lf_multi = frame_header->delta_lf_multi,
257		✗	.tx_mode = frame_header->tx_mode,
258		✗	.reference_select = frame_header->reference_select,
259		✗	.reduced_tx_set_used = frame_header->reduced_tx_set,
260		✗	.skip_mode_present = frame_header->skip_mode_present,
261			},
262			.loop_restoration_fields.bits = {
263		✗	.yframe_restoration_type = remap_lr_type[frame_header->lr_type[0]],
264		✗	.cbframe_restoration_type = remap_lr_type[frame_header->lr_type[1]],
265		✗	.crframe_restoration_type = remap_lr_type[frame_header->lr_type[2]],
266		✗	.lr_unit_shift = frame_header->lr_unit_shift,
267		✗	.lr_uv_shift = frame_header->lr_uv_shift,
268			},
269			.qmatrix_fields.bits = {
270		✗	.using_qmatrix = frame_header->using_qmatrix,
271		✗	.qm_y = frame_header->qm_y,
272		✗	.qm_u = frame_header->qm_u,
273		✗	.qm_v = frame_header->qm_v,
274			}
275			};
276
277		✗	for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
278		✗	if (pic_param.pic_info_fields.bits.frame_type == AV1_FRAME_KEY && frame_header->show_frame)
279		✗	pic_param.ref_frame_map[i] = VA_INVALID_ID;
280			else
281		✗	pic_param.ref_frame_map[i] = ctx->ref_tab[i].valid ?
282		✗	ff_vaapi_get_surface_id(ctx->ref_tab[i].frame) :
283		✗	vaapi_av1_surface_id(&s->ref[i]);
284			}
285		✗	for (int i = 0; i < AV1_REFS_PER_FRAME; i++) {
286		✗	pic_param.ref_frame_idx[i] = frame_header->ref_frame_idx[i];
287			}
288		✗	for (int i = 0; i < AV1_TOTAL_REFS_PER_FRAME; i++) {
289		✗	pic_param.ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
290			}
291		✗	for (int i = 0; i < 2; i++) {
292		✗	pic_param.mode_deltas[i] = frame_header->loop_filter_mode_deltas[i];
293			}
294		✗	for (int i = 0; i < (1 << frame_header->cdef_bits); i++) {
295		✗	pic_param.cdef_y_strengths[i] =
296		✗	(frame_header->cdef_y_pri_strength[i] << 2) +
297		✗	frame_header->cdef_y_sec_strength[i];
298		✗	pic_param.cdef_uv_strengths[i] =
299		✗	(frame_header->cdef_uv_pri_strength[i] << 2) +
300		✗	frame_header->cdef_uv_sec_strength[i];
301			}
302		✗	for (int i = 0; i < frame_header->tile_cols; i++) {
303		✗	pic_param.width_in_sbs_minus_1[i] =
304		✗	frame_header->width_in_sbs_minus_1[i];
305			}
306		✗	for (int i = 0; i < frame_header->tile_rows; i++) {
307		✗	pic_param.height_in_sbs_minus_1[i] =
308		✗	frame_header->height_in_sbs_minus_1[i];
309			}
310		✗	for (int i = AV1_REF_FRAME_LAST; i <= AV1_REF_FRAME_ALTREF; i++) {
311		✗	pic_param.wm[i - 1].invalid = s->cur_frame.gm_invalid[i];
312		✗	pic_param.wm[i - 1].wmtype = s->cur_frame.gm_type[i];
313		✗	for (int j = 0; j < 6; j++)
314		✗	pic_param.wm[i - 1].wmmat[j] = s->cur_frame.gm_params[i][j];
315			}
316		✗	for (int i = 0; i < AV1_MAX_SEGMENTS; i++) {
317		✗	for (int j = 0; j < AV1_SEG_LVL_MAX; j++) {
318		✗	pic_param.seg_info.feature_mask[i] |= (frame_header->feature_enabled[i][j] << j);
319		✗	if (segmentation_feature_signed[j])
320		✗	pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j],
321		✗	-segmentation_feature_max[j], segmentation_feature_max[j]);
322			else
323		✗	pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j],
324		✗	0, segmentation_feature_max[j]);
325			}
326			}
327		✗	if (apply_grain) {
328		✗	for (int i = 0; i < film_grain->num_y_points; i++) {
329		✗	pic_param.film_grain_info.point_y_value[i] =
330		✗	film_grain->point_y_value[i];
331		✗	pic_param.film_grain_info.point_y_scaling[i] =
332		✗	film_grain->point_y_scaling[i];
333			}
334		✗	for (int i = 0; i < film_grain->num_cb_points; i++) {
335		✗	pic_param.film_grain_info.point_cb_value[i] =
336		✗	film_grain->point_cb_value[i];
337		✗	pic_param.film_grain_info.point_cb_scaling[i] =
338		✗	film_grain->point_cb_scaling[i];
339			}
340		✗	for (int i = 0; i < film_grain->num_cr_points; i++) {
341		✗	pic_param.film_grain_info.point_cr_value[i] =
342		✗	film_grain->point_cr_value[i];
343		✗	pic_param.film_grain_info.point_cr_scaling[i] =
344		✗	film_grain->point_cr_scaling[i];
345			}
346		✗	for (int i = 0; i < 24; i++) {
347		✗	pic_param.film_grain_info.ar_coeffs_y[i] =
348		✗	film_grain->ar_coeffs_y_plus_128[i] - 128;
349			}
350		✗	for (int i = 0; i < 25; i++) {
351		✗	pic_param.film_grain_info.ar_coeffs_cb[i] =
352		✗	film_grain->ar_coeffs_cb_plus_128[i] - 128;
353		✗	pic_param.film_grain_info.ar_coeffs_cr[i] =
354		✗	film_grain->ar_coeffs_cr_plus_128[i] - 128;
355			}
356			}
357		✗	err = ff_vaapi_decode_make_param_buffer(avctx, pic,
358			VAPictureParameterBufferType,
359			&pic_param, sizeof(pic_param));
360		✗	if (err < 0)
361		✗	goto fail;
362
363		✗	return 0;
364
365		✗	fail:
366		✗	ff_vaapi_decode_cancel(avctx, pic);
367		✗	return err;
368			}
369
370		✗	static int vaapi_av1_end_frame(AVCodecContext *avctx)
371			{
372		✗	const AV1DecContext *s = avctx->priv_data;
373		✗	const AV1RawFrameHeader *header = s->raw_frame_header;
374		✗	const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
375		✗	VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
376		✗	VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
377
378		✗	int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
379			int ret;
380		✗	ret = ff_vaapi_decode_issue(avctx, pic);
381		✗	if (ret < 0)
382		✗	return ret;
383
384		✗	for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
385		✗	if (header->refresh_frame_flags & (1 << i)) {
386		✗	if (ctx->ref_tab[i].frame->buf[0])
387		✗	av_frame_unref(ctx->ref_tab[i].frame);
388
389		✗	if (apply_grain) {
390		✗	ret = av_frame_ref(ctx->ref_tab[i].frame, ctx->tmp_frame);
391		✗	if (ret < 0)
392		✗	return ret;
393		✗	ctx->ref_tab[i].valid = 1;
394			} else {
395		✗	ctx->ref_tab[i].valid = 0;
396			}
397			}
398			}
399
400		✗	return 0;
401			}
402
403		✗	static int vaapi_av1_decode_slice(AVCodecContext *avctx,
404			const uint8_t *buffer,
405			uint32_t size)
406			{
407		✗	const AV1DecContext *s = avctx->priv_data;
408		✗	VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
409		✗	VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
410			int err, nb_params;
411
412		✗	nb_params = s->tg_end - s->tg_start + 1;
413		✗	if (ctx->nb_slice_params < nb_params) {
414		✗	VASliceParameterBufferAV1 *tmp = av_realloc_array(ctx->slice_params,
415			nb_params,
416			sizeof(*ctx->slice_params));
417		✗	if (!tmp) {
418		✗	ctx->nb_slice_params = 0;
419		✗	err = AVERROR(ENOMEM);
420		✗	goto fail;
421			}
422		✗	ctx->slice_params = tmp;
423		✗	ctx->nb_slice_params = nb_params;
424			}
425
426		✗	for (int i = s->tg_start; i <= s->tg_end; i++) {
427		✗	ctx->slice_params[i - s->tg_start] = (VASliceParameterBufferAV1) {
428		✗	.slice_data_size = s->tile_group_info[i].tile_size,
429		✗	.slice_data_offset = s->tile_group_info[i].tile_offset,
430			.slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
431		✗	.tile_row = s->tile_group_info[i].tile_row,
432		✗	.tile_column = s->tile_group_info[i].tile_column,
433			};
434			}
435
436		✗	err = ff_vaapi_decode_make_slice_buffer(avctx, pic, ctx->slice_params, nb_params,
437			sizeof(VASliceParameterBufferAV1),
438			buffer,
439			size);
440		✗	if (err)
441		✗	goto fail;
442
443		✗	return 0;
444
445		✗	fail:
446		✗	ff_vaapi_decode_cancel(avctx, pic);
447		✗	return err;
448			}
449
450			const FFHWAccel ff_av1_vaapi_hwaccel = {
451			.p.name = "av1_vaapi",
452			.p.type = AVMEDIA_TYPE_VIDEO,
453			.p.id = AV_CODEC_ID_AV1,
454			.p.pix_fmt = AV_PIX_FMT_VAAPI,
455			.start_frame = vaapi_av1_start_frame,
456			.end_frame = vaapi_av1_end_frame,
457			.decode_slice = vaapi_av1_decode_slice,
458			.frame_priv_data_size = sizeof(VAAPIDecodePicture),
459			.init = vaapi_av1_decode_init,
460			.uninit = vaapi_av1_decode_uninit,
461			.frame_params = ff_vaapi_common_frame_params,
462			.priv_data_size = sizeof(VAAPIAV1DecContext),
463			.caps_internal = HWACCEL_CAP_ASYNC_SAFE,
464			};
465