1 /*
2 * AV1 HW decode acceleration through VDPAU
3 *
4 * Copyright (c) 2022 Manoj Gupta Bonda
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software Foundation,
20 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #include <vdpau/vdpau.h>
24 #include "libavutil/pixdesc.h"
25 #include "avcodec.h"
26 #include "internal.h"
27 #include "av1dec.h"
28 #include "hwconfig.h"
29 #include "vdpau.h"
30 #include "vdpau_internal.h"
31
get_bit_depth_from_seq(const AV1RawSequenceHeader * seq)32 static int get_bit_depth_from_seq(const AV1RawSequenceHeader *seq)
33 {
34 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth) {
35 return seq->color_config.twelve_bit ? 12 : 10;
36 } else if (seq->seq_profile <= 2 && seq->color_config.high_bitdepth) {
37 return 10;
38 } else {
39 return 8;
40 }
41 }
42
vdpau_av1_start_frame(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)43 static int vdpau_av1_start_frame(AVCodecContext *avctx,
44 const uint8_t *buffer, uint32_t size)
45 {
46 AV1DecContext *s = avctx->priv_data;
47 const AV1RawSequenceHeader *seq = s->raw_seq;
48 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
49 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
50
51 struct vdpau_picture_context *pic_ctx = s->cur_frame.hwaccel_picture_private;
52 int i,j;
53
54 unsigned char remap_lr_type[4] = { AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ };
55
56
57 VdpPictureInfoAV1 *info = &pic_ctx->info.av1;
58 const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
59 if (!pixdesc) {
60 return AV_PIX_FMT_NONE;
61 }
62
63 info->width = avctx->width;
64 info->height = avctx->height;
65
66
67 info->frame_offset = frame_header->order_hint;
68
69 /* Sequence Header */
70 info->profile = seq->seq_profile;
71 info->use_128x128_superblock = seq->use_128x128_superblock;
72 info->subsampling_x = seq->color_config.subsampling_x;
73 info->subsampling_y = seq->color_config.subsampling_y;
74 info->mono_chrome = seq->color_config.mono_chrome;
75 info->bit_depth_minus8 = get_bit_depth_from_seq(seq) - 8;
76 info->enable_filter_intra = seq->enable_filter_intra;
77 info->enable_intra_edge_filter = seq->enable_intra_edge_filter;
78 info->enable_interintra_compound = seq->enable_interintra_compound;
79 info->enable_masked_compound = seq->enable_masked_compound;
80 info->enable_dual_filter = seq->enable_dual_filter;
81 info->enable_order_hint = seq->enable_order_hint;
82 info->order_hint_bits_minus1 = seq->order_hint_bits_minus_1;
83 info->enable_jnt_comp = seq->enable_jnt_comp;
84 info->enable_superres = seq->enable_superres;
85 info->enable_cdef = seq->enable_cdef;
86 info->enable_restoration = seq->enable_restoration;
87 info->enable_fgs = seq->film_grain_params_present;
88
89 /* Frame Header */
90 info->frame_type = frame_header->frame_type;
91 info->show_frame = frame_header->show_frame;
92 info->disable_cdf_update = frame_header->disable_cdf_update;
93 info->allow_screen_content_tools = frame_header->allow_screen_content_tools;
94 info->force_integer_mv = frame_header->force_integer_mv ||
95 frame_header->frame_type == AV1_FRAME_INTRA_ONLY ||
96 frame_header->frame_type == AV1_FRAME_KEY;
97 info->coded_denom = frame_header->coded_denom;
98 info->allow_intrabc = frame_header->allow_intrabc;
99 info->allow_high_precision_mv = frame_header->allow_high_precision_mv;
100 info->interp_filter = frame_header->interpolation_filter;
101 info->switchable_motion_mode = frame_header->is_motion_mode_switchable;
102 info->use_ref_frame_mvs = frame_header->use_ref_frame_mvs;
103 info->disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf;
104 info->delta_q_present = frame_header->delta_q_present;
105 info->delta_q_res = frame_header->delta_q_res;
106 info->using_qmatrix = frame_header->using_qmatrix;
107 info->coded_lossless = s->cur_frame.coded_lossless;
108 info->use_superres = frame_header->use_superres;
109 info->tx_mode = frame_header->tx_mode;
110 info->reference_mode = frame_header->reference_select;
111 info->allow_warped_motion = frame_header->allow_warped_motion;
112 info->reduced_tx_set = frame_header->reduced_tx_set;
113 info->skip_mode = frame_header->skip_mode_present;
114
115 /* Tiling Info */
116 info->num_tile_cols = frame_header->tile_cols;
117 info->num_tile_rows = frame_header->tile_rows;
118 info->context_update_tile_id = frame_header->context_update_tile_id;
119
120 /* CDEF */
121 info->cdef_damping_minus_3 = frame_header->cdef_damping_minus_3;
122 info->cdef_bits = frame_header->cdef_bits;
123
124 /* SkipModeFrames */
125 info->SkipModeFrame0 = frame_header->skip_mode_present ?
126 s->cur_frame.skip_mode_frame_idx[0] : 0;
127 info->SkipModeFrame1 = frame_header->skip_mode_present ?
128 s->cur_frame.skip_mode_frame_idx[1] : 0;
129
130 /* QP Information */
131 info->base_qindex = frame_header->base_q_idx;
132 info->qp_y_dc_delta_q = frame_header->delta_q_y_dc;
133 info->qp_u_dc_delta_q = frame_header->delta_q_u_dc;
134 info->qp_v_dc_delta_q = frame_header->delta_q_v_dc;
135 info->qp_u_ac_delta_q = frame_header->delta_q_u_ac;
136 info->qp_v_ac_delta_q = frame_header->delta_q_v_ac;
137 info->qm_y = frame_header->qm_y;
138 info->qm_u = frame_header->qm_u;
139 info->qm_v = frame_header->qm_v;
140
141 /* Segmentation */
142 info->segmentation_enabled = frame_header->segmentation_enabled;
143 info->segmentation_update_map = frame_header->segmentation_update_map;
144 info->segmentation_update_data = frame_header->segmentation_update_data;
145 info->segmentation_temporal_update = frame_header->segmentation_temporal_update;
146
147 /* Loopfilter */
148 info->loop_filter_level[0] = frame_header->loop_filter_level[0];
149 info->loop_filter_level[1] = frame_header->loop_filter_level[1];
150 info->loop_filter_level_u = frame_header->loop_filter_level[2];
151 info->loop_filter_level_v = frame_header->loop_filter_level[3];
152 info->loop_filter_sharpness = frame_header->loop_filter_sharpness;
153 info->loop_filter_delta_enabled = frame_header->loop_filter_delta_enabled;
154 info->loop_filter_delta_update = frame_header->loop_filter_delta_update;
155 info->loop_filter_mode_deltas[0] = frame_header->loop_filter_mode_deltas[0];
156 info->loop_filter_mode_deltas[1] = frame_header->loop_filter_mode_deltas[1];
157 info->delta_lf_present = frame_header->delta_lf_present;
158 info->delta_lf_res = frame_header->delta_lf_res;
159 info->delta_lf_multi = frame_header->delta_lf_multi;
160
161 /* Restoration */
162 info->lr_type[0] = remap_lr_type[frame_header->lr_type[0]];
163 info->lr_type[1] = remap_lr_type[frame_header->lr_type[1]];
164 info->lr_type[2] = remap_lr_type[frame_header->lr_type[2]];
165 info->lr_unit_size[0] = 1 + frame_header->lr_unit_shift;
166 info->lr_unit_size[1] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift;
167 info->lr_unit_size[2] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift;
168
169 /* Reference Frames */
170 info->temporal_layer_id = s->cur_frame.temporal_id;
171 info->spatial_layer_id = s->cur_frame.spatial_id;
172
173 /* Film Grain Params */
174 info->apply_grain = film_grain->apply_grain;
175 info->overlap_flag = film_grain->overlap_flag;
176 info->scaling_shift_minus8 = film_grain->grain_scaling_minus_8;
177 info->chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma;
178 info->ar_coeff_lag = film_grain->ar_coeff_lag;
179 info->ar_coeff_shift_minus6 = film_grain->ar_coeff_shift_minus_6;
180 info->grain_scale_shift = film_grain->grain_scale_shift;
181 info->clip_to_restricted_range = film_grain->clip_to_restricted_range;
182 info->num_y_points = film_grain->num_y_points;
183 info->num_cb_points = film_grain->num_cb_points;
184 info->num_cr_points = film_grain->num_cr_points;
185 info->random_seed = film_grain->grain_seed;
186 info->cb_mult = film_grain->cb_mult;
187 info->cb_luma_mult = film_grain->cb_luma_mult;
188 info->cb_offset = film_grain->cb_offset;
189 info->cr_mult = film_grain->cr_mult;
190 info->cr_luma_mult = film_grain->cr_luma_mult;
191 info->cr_offset = film_grain->cr_offset;
192
193 /* Tiling Info */
194 for (i = 0; i < frame_header->tile_cols; ++i) {
195 info->tile_widths[i] = frame_header->width_in_sbs_minus_1[i] + 1;
196 }
197 for (i = 0; i < frame_header->tile_rows; ++i) {
198 info->tile_heights[i] = frame_header->height_in_sbs_minus_1[i] + 1;
199 }
200
201 /* CDEF */
202 for (i = 0; i < (1 << frame_header->cdef_bits); ++i) {
203 info->cdef_y_strength[i] = (frame_header->cdef_y_pri_strength[i] & 0x0F) | (frame_header->cdef_y_sec_strength[i] << 4);
204 info->cdef_uv_strength[i] = (frame_header->cdef_uv_pri_strength[i] & 0x0F) | (frame_header->cdef_uv_sec_strength[i] << 4);
205 }
206
207
208 /* Segmentation */
209 for (i = 0; i < AV1_MAX_SEGMENTS; ++i) {
210 info->segmentation_feature_mask[i] = 0;
211 for (j = 0; j < AV1_SEG_LVL_MAX; ++j) {
212 info->segmentation_feature_mask[i] |= frame_header->feature_enabled[i][j] << j;
213 info->segmentation_feature_data[i][j] = frame_header->feature_value[i][j];
214 }
215 }
216
217 for (i = 0; i < AV1_NUM_REF_FRAMES; ++i) {
218 /* Loopfilter */
219 info->loop_filter_ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
220
221 /* Reference Frames */
222 info->ref_frame_map[i] = ff_vdpau_get_surface_id(s->ref[i].f) ? ff_vdpau_get_surface_id(s->ref[i].f) : VDP_INVALID_HANDLE;
223 }
224
225 if (frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE) {
226 info->primary_ref_frame = -1;
227 } else {
228 int8_t pri_ref_idx = frame_header->ref_frame_idx[frame_header->primary_ref_frame];
229 info->primary_ref_frame = info->ref_frame_map[pri_ref_idx];
230 }
231
232 for (i = 0; i < AV1_REFS_PER_FRAME; ++i) {
233 /* Ref Frame List */
234 int8_t ref_idx = frame_header->ref_frame_idx[i];
235 AVFrame *ref_frame = s->ref[ref_idx].f;
236
237 info->ref_frame[i].index = info->ref_frame_map[ref_idx];
238 info->ref_frame[i].width = ref_frame->width;
239 info->ref_frame[i].height = ref_frame->height;
240
241 /* Global Motion */
242 info->global_motion[i].invalid = !frame_header->is_global[AV1_REF_FRAME_LAST + i];
243 info->global_motion[i].wmtype = s->cur_frame.gm_type[AV1_REF_FRAME_LAST + i];
244 for (j = 0; j < 6; ++j) {
245 info->global_motion[i].wmmat[j] = s->cur_frame.gm_params[AV1_REF_FRAME_LAST + i][j];
246 }
247 }
248
249 /* Film Grain Params */
250 if (film_grain->apply_grain) {
251 for (i = 0; i < 14; ++i) {
252 info->scaling_points_y[i][0] = film_grain->point_y_value[i];
253 info->scaling_points_y[i][1] = film_grain->point_y_scaling[i];
254 }
255 for (i = 0; i < 10; ++i) {
256 info->scaling_points_cb[i][0] = film_grain->point_cb_value[i];
257 info->scaling_points_cb[i][1] = film_grain->point_cb_scaling[i];
258 info->scaling_points_cr[i][0] = film_grain->point_cr_value[i];
259 info->scaling_points_cr[i][1] = film_grain->point_cr_scaling[i];
260 }
261 for (i = 0; i < 24; ++i) {
262 info->ar_coeffs_y[i] = (short)film_grain->ar_coeffs_y_plus_128[i] - 128;
263 }
264 for (i = 0; i < 25; ++i) {
265 info->ar_coeffs_cb[i] = (short)film_grain->ar_coeffs_cb_plus_128[i] - 128;
266 info->ar_coeffs_cr[i] = (short)film_grain->ar_coeffs_cr_plus_128[i] - 128;
267 }
268 }
269
270
271 return ff_vdpau_common_start_frame(pic_ctx, buffer, size);
272
273 }
274
vdpau_av1_decode_slice(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)275 static int vdpau_av1_decode_slice(AVCodecContext *avctx,
276 const uint8_t *buffer, uint32_t size)
277 {
278 const AV1DecContext *s = avctx->priv_data;
279 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
280 struct vdpau_picture_context *pic_ctx = s->cur_frame.hwaccel_picture_private;
281 VdpPictureInfoAV1 *info = &pic_ctx->info.av1;
282 int val;
283 int nb_slices;
284 VdpBitstreamBuffer *buffers = pic_ctx->bitstream_buffers;
285 int bitstream_len = 0;
286
287 nb_slices = frame_header->tile_cols * frame_header->tile_rows;
288 /* Shortcut if all tiles are in the same buffer*/
289 if (nb_slices == s->tg_end - s->tg_start + 1) {
290 for (int i = 0; i < nb_slices; ++i) {
291 info->tile_info[i*2 ] = s->tile_group_info[i].tile_offset;
292 info->tile_info[i*2 + 1] = info->tile_info[i*2] + s->tile_group_info[i].tile_size;
293 }
294 val = ff_vdpau_add_buffer(pic_ctx, buffer, size);
295 if (val) {
296 return val;
297 }
298
299 return 0;
300 }
301
302 for(int i = 0; i < pic_ctx->bitstream_buffers_used; i++) {
303 bitstream_len += buffers->bitstream_bytes;
304 buffers++;
305 }
306
307 for (uint32_t tile_num = s->tg_start; tile_num <= s->tg_end; ++tile_num) {
308 info->tile_info[tile_num*2 ] = bitstream_len + s->tile_group_info[tile_num].tile_offset;
309 info->tile_info[tile_num*2 + 1] = info->tile_info[tile_num*2] + s->tile_group_info[tile_num].tile_size;
310 }
311
312 val = ff_vdpau_add_buffer(pic_ctx, buffer, size);
313 if (val) {
314 return val;
315 }
316
317 return 0;
318 }
319
vdpau_av1_end_frame(AVCodecContext * avctx)320 static int vdpau_av1_end_frame(AVCodecContext *avctx)
321 {
322 const AV1DecContext *s = avctx->priv_data;
323 struct vdpau_picture_context *pic_ctx = s->cur_frame.hwaccel_picture_private;
324
325 int val;
326
327 val = ff_vdpau_common_end_frame(avctx, s->cur_frame.f, pic_ctx);
328 if (val < 0)
329 return val;
330
331 return 0;
332 }
333
vdpau_av1_init(AVCodecContext * avctx)334 static int vdpau_av1_init(AVCodecContext *avctx)
335 {
336 VdpDecoderProfile profile;
337 uint32_t level = avctx->level;
338
339 switch (avctx->profile) {
340 case FF_PROFILE_AV1_MAIN:
341 profile = VDP_DECODER_PROFILE_AV1_MAIN;
342 break;
343 case FF_PROFILE_AV1_HIGH:
344 profile = VDP_DECODER_PROFILE_AV1_HIGH;
345 break;
346 case FF_PROFILE_AV1_PROFESSIONAL:
347 profile = VDP_DECODER_PROFILE_AV1_PROFESSIONAL;
348 break;
349 default:
350 return AVERROR(ENOTSUP);
351 }
352
353 return ff_vdpau_common_init(avctx, profile, level);
354 }
355
356 const AVHWAccel ff_av1_vdpau_hwaccel = {
357 .name = "av1_vdpau",
358 .type = AVMEDIA_TYPE_VIDEO,
359 .id = AV_CODEC_ID_AV1,
360 .pix_fmt = AV_PIX_FMT_VDPAU,
361 .start_frame = vdpau_av1_start_frame,
362 .end_frame = vdpau_av1_end_frame,
363 .decode_slice = vdpau_av1_decode_slice,
364 .frame_priv_data_size = sizeof(struct vdpau_picture_context),
365 .init = vdpau_av1_init,
366 .uninit = ff_vdpau_common_uninit,
367 .frame_params = ff_vdpau_common_frame_params,
368 .priv_data_size = sizeof(VDPAUContext),
369 .caps_internal = HWACCEL_CAP_ASYNC_SAFE,
370 };
371