1 /*
2 * AV1 HW decode acceleration through NVDEC
3 *
4 * Copyright (c) 2020 Timo Rothenpieler
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
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #include "avcodec.h"
24 #include "nvdec.h"
25 #include "decode.h"
26 #include "internal.h"
27 #include "av1dec.h"
28
29
get_bit_depth_from_seq(const AV1RawSequenceHeader * seq)30 static int get_bit_depth_from_seq(const AV1RawSequenceHeader *seq)
31 {
32 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
33 return seq->color_config.twelve_bit ? 12 : 10;
34 else if (seq->seq_profile <= 2 && seq->color_config.high_bitdepth)
35 return 10;
36 else
37 return 8;
38 }
39
nvdec_av1_start_frame(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)40 static int nvdec_av1_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
41 {
42 const AV1DecContext *s = avctx->priv_data;
43 const AV1RawSequenceHeader *seq = s->raw_seq;
44 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
45 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain;
46
47 NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
48 CUVIDPICPARAMS *pp = &ctx->pic_params;
49 CUVIDAV1PICPARAMS *ppc = &pp->CodecSpecific.av1;
50 FrameDecodeData *fdd;
51 NVDECFrame *cf;
52 AVFrame *cur_frame = s->cur_frame.f;
53
54 unsigned char remap_lr_type[4] = { AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ };
55
56 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain;
57 int ret, i, j;
58
59 ret = ff_nvdec_start_frame_sep_ref(avctx, cur_frame, apply_grain);
60 if (ret < 0)
61 return ret;
62
63 fdd = (FrameDecodeData*)cur_frame->private_ref->data;
64 cf = (NVDECFrame*)fdd->hwaccel_priv;
65
66 *pp = (CUVIDPICPARAMS) {
67 .PicWidthInMbs = (cur_frame->width + 15) / 16,
68 .FrameHeightInMbs = (cur_frame->height + 15) / 16,
69 .CurrPicIdx = cf->idx,
70 .ref_pic_flag = !!frame_header->refresh_frame_flags,
71 .intra_pic_flag = frame_header->frame_type == AV1_FRAME_INTRA_ONLY ||
72 frame_header->frame_type == AV1_FRAME_KEY,
73
74 .CodecSpecific.av1 = {
75 .width = cur_frame->width,
76 .height = cur_frame->height,
77
78 .frame_offset = frame_header->order_hint,
79 .decodePicIdx = cf->ref_idx,
80
81 /* Sequence Header */
82 .profile = seq->seq_profile,
83 .use_128x128_superblock = seq->use_128x128_superblock,
84 .subsampling_x = seq->color_config.subsampling_x,
85 .subsampling_y = seq->color_config.subsampling_y,
86 .mono_chrome = seq->color_config.mono_chrome,
87 .bit_depth_minus8 = get_bit_depth_from_seq(seq) - 8,
88 .enable_filter_intra = seq->enable_filter_intra,
89 .enable_intra_edge_filter = seq->enable_intra_edge_filter,
90 .enable_interintra_compound = seq->enable_interintra_compound,
91 .enable_masked_compound = seq->enable_masked_compound,
92 .enable_dual_filter = seq->enable_dual_filter,
93 .enable_order_hint = seq->enable_order_hint,
94 .order_hint_bits_minus1 = seq->order_hint_bits_minus_1,
95 .enable_jnt_comp = seq->enable_jnt_comp,
96 .enable_superres = seq->enable_superres,
97 .enable_cdef = seq->enable_cdef,
98 .enable_restoration = seq->enable_restoration,
99 .enable_fgs = seq->film_grain_params_present &&
100 !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN),
101
102 /* Frame Header */
103 .frame_type = frame_header->frame_type,
104 .show_frame = frame_header->show_frame,
105 .disable_cdf_update = frame_header->disable_cdf_update,
106 .allow_screen_content_tools = frame_header->allow_screen_content_tools,
107 .force_integer_mv = frame_header->force_integer_mv ||
108 frame_header->frame_type == AV1_FRAME_INTRA_ONLY ||
109 frame_header->frame_type == AV1_FRAME_KEY,
110 .coded_denom = frame_header->coded_denom,
111 .allow_intrabc = frame_header->allow_intrabc,
112 .allow_high_precision_mv = frame_header->allow_high_precision_mv,
113 .interp_filter = frame_header->interpolation_filter,
114 .switchable_motion_mode = frame_header->is_motion_mode_switchable,
115 .use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
116 .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
117 .delta_q_present = frame_header->delta_q_present,
118 .delta_q_res = frame_header->delta_q_res,
119 .using_qmatrix = frame_header->using_qmatrix,
120 .coded_lossless = s->cur_frame.coded_lossless,
121 .use_superres = frame_header->use_superres,
122 .tx_mode = frame_header->tx_mode,
123 .reference_mode = frame_header->reference_select,
124 .allow_warped_motion = frame_header->allow_warped_motion,
125 .reduced_tx_set = frame_header->reduced_tx_set,
126 .skip_mode = frame_header->skip_mode_present,
127
128 /* Tiling Info */
129 .num_tile_cols = frame_header->tile_cols,
130 .num_tile_rows = frame_header->tile_rows,
131 .context_update_tile_id = frame_header->context_update_tile_id,
132
133 /* CDEF */
134 .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
135 .cdef_bits = frame_header->cdef_bits,
136
137 /* SkipModeFrames */
138 .SkipModeFrame0 = frame_header->skip_mode_present ?
139 s->cur_frame.skip_mode_frame_idx[0] : 0,
140 .SkipModeFrame1 = frame_header->skip_mode_present ?
141 s->cur_frame.skip_mode_frame_idx[1] : 0,
142
143 /* QP Information */
144 .base_qindex = frame_header->base_q_idx,
145 .qp_y_dc_delta_q = frame_header->delta_q_y_dc,
146 .qp_u_dc_delta_q = frame_header->delta_q_u_dc,
147 .qp_v_dc_delta_q = frame_header->delta_q_v_dc,
148 .qp_u_ac_delta_q = frame_header->delta_q_u_ac,
149 .qp_v_ac_delta_q = frame_header->delta_q_v_ac,
150 .qm_y = frame_header->qm_y,
151 .qm_u = frame_header->qm_u,
152 .qm_v = frame_header->qm_v,
153
154 /* Segmentation */
155 .segmentation_enabled = frame_header->segmentation_enabled,
156 .segmentation_update_map = frame_header->segmentation_update_map,
157 .segmentation_update_data = frame_header->segmentation_update_data,
158 .segmentation_temporal_update = frame_header->segmentation_temporal_update,
159
160 /* Loopfilter */
161 .loop_filter_level[0] = frame_header->loop_filter_level[0],
162 .loop_filter_level[1] = frame_header->loop_filter_level[1],
163 .loop_filter_level_u = frame_header->loop_filter_level[2],
164 .loop_filter_level_v = frame_header->loop_filter_level[3],
165 .loop_filter_sharpness = frame_header->loop_filter_sharpness,
166 .loop_filter_delta_enabled = frame_header->loop_filter_delta_enabled,
167 .loop_filter_delta_update = frame_header->loop_filter_delta_update,
168 .loop_filter_mode_deltas[0] = frame_header->loop_filter_mode_deltas[0],
169 .loop_filter_mode_deltas[1] = frame_header->loop_filter_mode_deltas[1],
170 .delta_lf_present = frame_header->delta_lf_present,
171 .delta_lf_res = frame_header->delta_lf_res,
172 .delta_lf_multi = frame_header->delta_lf_multi,
173
174 /* Restoration */
175 .lr_type[0] = remap_lr_type[frame_header->lr_type[0]],
176 .lr_type[1] = remap_lr_type[frame_header->lr_type[1]],
177 .lr_type[2] = remap_lr_type[frame_header->lr_type[2]],
178 .lr_unit_size[0] = 1 + frame_header->lr_unit_shift,
179 .lr_unit_size[1] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift,
180 .lr_unit_size[2] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift,
181
182 /* Reference Frames */
183 .temporal_layer_id = s->cur_frame.temporal_id,
184 .spatial_layer_id = s->cur_frame.spatial_id,
185
186 /* Film Grain Params */
187 .apply_grain = apply_grain,
188 .overlap_flag = film_grain->overlap_flag,
189 .scaling_shift_minus8 = film_grain->grain_scaling_minus_8,
190 .chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma,
191 .ar_coeff_lag = film_grain->ar_coeff_lag,
192 .ar_coeff_shift_minus6 = film_grain->ar_coeff_shift_minus_6,
193 .grain_scale_shift = film_grain->grain_scale_shift,
194 .clip_to_restricted_range = film_grain->clip_to_restricted_range,
195 .num_y_points = film_grain->num_y_points,
196 .num_cb_points = film_grain->num_cb_points,
197 .num_cr_points = film_grain->num_cr_points,
198 .random_seed = film_grain->grain_seed,
199 .cb_mult = film_grain->cb_mult,
200 .cb_luma_mult = film_grain->cb_luma_mult,
201 .cb_offset = film_grain->cb_offset,
202 .cr_mult = film_grain->cr_mult,
203 .cr_luma_mult = film_grain->cr_luma_mult,
204 .cr_offset = film_grain->cr_offset
205 }
206 };
207
208 /* Tiling Info */
209 for (i = 0; i < frame_header->tile_cols; ++i) {
210 ppc->tile_widths[i] = frame_header->width_in_sbs_minus_1[i] + 1;
211 }
212 for (i = 0; i < frame_header->tile_rows; ++i) {
213 ppc->tile_heights[i] = frame_header->height_in_sbs_minus_1[i] + 1;
214 }
215
216 /* CDEF */
217 for (i = 0; i < (1 << frame_header->cdef_bits); ++i) {
218 ppc->cdef_y_strength[i] = (frame_header->cdef_y_pri_strength[i] & 0x0F) | (frame_header->cdef_y_sec_strength[i] << 4);
219 ppc->cdef_uv_strength[i] = (frame_header->cdef_uv_pri_strength[i] & 0x0F) | (frame_header->cdef_uv_sec_strength[i] << 4);
220 }
221
222 /* Segmentation */
223 for (i = 0; i < AV1_MAX_SEGMENTS; ++i) {
224 ppc->segmentation_feature_mask[i] = 0;
225 for (j = 0; j < AV1_SEG_LVL_MAX; ++j) {
226 ppc->segmentation_feature_mask[i] |= frame_header->feature_enabled[i][j] << j;
227 ppc->segmentation_feature_data[i][j] = frame_header->feature_value[i][j];
228 }
229 }
230
231 for (i = 0; i < AV1_NUM_REF_FRAMES; ++i) {
232 /* Loopfilter */
233 ppc->loop_filter_ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
234
235 /* Reference Frames */
236 ppc->ref_frame_map[i] = ff_nvdec_get_ref_idx(s->ref[i].f);
237 }
238
239 if (frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE) {
240 ppc->primary_ref_frame = -1;
241 } else {
242 int8_t pri_ref_idx = frame_header->ref_frame_idx[frame_header->primary_ref_frame];
243 ppc->primary_ref_frame = ppc->ref_frame_map[pri_ref_idx];
244 }
245
246 for (i = 0; i < AV1_REFS_PER_FRAME; ++i) {
247 /* Ref Frame List */
248 int8_t ref_idx = frame_header->ref_frame_idx[i];
249 AVFrame *ref_frame = s->ref[ref_idx].f;
250
251 ppc->ref_frame[i].index = ppc->ref_frame_map[ref_idx];
252 ppc->ref_frame[i].width = ref_frame->width;
253 ppc->ref_frame[i].height = ref_frame->height;
254
255 /* Global Motion */
256 ppc->global_motion[i].invalid = !frame_header->is_global[AV1_REF_FRAME_LAST + i];
257 ppc->global_motion[i].wmtype = s->cur_frame.gm_type[AV1_REF_FRAME_LAST + i];
258 for (j = 0; j < 6; ++j) {
259 ppc->global_motion[i].wmmat[j] = s->cur_frame.gm_params[AV1_REF_FRAME_LAST + i][j];
260 }
261 }
262
263 /* Film Grain Params */
264 if (apply_grain) {
265 for (i = 0; i < 14; ++i) {
266 ppc->scaling_points_y[i][0] = film_grain->point_y_value[i];
267 ppc->scaling_points_y[i][1] = film_grain->point_y_scaling[i];
268 }
269 for (i = 0; i < 10; ++i) {
270 ppc->scaling_points_cb[i][0] = film_grain->point_cb_value[i];
271 ppc->scaling_points_cb[i][1] = film_grain->point_cb_scaling[i];
272 ppc->scaling_points_cr[i][0] = film_grain->point_cr_value[i];
273 ppc->scaling_points_cr[i][1] = film_grain->point_cr_scaling[i];
274 }
275 for (i = 0; i < 24; ++i) {
276 ppc->ar_coeffs_y[i] = (short)film_grain->ar_coeffs_y_plus_128[i] - 128;
277 }
278 for (i = 0; i < 25; ++i) {
279 ppc->ar_coeffs_cb[i] = (short)film_grain->ar_coeffs_cb_plus_128[i] - 128;
280 ppc->ar_coeffs_cr[i] = (short)film_grain->ar_coeffs_cr_plus_128[i] - 128;
281 }
282 }
283
284 return 0;
285 }
286
nvdec_av1_decode_slice(AVCodecContext * avctx,const uint8_t * buffer,uint32_t size)287 static int nvdec_av1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
288 {
289 const AV1DecContext *s = avctx->priv_data;
290 const AV1RawFrameHeader *frame_header = s->raw_frame_header;
291 NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
292 void *tmp;
293
294 ctx->nb_slices = frame_header->tile_cols * frame_header->tile_rows;
295
296 tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,
297 ctx->nb_slices * 2 * sizeof(*ctx->slice_offsets));
298 if (!tmp) {
299 return AVERROR(ENOMEM);
300 }
301 ctx->slice_offsets = tmp;
302
303 /* Shortcut if all tiles are in the same buffer */
304 if (ctx->nb_slices == s->tg_end - s->tg_start + 1) {
305 ctx->bitstream = (uint8_t*)buffer;
306 ctx->bitstream_len = size;
307
308 for (int i = 0; i < ctx->nb_slices; ++i) {
309 ctx->slice_offsets[i*2 ] = s->tile_group_info[i].tile_offset;
310 ctx->slice_offsets[i*2 + 1] = ctx->slice_offsets[i*2] + s->tile_group_info[i].tile_size;
311 }
312
313 return 0;
314 }
315
316 tmp = av_fast_realloc(ctx->bitstream_internal, &ctx->bitstream_allocated,
317 ctx->bitstream_len + size);
318 if (!tmp) {
319 return AVERROR(ENOMEM);
320 }
321 ctx->bitstream = ctx->bitstream_internal = tmp;
322
323 memcpy(ctx->bitstream + ctx->bitstream_len, buffer, size);
324
325 for (uint32_t tile_num = s->tg_start; tile_num <= s->tg_end; ++tile_num) {
326 ctx->slice_offsets[tile_num*2 ] = ctx->bitstream_len + s->tile_group_info[tile_num].tile_offset;
327 ctx->slice_offsets[tile_num*2 + 1] = ctx->slice_offsets[tile_num*2] + s->tile_group_info[tile_num].tile_size;
328 }
329 ctx->bitstream_len += size;
330
331 return 0;
332 }
333
nvdec_av1_frame_params(AVCodecContext * avctx,AVBufferRef * hw_frames_ctx)334 static int nvdec_av1_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx)
335 {
336 /* Maximum of 8 reference frames, but potentially stored twice due to film grain */
337 return ff_nvdec_frame_params(avctx, hw_frames_ctx, 8 * 2, 0);
338 }
339
340 const AVHWAccel ff_av1_nvdec_hwaccel = {
341 .name = "av1_nvdec",
342 .type = AVMEDIA_TYPE_VIDEO,
343 .id = AV_CODEC_ID_AV1,
344 .pix_fmt = AV_PIX_FMT_CUDA,
345 .start_frame = nvdec_av1_start_frame,
346 .end_frame = ff_nvdec_simple_end_frame,
347 .decode_slice = nvdec_av1_decode_slice,
348 .frame_params = nvdec_av1_frame_params,
349 .init = ff_nvdec_decode_init,
350 .uninit = ff_nvdec_decode_uninit,
351 .priv_data_size = sizeof(NVDECContext),
352 };
353