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/pixdesc.h"
23 #include "hwconfig.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 } VAAPIAV1DecContext;
47
vaapi_av1_surface_id(AV1Frame * vf)48 static VASurfaceID vaapi_av1_surface_id(AV1Frame *vf)
49 {
50 if (vf)
51 return ff_vaapi_get_surface_id(vf->f);
52 else
53 return VA_INVALID_SURFACE;
54 }
55
vaapi_av1_get_bit_depth_idx(AVCodecContext * avctx)56 static int8_t vaapi_av1_get_bit_depth_idx(AVCodecContext *avctx)
57 {
58 AV1DecContext *s = avctx->priv_data;
59 const AV1RawSequenceHeader *seq = s->raw_seq;
60 int8_t bit_depth = 8;
61
62 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
63 bit_depth = seq->color_config.twelve_bit ? 12 : 10;
64 else if (seq->seq_profile <= 2)
65 bit_depth = seq->color_config.high_bitdepth ? 10 : 8;
66 else {
67 av_log(avctx, AV_LOG_ERROR,
68 "Couldn't get bit depth from profile:%d.\n", seq->seq_profile);
69 return -1;
70 }
71 return bit_depth == 8 ? 0 : bit_depth == 10 ? 1 : 2;
72 }
73
vaapi_av1_decode_init(AVCodecContext * avctx)74 static int vaapi_av1_decode_init(AVCodecContext *avctx)
75 {
76 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
77
78 ctx->tmp_frame = av_frame_alloc();
79 if (!ctx->tmp_frame) {
80 av_log(avctx, AV_LOG_ERROR,
81 "Failed to allocate frame.\n");
82 return AVERROR(ENOMEM);
83 }
84
85 for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++) {
86 ctx->ref_tab[i].frame = av_frame_alloc();
87 if (!ctx->ref_tab[i].frame) {
88 av_log(avctx, AV_LOG_ERROR,
89 "Failed to allocate reference table frame %d.\n", i);
90 return AVERROR(ENOMEM);
91 }
92 ctx->ref_tab[i].valid = 0;
93 }
94
95 return ff_vaapi_decode_init(avctx);
96 }
97
vaapi_av1_decode_uninit(AVCodecContext * avctx)98 static int vaapi_av1_decode_uninit(AVCodecContext *avctx)
99 {
100 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
101
102 if (ctx->tmp_frame->buf[0])
103 ff_thread_release_buffer(avctx, ctx->tmp_frame);
104 av_frame_free(&ctx->tmp_frame);
105
106 for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++) {
107 if (ctx->ref_tab[i].frame->buf[0])
108 ff_thread_release_buffer(avctx, ctx->ref_tab[i].frame);
109 av_frame_free(&ctx->ref_tab[i].frame);
110 }
111
112 return ff_vaapi_decode_uninit(avctx);
113 }
114
115
vaapi_av1_start_frame(AVCodecContext * avctx,av_unused const uint8_t * buffer,av_unused uint32_t size)116 static int vaapi_av1_start_frame(AVCodecContext *avctx,
117 av_unused const uint8_t *buffer,
118 av_unused uint32_t size)
119 {
120 AV1DecContext *s = avctx->priv_data;
121 const AV1RawSequenceHeader *seq = s->raw_seq;
122 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
123 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
124 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
125 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
126 VADecPictureParameterBufferAV1 pic_param;
127 int8_t bit_depth_idx;
128 int err = 0;
129 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
130 uint8_t remap_lr_type[4] = {AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ};
131 uint8_t segmentation_feature_signed[AV1_SEG_LVL_MAX] = {1, 1, 1, 1, 1, 0, 0, 0};
132 uint8_t segmentation_feature_max[AV1_SEG_LVL_MAX] = {255, AV1_MAX_LOOP_FILTER,
133 AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, 7 , 0 , 0 };
134
135 bit_depth_idx = vaapi_av1_get_bit_depth_idx(avctx);
136 if (bit_depth_idx < 0)
137 goto fail;
138
139 if (apply_grain) {
140 if (ctx->tmp_frame->buf[0])
141 ff_thread_release_buffer(avctx, ctx->tmp_frame);
142 err = ff_thread_get_buffer(avctx, ctx->tmp_frame, AV_GET_BUFFER_FLAG_REF);
143 if (err < 0)
144 goto fail;
145 pic->output_surface = ff_vaapi_get_surface_id(ctx->tmp_frame);
146 } else {
147 pic->output_surface = vaapi_av1_surface_id(&s->cur_frame);
148 }
149
150 memset(&pic_param, 0, sizeof(VADecPictureParameterBufferAV1));
151 pic_param = (VADecPictureParameterBufferAV1) {
152 .profile = seq->seq_profile,
153 .order_hint_bits_minus_1 = seq->order_hint_bits_minus_1,
154 .bit_depth_idx = bit_depth_idx,
155 .matrix_coefficients = seq->color_config.matrix_coefficients,
156 .current_frame = pic->output_surface,
157 .current_display_picture = vaapi_av1_surface_id(&s->cur_frame),
158 .frame_width_minus1 = frame_header->frame_width_minus_1,
159 .frame_height_minus1 = frame_header->frame_height_minus_1,
160 .primary_ref_frame = frame_header->primary_ref_frame,
161 .order_hint = frame_header->order_hint,
162 .tile_cols = frame_header->tile_cols,
163 .tile_rows = frame_header->tile_rows,
164 .context_update_tile_id = frame_header->context_update_tile_id,
165 .superres_scale_denominator = frame_header->use_superres ?
166 frame_header->coded_denom + AV1_SUPERRES_DENOM_MIN :
167 AV1_SUPERRES_NUM,
168 .interp_filter = frame_header->interpolation_filter,
169 .filter_level[0] = frame_header->loop_filter_level[0],
170 .filter_level[1] = frame_header->loop_filter_level[1],
171 .filter_level_u = frame_header->loop_filter_level[2],
172 .filter_level_v = frame_header->loop_filter_level[3],
173 .base_qindex = frame_header->base_q_idx,
174 .y_dc_delta_q = frame_header->delta_q_y_dc,
175 .u_dc_delta_q = frame_header->delta_q_u_dc,
176 .u_ac_delta_q = frame_header->delta_q_u_ac,
177 .v_dc_delta_q = frame_header->delta_q_v_dc,
178 .v_ac_delta_q = frame_header->delta_q_v_ac,
179 .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
180 .cdef_bits = frame_header->cdef_bits,
181 .seq_info_fields.fields = {
182 .still_picture = seq->still_picture,
183 .use_128x128_superblock = seq->use_128x128_superblock,
184 .enable_filter_intra = seq->enable_filter_intra,
185 .enable_intra_edge_filter = seq->enable_intra_edge_filter,
186 .enable_interintra_compound = seq->enable_interintra_compound,
187 .enable_masked_compound = seq->enable_masked_compound,
188 .enable_dual_filter = seq->enable_dual_filter,
189 .enable_order_hint = seq->enable_order_hint,
190 .enable_jnt_comp = seq->enable_jnt_comp,
191 .enable_cdef = seq->enable_cdef,
192 .mono_chrome = seq->color_config.mono_chrome,
193 .color_range = seq->color_config.color_range,
194 .subsampling_x = seq->color_config.subsampling_x,
195 .subsampling_y = seq->color_config.subsampling_y,
196 .chroma_sample_position = seq->color_config.chroma_sample_position,
197 .film_grain_params_present = seq->film_grain_params_present &&
198 !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN),
199 },
200 .seg_info.segment_info_fields.bits = {
201 .enabled = frame_header->segmentation_enabled,
202 .update_map = frame_header->segmentation_update_map,
203 .temporal_update = frame_header->segmentation_temporal_update,
204 .update_data = frame_header->segmentation_update_data,
205 },
206 .film_grain_info = {
207 .film_grain_info_fields.bits = {
208 .apply_grain = apply_grain,
209 .chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma,
210 .grain_scaling_minus_8 = film_grain->grain_scaling_minus_8,
211 .ar_coeff_lag = film_grain->ar_coeff_lag,
212 .ar_coeff_shift_minus_6 = film_grain->ar_coeff_shift_minus_6,
213 .grain_scale_shift = film_grain->grain_scale_shift,
214 .overlap_flag = film_grain->overlap_flag,
215 .clip_to_restricted_range = film_grain->clip_to_restricted_range,
216 },
217 .grain_seed = film_grain->grain_seed,
218 .num_y_points = film_grain->num_y_points,
219 .num_cb_points = film_grain->num_cb_points,
220 .num_cr_points = film_grain->num_cr_points,
221 .cb_mult = film_grain->cb_mult,
222 .cb_luma_mult = film_grain->cb_luma_mult,
223 .cb_offset = film_grain->cb_offset,
224 .cr_mult = film_grain->cr_mult,
225 .cr_luma_mult = film_grain->cr_luma_mult,
226 .cr_offset = film_grain->cr_offset,
227 },
228 .pic_info_fields.bits = {
229 .frame_type = frame_header->frame_type,
230 .show_frame = frame_header->show_frame,
231 .showable_frame = frame_header->showable_frame,
232 .error_resilient_mode = frame_header->error_resilient_mode,
233 .disable_cdf_update = frame_header->disable_cdf_update,
234 .allow_screen_content_tools = frame_header->allow_screen_content_tools,
235 .force_integer_mv = frame_header->force_integer_mv,
236 .allow_intrabc = frame_header->allow_intrabc,
237 .use_superres = frame_header->use_superres,
238 .allow_high_precision_mv = frame_header->allow_high_precision_mv,
239 .is_motion_mode_switchable = frame_header->is_motion_mode_switchable,
240 .use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
241 .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
242 .uniform_tile_spacing_flag = frame_header->uniform_tile_spacing_flag,
243 .allow_warped_motion = frame_header->allow_warped_motion,
244 },
245 .loop_filter_info_fields.bits = {
246 .sharpness_level = frame_header->loop_filter_sharpness,
247 .mode_ref_delta_enabled = frame_header->loop_filter_delta_enabled,
248 .mode_ref_delta_update = frame_header->loop_filter_delta_update,
249 },
250 .mode_control_fields.bits = {
251 .delta_q_present_flag = frame_header->delta_q_present,
252 .log2_delta_q_res = frame_header->delta_q_res,
253 .delta_lf_present_flag = frame_header->delta_lf_present,
254 .log2_delta_lf_res = frame_header->delta_lf_res,
255 .delta_lf_multi = frame_header->delta_lf_multi,
256 .tx_mode = frame_header->tx_mode,
257 .reference_select = frame_header->reference_select,
258 .reduced_tx_set_used = frame_header->reduced_tx_set,
259 .skip_mode_present = frame_header->skip_mode_present,
260 },
261 .loop_restoration_fields.bits = {
262 .yframe_restoration_type = remap_lr_type[frame_header->lr_type[0]],
263 .cbframe_restoration_type = remap_lr_type[frame_header->lr_type[1]],
264 .crframe_restoration_type = remap_lr_type[frame_header->lr_type[2]],
265 .lr_unit_shift = frame_header->lr_unit_shift,
266 .lr_uv_shift = frame_header->lr_uv_shift,
267 },
268 .qmatrix_fields.bits = {
269 .using_qmatrix = frame_header->using_qmatrix,
270 .qm_y = frame_header->qm_y,
271 .qm_u = frame_header->qm_u,
272 .qm_v = frame_header->qm_v,
273 }
274 };
275
276 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
277 if (pic_param.pic_info_fields.bits.frame_type == AV1_FRAME_KEY)
278 pic_param.ref_frame_map[i] = VA_INVALID_ID;
279 else
280 pic_param.ref_frame_map[i] = ctx->ref_tab[i].valid ?
281 ff_vaapi_get_surface_id(ctx->ref_tab[i].frame) :
282 vaapi_av1_surface_id(&s->ref[i]);
283 }
284 for (int i = 0; i < AV1_REFS_PER_FRAME; i++) {
285 pic_param.ref_frame_idx[i] = frame_header->ref_frame_idx[i];
286 }
287 for (int i = 0; i < AV1_TOTAL_REFS_PER_FRAME; i++) {
288 pic_param.ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
289 }
290 for (int i = 0; i < 2; i++) {
291 pic_param.mode_deltas[i] = frame_header->loop_filter_mode_deltas[i];
292 }
293 for (int i = 0; i < (1 << frame_header->cdef_bits); i++) {
294 pic_param.cdef_y_strengths[i] =
295 (frame_header->cdef_y_pri_strength[i] << 2) +
296 frame_header->cdef_y_sec_strength[i];
297 pic_param.cdef_uv_strengths[i] =
298 (frame_header->cdef_uv_pri_strength[i] << 2) +
299 frame_header->cdef_uv_sec_strength[i];
300 }
301 for (int i = 0; i < frame_header->tile_cols; i++) {
302 pic_param.width_in_sbs_minus_1[i] =
303 frame_header->width_in_sbs_minus_1[i];
304 }
305 for (int i = 0; i < frame_header->tile_rows; i++) {
306 pic_param.height_in_sbs_minus_1[i] =
307 frame_header->height_in_sbs_minus_1[i];
308 }
309 for (int i = AV1_REF_FRAME_LAST; i <= AV1_REF_FRAME_ALTREF; i++) {
310 pic_param.wm[i - 1].invalid = s->cur_frame.gm_invalid[i];
311 pic_param.wm[i - 1].wmtype = s->cur_frame.gm_type[i];
312 for (int j = 0; j < 6; j++)
313 pic_param.wm[i - 1].wmmat[j] = s->cur_frame.gm_params[i][j];
314 }
315 for (int i = 0; i < AV1_MAX_SEGMENTS; i++) {
316 for (int j = 0; j < AV1_SEG_LVL_MAX; j++) {
317 pic_param.seg_info.feature_mask[i] |= (frame_header->feature_enabled[i][j] << j);
318 if (segmentation_feature_signed[j])
319 pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j],
320 -segmentation_feature_max[j], segmentation_feature_max[j]);
321 else
322 pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j],
323 0, segmentation_feature_max[j]);
324 }
325 }
326 if (apply_grain) {
327 for (int i = 0; i < film_grain->num_y_points; i++) {
328 pic_param.film_grain_info.point_y_value[i] =
329 film_grain->point_y_value[i];
330 pic_param.film_grain_info.point_y_scaling[i] =
331 film_grain->point_y_scaling[i];
332 }
333 for (int i = 0; i < film_grain->num_cb_points; i++) {
334 pic_param.film_grain_info.point_cb_value[i] =
335 film_grain->point_cb_value[i];
336 pic_param.film_grain_info.point_cb_scaling[i] =
337 film_grain->point_cb_scaling[i];
338 }
339 for (int i = 0; i < film_grain->num_cr_points; i++) {
340 pic_param.film_grain_info.point_cr_value[i] =
341 film_grain->point_cr_value[i];
342 pic_param.film_grain_info.point_cr_scaling[i] =
343 film_grain->point_cr_scaling[i];
344 }
345 for (int i = 0; i < 24; i++) {
346 pic_param.film_grain_info.ar_coeffs_y[i] =
347 film_grain->ar_coeffs_y_plus_128[i] - 128;
348 }
349 for (int i = 0; i < 25; i++) {
350 pic_param.film_grain_info.ar_coeffs_cb[i] =
351 film_grain->ar_coeffs_cb_plus_128[i] - 128;
352 pic_param.film_grain_info.ar_coeffs_cr[i] =
353 film_grain->ar_coeffs_cr_plus_128[i] - 128;
354 }
355 }
356 err = ff_vaapi_decode_make_param_buffer(avctx, pic,
357 VAPictureParameterBufferType,
358 &pic_param, sizeof(pic_param));
359 if (err < 0)
360 goto fail;
361
362 return 0;
363
364 fail:
365 ff_vaapi_decode_cancel(avctx, pic);
366 return err;
367 }
368
vaapi_av1_end_frame(AVCodecContext * avctx)369 static int vaapi_av1_end_frame(AVCodecContext *avctx)
370 {
371 const AV1DecContext *s = avctx->priv_data;
372 const AV1RawFrameHeader *header = s->raw_frame_header;
373 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
374 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
375 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data;
376
377 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
378 int ret;
379 ret = ff_vaapi_decode_issue(avctx, pic);
380 if (ret < 0)
381 return ret;
382
383 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
384 if (header->refresh_frame_flags & (1 << i)) {
385 if (ctx->ref_tab[i].frame->buf[0])
386 ff_thread_release_buffer(avctx, ctx->ref_tab[i].frame);
387
388 if (apply_grain) {
389 ret = av_frame_ref(ctx->ref_tab[i].frame, ctx->tmp_frame);
390 if (ret < 0)
391 return ret;
392 ctx->ref_tab[i].valid = 1;
393 } else {
394 ctx->ref_tab[i].valid = 0;
395 }
396 }
397 }
398
399 return 0;
400 }
401
vaapi_av1_decode_slice(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)402 static int vaapi_av1_decode_slice(AVCodecContext *avctx,
403 const uint8_t *buffer,
404 uint32_t size)
405 {
406 const AV1DecContext *s = avctx->priv_data;
407 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
408 VASliceParameterBufferAV1 slice_param;
409 int err = 0;
410
411 for (int i = s->tg_start; i <= s->tg_end; i++) {
412 memset(&slice_param, 0, sizeof(VASliceParameterBufferAV1));
413
414 slice_param = (VASliceParameterBufferAV1) {
415 .slice_data_size = s->tile_group_info[i].tile_size,
416 .slice_data_offset = s->tile_group_info[i].tile_offset,
417 .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
418 .tile_row = s->tile_group_info[i].tile_row,
419 .tile_column = s->tile_group_info[i].tile_column,
420 .tg_start = s->tg_start,
421 .tg_end = s->tg_end,
422 };
423
424 err = ff_vaapi_decode_make_slice_buffer(avctx, pic, &slice_param,
425 sizeof(VASliceParameterBufferAV1),
426 buffer,
427 size);
428 if (err) {
429 ff_vaapi_decode_cancel(avctx, pic);
430 return err;
431 }
432 }
433
434 return 0;
435 }
436
437 const AVHWAccel ff_av1_vaapi_hwaccel = {
438 .name = "av1_vaapi",
439 .type = AVMEDIA_TYPE_VIDEO,
440 .id = AV_CODEC_ID_AV1,
441 .pix_fmt = AV_PIX_FMT_VAAPI,
442 .start_frame = vaapi_av1_start_frame,
443 .end_frame = vaapi_av1_end_frame,
444 .decode_slice = vaapi_av1_decode_slice,
445 .frame_priv_data_size = sizeof(VAAPIDecodePicture),
446 .init = vaapi_av1_decode_init,
447 .uninit = vaapi_av1_decode_uninit,
448 .frame_params = ff_vaapi_common_frame_params,
449 .priv_data_size = sizeof(VAAPIAV1DecContext),
450 .caps_internal = HWACCEL_CAP_ASYNC_SAFE,
451 };
452