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 "av1dec.h"
25
vaapi_av1_surface_id(AV1Frame * vf)26 static VASurfaceID vaapi_av1_surface_id(AV1Frame *vf)
27 {
28 if (vf)
29 return ff_vaapi_get_surface_id(vf->tf.f);
30 else
31 return VA_INVALID_SURFACE;
32 }
33
vaapi_av1_get_bit_depth_idx(AVCodecContext * avctx)34 static int8_t vaapi_av1_get_bit_depth_idx(AVCodecContext *avctx)
35 {
36 AV1DecContext *s = avctx->priv_data;
37 const AV1RawSequenceHeader *seq = s->raw_seq;
38 int8_t bit_depth = 8;
39
40 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
41 bit_depth = seq->color_config.twelve_bit ? 12 : 10;
42 else if (seq->seq_profile <= 2)
43 bit_depth = seq->color_config.high_bitdepth ? 10 : 8;
44 else {
45 av_log(avctx, AV_LOG_ERROR,
46 "Couldn't get bit depth from profile:%d.\n", seq->seq_profile);
47 return -1;
48 }
49 return bit_depth == 8 ? 0 : bit_depth == 10 ? 1 : 2;
50 }
51
vaapi_av1_start_frame(AVCodecContext * avctx,av_unused const uint8_t * buffer,av_unused uint32_t size)52 static int vaapi_av1_start_frame(AVCodecContext *avctx,
53 av_unused const uint8_t *buffer,
54 av_unused uint32_t size)
55 {
56 AV1DecContext *s = avctx->priv_data;
57 const AV1RawSequenceHeader *seq = s->raw_seq;
58 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
59 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
60 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
61 VADecPictureParameterBufferAV1 pic_param;
62 int8_t bit_depth_idx;
63 int err = 0;
64 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
65 uint8_t remap_lr_type[4] = {AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ};
66
67 pic->output_surface = vaapi_av1_surface_id(&s->cur_frame);
68
69 bit_depth_idx = vaapi_av1_get_bit_depth_idx(avctx);
70 if (bit_depth_idx < 0)
71 goto fail;
72
73 memset(&pic_param, 0, sizeof(VADecPictureParameterBufferAV1));
74 pic_param = (VADecPictureParameterBufferAV1) {
75 .profile = seq->seq_profile,
76 .order_hint_bits_minus_1 = seq->order_hint_bits_minus_1,
77 .bit_depth_idx = bit_depth_idx,
78 .current_frame = pic->output_surface,
79 .current_display_picture = pic->output_surface,
80 .frame_width_minus1 = frame_header->frame_width_minus_1,
81 .frame_height_minus1 = frame_header->frame_height_minus_1,
82 .primary_ref_frame = frame_header->primary_ref_frame,
83 .order_hint = frame_header->order_hint,
84 .tile_cols = frame_header->tile_cols,
85 .tile_rows = frame_header->tile_rows,
86 .context_update_tile_id = frame_header->context_update_tile_id,
87 .interp_filter = frame_header->interpolation_filter,
88 .filter_level[0] = frame_header->loop_filter_level[0],
89 .filter_level[1] = frame_header->loop_filter_level[1],
90 .filter_level_u = frame_header->loop_filter_level[2],
91 .filter_level_v = frame_header->loop_filter_level[3],
92 .base_qindex = frame_header->base_q_idx,
93 .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
94 .cdef_bits = frame_header->cdef_bits,
95 .seq_info_fields.fields = {
96 .still_picture = seq->still_picture,
97 .use_128x128_superblock = seq->use_128x128_superblock,
98 .enable_filter_intra = seq->enable_filter_intra,
99 .enable_intra_edge_filter = seq->enable_intra_edge_filter,
100 .enable_interintra_compound = seq->enable_interintra_compound,
101 .enable_masked_compound = seq->enable_masked_compound,
102 .enable_dual_filter = seq->enable_dual_filter,
103 .enable_order_hint = seq->enable_order_hint,
104 .enable_jnt_comp = seq->enable_jnt_comp,
105 .enable_cdef = seq->enable_cdef,
106 .mono_chrome = seq->color_config.mono_chrome,
107 .color_range = seq->color_config.color_range,
108 .subsampling_x = seq->color_config.subsampling_x,
109 .subsampling_y = seq->color_config.subsampling_y,
110 .chroma_sample_position = seq->color_config.chroma_sample_position,
111 .film_grain_params_present = seq->film_grain_params_present &&
112 !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN),
113 },
114 .seg_info.segment_info_fields.bits = {
115 .enabled = frame_header->segmentation_enabled,
116 .update_map = frame_header->segmentation_update_map,
117 .temporal_update = frame_header->segmentation_temporal_update,
118 .update_data = frame_header->segmentation_update_data,
119 },
120 .film_grain_info = {
121 .film_grain_info_fields.bits = {
122 .apply_grain = apply_grain,
123 .chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma,
124 .grain_scaling_minus_8 = film_grain->grain_scaling_minus_8,
125 .ar_coeff_lag = film_grain->ar_coeff_lag,
126 .ar_coeff_shift_minus_6 = film_grain->ar_coeff_shift_minus_6,
127 .grain_scale_shift = film_grain->grain_scale_shift,
128 .overlap_flag = film_grain->overlap_flag,
129 .clip_to_restricted_range = film_grain->clip_to_restricted_range,
130 },
131 .grain_seed = film_grain->grain_seed,
132 .num_y_points = film_grain->num_y_points,
133 .num_cb_points = film_grain->num_cb_points,
134 .num_cr_points = film_grain->num_cr_points,
135 .cb_mult = film_grain->cb_mult,
136 .cb_luma_mult = film_grain->cb_luma_mult,
137 .cb_offset = film_grain->cb_offset,
138 .cr_mult = film_grain->cr_mult,
139 .cr_luma_mult = film_grain->cr_luma_mult,
140 .cr_offset = film_grain->cr_offset,
141 },
142 .pic_info_fields.bits = {
143 .frame_type = frame_header->frame_type,
144 .show_frame = frame_header->show_frame,
145 .showable_frame = frame_header->showable_frame,
146 .error_resilient_mode = frame_header->error_resilient_mode,
147 .disable_cdf_update = frame_header->disable_cdf_update,
148 .allow_screen_content_tools = frame_header->allow_screen_content_tools,
149 .force_integer_mv = frame_header->force_integer_mv,
150 .allow_intrabc = frame_header->allow_intrabc,
151 .use_superres = frame_header->use_superres,
152 .allow_high_precision_mv = frame_header->allow_high_precision_mv,
153 .is_motion_mode_switchable = frame_header->is_motion_mode_switchable,
154 .use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
155 .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
156 .uniform_tile_spacing_flag = frame_header->uniform_tile_spacing_flag,
157 .allow_warped_motion = frame_header->allow_warped_motion,
158 },
159 .loop_filter_info_fields.bits = {
160 .sharpness_level = frame_header->loop_filter_sharpness,
161 .mode_ref_delta_enabled = frame_header->loop_filter_delta_enabled,
162 .mode_ref_delta_update = frame_header->loop_filter_delta_update,
163 },
164 .mode_control_fields.bits = {
165 .delta_q_present_flag = frame_header->delta_q_present,
166 .log2_delta_q_res = frame_header->delta_q_res,
167 .tx_mode = frame_header->tx_mode,
168 .reference_select = frame_header->reference_select,
169 .reduced_tx_set_used = frame_header->reduced_tx_set,
170 .skip_mode_present = frame_header->skip_mode_present,
171 },
172 .loop_restoration_fields.bits = {
173 .yframe_restoration_type = remap_lr_type[frame_header->lr_type[0]],
174 .cbframe_restoration_type = remap_lr_type[frame_header->lr_type[1]],
175 .crframe_restoration_type = remap_lr_type[frame_header->lr_type[2]],
176 .lr_unit_shift = frame_header->lr_unit_shift,
177 .lr_uv_shift = frame_header->lr_uv_shift,
178 },
179 .qmatrix_fields.bits = {
180 .using_qmatrix = frame_header->using_qmatrix,
181 }
182 };
183
184 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) {
185 if (pic_param.pic_info_fields.bits.frame_type == AV1_FRAME_KEY)
186 pic_param.ref_frame_map[i] = VA_INVALID_ID;
187 else
188 pic_param.ref_frame_map[i] = vaapi_av1_surface_id(&s->ref[i]);
189 }
190 for (int i = 0; i < AV1_REFS_PER_FRAME; i++) {
191 pic_param.ref_frame_idx[i] = frame_header->ref_frame_idx[i];
192 }
193 for (int i = 0; i < AV1_TOTAL_REFS_PER_FRAME; i++) {
194 pic_param.ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
195 }
196 for (int i = 0; i < 2; i++) {
197 pic_param.mode_deltas[i] = frame_header->loop_filter_mode_deltas[i];
198 }
199 for (int i = 0; i < (1 << frame_header->cdef_bits); i++) {
200 pic_param.cdef_y_strengths[i] =
201 (frame_header->cdef_y_pri_strength[i] << 2) +
202 frame_header->cdef_y_sec_strength[i];
203 pic_param.cdef_uv_strengths[i] =
204 (frame_header->cdef_uv_pri_strength[i] << 2) +
205 frame_header->cdef_uv_sec_strength[i];
206 }
207 for (int i = 0; i < frame_header->tile_cols; i++) {
208 pic_param.width_in_sbs_minus_1[i] =
209 frame_header->width_in_sbs_minus_1[i];
210 }
211 for (int i = 0; i < frame_header->tile_rows; i++) {
212 pic_param.height_in_sbs_minus_1[i] =
213 frame_header->height_in_sbs_minus_1[i];
214 }
215 for (int i = AV1_REF_FRAME_LAST; i <= AV1_REF_FRAME_ALTREF; i++) {
216 pic_param.wm[i - 1].wmtype = s->cur_frame.gm_type[i];
217 for (int j = 0; j < 6; j++)
218 pic_param.wm[i - 1].wmmat[j] = s->cur_frame.gm_params[i][j];
219 }
220 if (apply_grain) {
221 for (int i = 0; i < film_grain->num_y_points; i++) {
222 pic_param.film_grain_info.point_y_value[i] =
223 film_grain->point_y_value[i];
224 pic_param.film_grain_info.point_y_scaling[i] =
225 film_grain->point_y_scaling[i];
226 }
227 for (int i = 0; i < film_grain->num_cb_points; i++) {
228 pic_param.film_grain_info.point_cb_value[i] =
229 film_grain->point_cb_value[i];
230 pic_param.film_grain_info.point_cb_scaling[i] =
231 film_grain->point_cb_scaling[i];
232 }
233 for (int i = 0; i < film_grain->num_cr_points; i++) {
234 pic_param.film_grain_info.point_cr_value[i] =
235 film_grain->point_cr_value[i];
236 pic_param.film_grain_info.point_cr_scaling[i] =
237 film_grain->point_cr_scaling[i];
238 }
239 for (int i = 0; i < 24; i++) {
240 pic_param.film_grain_info.ar_coeffs_y[i] =
241 film_grain->ar_coeffs_y_plus_128[i] - 128;
242 }
243 for (int i = 0; i < 25; i++) {
244 pic_param.film_grain_info.ar_coeffs_cb[i] =
245 film_grain->ar_coeffs_cb_plus_128[i] - 128;
246 pic_param.film_grain_info.ar_coeffs_cr[i] =
247 film_grain->ar_coeffs_cr_plus_128[i] - 128;
248 }
249 }
250 err = ff_vaapi_decode_make_param_buffer(avctx, pic,
251 VAPictureParameterBufferType,
252 &pic_param, sizeof(pic_param));
253 if (err < 0)
254 goto fail;
255
256 return 0;
257
258 fail:
259 ff_vaapi_decode_cancel(avctx, pic);
260 return err;
261 }
262
vaapi_av1_end_frame(AVCodecContext * avctx)263 static int vaapi_av1_end_frame(AVCodecContext *avctx)
264 {
265 const AV1DecContext *s = avctx->priv_data;
266 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
267 return ff_vaapi_decode_issue(avctx, pic);
268 }
269
vaapi_av1_decode_slice(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)270 static int vaapi_av1_decode_slice(AVCodecContext *avctx,
271 const uint8_t *buffer,
272 uint32_t size)
273 {
274 const AV1DecContext *s = avctx->priv_data;
275 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private;
276 VASliceParameterBufferAV1 slice_param;
277 int err = 0;
278
279 for (int i = s->tg_start; i <= s->tg_end; i++) {
280 memset(&slice_param, 0, sizeof(VASliceParameterBufferAV1));
281
282 slice_param = (VASliceParameterBufferAV1) {
283 .slice_data_size = s->tile_group_info[i].tile_size,
284 .slice_data_offset = s->tile_group_info[i].tile_offset,
285 .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
286 .tile_row = s->tile_group_info[i].tile_row,
287 .tile_column = s->tile_group_info[i].tile_column,
288 .tg_start = s->tg_start,
289 .tg_end = s->tg_end,
290 };
291
292 err = ff_vaapi_decode_make_slice_buffer(avctx, pic, &slice_param,
293 sizeof(VASliceParameterBufferAV1),
294 buffer,
295 size);
296 if (err) {
297 ff_vaapi_decode_cancel(avctx, pic);
298 return err;
299 }
300 }
301
302 return 0;
303 }
304
305 const AVHWAccel ff_av1_vaapi_hwaccel = {
306 .name = "av1_vaapi",
307 .type = AVMEDIA_TYPE_VIDEO,
308 .id = AV_CODEC_ID_AV1,
309 .pix_fmt = AV_PIX_FMT_VAAPI,
310 .start_frame = vaapi_av1_start_frame,
311 .end_frame = vaapi_av1_end_frame,
312 .decode_slice = vaapi_av1_decode_slice,
313 .frame_priv_data_size = sizeof(VAAPIDecodePicture),
314 .init = ff_vaapi_decode_init,
315 .uninit = ff_vaapi_decode_uninit,
316 .frame_params = ff_vaapi_common_frame_params,
317 .priv_data_size = sizeof(VAAPIDecodeContext),
318 .caps_internal = HWACCEL_CAP_ASYNC_SAFE,
319 };
320