FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavcodec/vaapi_av1.c
Date: 2022-01-18 10:05:52
Exec Total Coverage
Lines: 0 273 0.0%
Branches: 0 108 0.0%

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