1 |
|
|
/* |
2 |
|
|
* HEVC HW decode acceleration through VA API |
3 |
|
|
* |
4 |
|
|
* Copyright (C) 2015 Timo Rothenpieler <timo@rothenpieler.org> |
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 <va/va.h> |
24 |
|
|
#include <va/va_dec_hevc.h> |
25 |
|
|
|
26 |
|
|
#include "avcodec.h" |
27 |
|
|
#include "hevcdec.h" |
28 |
|
|
#include "hwconfig.h" |
29 |
|
|
#include "vaapi_decode.h" |
30 |
|
|
#include "vaapi_hevc.h" |
31 |
|
|
#include "h265_profile_level.h" |
32 |
|
|
|
33 |
|
|
typedef struct VAAPIDecodePictureHEVC { |
34 |
|
|
#if VA_CHECK_VERSION(1, 2, 0) |
35 |
|
|
VAPictureParameterBufferHEVCExtension pic_param; |
36 |
|
|
VASliceParameterBufferHEVCExtension last_slice_param; |
37 |
|
|
#else |
38 |
|
|
VAPictureParameterBufferHEVC pic_param; |
39 |
|
|
VASliceParameterBufferHEVC last_slice_param; |
40 |
|
|
#endif |
41 |
|
|
const uint8_t *last_buffer; |
42 |
|
|
size_t last_size; |
43 |
|
|
|
44 |
|
|
VAAPIDecodePicture pic; |
45 |
|
|
} VAAPIDecodePictureHEVC; |
46 |
|
|
|
47 |
|
|
static void init_vaapi_pic(VAPictureHEVC *va_pic) |
48 |
|
|
{ |
49 |
|
|
va_pic->picture_id = VA_INVALID_ID; |
50 |
|
|
va_pic->flags = VA_PICTURE_HEVC_INVALID; |
51 |
|
|
va_pic->pic_order_cnt = 0; |
52 |
|
|
} |
53 |
|
|
|
54 |
|
|
static void fill_vaapi_pic(VAPictureHEVC *va_pic, const HEVCFrame *pic, int rps_type) |
55 |
|
|
{ |
56 |
|
|
va_pic->picture_id = ff_vaapi_get_surface_id(pic->frame); |
57 |
|
|
va_pic->pic_order_cnt = pic->poc; |
58 |
|
|
va_pic->flags = rps_type; |
59 |
|
|
|
60 |
|
|
if (pic->flags & HEVC_FRAME_FLAG_LONG_REF) |
61 |
|
|
va_pic->flags |= VA_PICTURE_HEVC_LONG_TERM_REFERENCE; |
62 |
|
|
|
63 |
|
|
if (pic->frame->interlaced_frame) { |
64 |
|
|
va_pic->flags |= VA_PICTURE_HEVC_FIELD_PIC; |
65 |
|
|
|
66 |
|
|
if (!pic->frame->top_field_first) |
67 |
|
|
va_pic->flags |= VA_PICTURE_HEVC_BOTTOM_FIELD; |
68 |
|
|
} |
69 |
|
|
} |
70 |
|
|
|
71 |
|
|
static int find_frame_rps_type(const HEVCContext *h, const HEVCFrame *pic) |
72 |
|
|
{ |
73 |
|
|
VASurfaceID pic_surf = ff_vaapi_get_surface_id(pic->frame); |
74 |
|
|
int i; |
75 |
|
|
|
76 |
|
|
for (i = 0; i < h->rps[ST_CURR_BEF].nb_refs; i++) { |
77 |
|
|
if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_BEF].ref[i]->frame)) |
78 |
|
|
return VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE; |
79 |
|
|
} |
80 |
|
|
|
81 |
|
|
for (i = 0; i < h->rps[ST_CURR_AFT].nb_refs; i++) { |
82 |
|
|
if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_AFT].ref[i]->frame)) |
83 |
|
|
return VA_PICTURE_HEVC_RPS_ST_CURR_AFTER; |
84 |
|
|
} |
85 |
|
|
|
86 |
|
|
for (i = 0; i < h->rps[LT_CURR].nb_refs; i++) { |
87 |
|
|
if (pic_surf == ff_vaapi_get_surface_id(h->rps[LT_CURR].ref[i]->frame)) |
88 |
|
|
return VA_PICTURE_HEVC_RPS_LT_CURR; |
89 |
|
|
} |
90 |
|
|
|
91 |
|
|
return 0; |
92 |
|
|
} |
93 |
|
|
|
94 |
|
|
static void fill_vaapi_reference_frames(const HEVCContext *h, VAPictureParameterBufferHEVC *pp) |
95 |
|
|
{ |
96 |
|
|
const HEVCFrame *current_picture = h->ref; |
97 |
|
|
int i, j, rps_type; |
98 |
|
|
|
99 |
|
|
for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); i++) { |
100 |
|
|
const HEVCFrame *frame = NULL; |
101 |
|
|
|
102 |
|
|
while (!frame && j < FF_ARRAY_ELEMS(h->DPB)) { |
103 |
|
|
if (&h->DPB[j] != current_picture && (h->DPB[j].flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF))) |
104 |
|
|
frame = &h->DPB[j]; |
105 |
|
|
j++; |
106 |
|
|
} |
107 |
|
|
|
108 |
|
|
init_vaapi_pic(&pp->ReferenceFrames[i]); |
109 |
|
|
|
110 |
|
|
if (frame) { |
111 |
|
|
rps_type = find_frame_rps_type(h, frame); |
112 |
|
|
fill_vaapi_pic(&pp->ReferenceFrames[i], frame, rps_type); |
113 |
|
|
} |
114 |
|
|
} |
115 |
|
|
} |
116 |
|
|
|
117 |
|
|
static int vaapi_hevc_start_frame(AVCodecContext *avctx, |
118 |
|
|
av_unused const uint8_t *buffer, |
119 |
|
|
av_unused uint32_t size) |
120 |
|
|
{ |
121 |
|
|
const HEVCContext *h = avctx->priv_data; |
122 |
|
|
VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private; |
123 |
|
|
const HEVCSPS *sps = h->ps.sps; |
124 |
|
|
const HEVCPPS *pps = h->ps.pps; |
125 |
|
|
|
126 |
|
|
const ScalingList *scaling_list = NULL; |
127 |
|
|
int pic_param_size, err, i; |
128 |
|
|
|
129 |
|
|
VAPictureParameterBufferHEVC *pic_param = (VAPictureParameterBufferHEVC *)&pic->pic_param; |
130 |
|
|
|
131 |
|
|
pic->pic.output_surface = ff_vaapi_get_surface_id(h->ref->frame); |
132 |
|
|
|
133 |
|
|
*pic_param = (VAPictureParameterBufferHEVC) { |
134 |
|
|
.pic_width_in_luma_samples = sps->width, |
135 |
|
|
.pic_height_in_luma_samples = sps->height, |
136 |
|
|
.log2_min_luma_coding_block_size_minus3 = sps->log2_min_cb_size - 3, |
137 |
|
|
.sps_max_dec_pic_buffering_minus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1, |
138 |
|
|
.log2_diff_max_min_luma_coding_block_size = sps->log2_diff_max_min_coding_block_size, |
139 |
|
|
.log2_min_transform_block_size_minus2 = sps->log2_min_tb_size - 2, |
140 |
|
|
.log2_diff_max_min_transform_block_size = sps->log2_max_trafo_size - sps->log2_min_tb_size, |
141 |
|
|
.max_transform_hierarchy_depth_inter = sps->max_transform_hierarchy_depth_inter, |
142 |
|
|
.max_transform_hierarchy_depth_intra = sps->max_transform_hierarchy_depth_intra, |
143 |
|
|
.num_short_term_ref_pic_sets = sps->nb_st_rps, |
144 |
|
|
.num_long_term_ref_pic_sps = sps->num_long_term_ref_pics_sps, |
145 |
|
|
.num_ref_idx_l0_default_active_minus1 = pps->num_ref_idx_l0_default_active - 1, |
146 |
|
|
.num_ref_idx_l1_default_active_minus1 = pps->num_ref_idx_l1_default_active - 1, |
147 |
|
|
.init_qp_minus26 = pps->pic_init_qp_minus26, |
148 |
|
|
.pps_cb_qp_offset = pps->cb_qp_offset, |
149 |
|
|
.pps_cr_qp_offset = pps->cr_qp_offset, |
150 |
|
|
.pcm_sample_bit_depth_luma_minus1 = sps->pcm.bit_depth - 1, |
151 |
|
|
.pcm_sample_bit_depth_chroma_minus1 = sps->pcm.bit_depth_chroma - 1, |
152 |
|
|
.log2_min_pcm_luma_coding_block_size_minus3 = sps->pcm.log2_min_pcm_cb_size - 3, |
153 |
|
|
.log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size, |
154 |
|
|
.diff_cu_qp_delta_depth = pps->diff_cu_qp_delta_depth, |
155 |
|
|
.pps_beta_offset_div2 = pps->beta_offset / 2, |
156 |
|
|
.pps_tc_offset_div2 = pps->tc_offset / 2, |
157 |
|
|
.log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level - 2, |
158 |
|
|
.bit_depth_luma_minus8 = sps->bit_depth - 8, |
159 |
|
|
.bit_depth_chroma_minus8 = sps->bit_depth - 8, |
160 |
|
|
.log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4, |
161 |
|
|
.num_extra_slice_header_bits = pps->num_extra_slice_header_bits, |
162 |
|
|
.pic_fields.bits = { |
163 |
|
|
.chroma_format_idc = sps->chroma_format_idc, |
164 |
|
|
.tiles_enabled_flag = pps->tiles_enabled_flag, |
165 |
|
|
.separate_colour_plane_flag = sps->separate_colour_plane_flag, |
166 |
|
|
.pcm_enabled_flag = sps->pcm_enabled_flag, |
167 |
|
|
.scaling_list_enabled_flag = sps->scaling_list_enable_flag, |
168 |
|
|
.transform_skip_enabled_flag = pps->transform_skip_enabled_flag, |
169 |
|
|
.amp_enabled_flag = sps->amp_enabled_flag, |
170 |
|
|
.strong_intra_smoothing_enabled_flag = sps->sps_strong_intra_smoothing_enable_flag, |
171 |
|
|
.sign_data_hiding_enabled_flag = pps->sign_data_hiding_flag, |
172 |
|
|
.constrained_intra_pred_flag = pps->constrained_intra_pred_flag, |
173 |
|
|
.cu_qp_delta_enabled_flag = pps->cu_qp_delta_enabled_flag, |
174 |
|
|
.weighted_pred_flag = pps->weighted_pred_flag, |
175 |
|
|
.weighted_bipred_flag = pps->weighted_bipred_flag, |
176 |
|
|
.transquant_bypass_enabled_flag = pps->transquant_bypass_enable_flag, |
177 |
|
|
.entropy_coding_sync_enabled_flag = pps->entropy_coding_sync_enabled_flag, |
178 |
|
|
.pps_loop_filter_across_slices_enabled_flag = pps->seq_loop_filter_across_slices_enabled_flag, |
179 |
|
|
.loop_filter_across_tiles_enabled_flag = pps->loop_filter_across_tiles_enabled_flag, |
180 |
|
|
.pcm_loop_filter_disabled_flag = sps->pcm.loop_filter_disable_flag, |
181 |
|
|
}, |
182 |
|
|
.slice_parsing_fields.bits = { |
183 |
|
|
.lists_modification_present_flag = pps->lists_modification_present_flag, |
184 |
|
|
.long_term_ref_pics_present_flag = sps->long_term_ref_pics_present_flag, |
185 |
|
|
.sps_temporal_mvp_enabled_flag = sps->sps_temporal_mvp_enabled_flag, |
186 |
|
|
.cabac_init_present_flag = pps->cabac_init_present_flag, |
187 |
|
|
.output_flag_present_flag = pps->output_flag_present_flag, |
188 |
|
|
.dependent_slice_segments_enabled_flag = pps->dependent_slice_segments_enabled_flag, |
189 |
|
|
.pps_slice_chroma_qp_offsets_present_flag = pps->pic_slice_level_chroma_qp_offsets_present_flag, |
190 |
|
|
.sample_adaptive_offset_enabled_flag = sps->sao_enabled, |
191 |
|
|
.deblocking_filter_override_enabled_flag = pps->deblocking_filter_override_enabled_flag, |
192 |
|
|
.pps_disable_deblocking_filter_flag = pps->disable_dbf, |
193 |
|
|
.slice_segment_header_extension_present_flag = pps->slice_header_extension_present_flag, |
194 |
|
|
.RapPicFlag = IS_IRAP(h), |
195 |
|
|
.IdrPicFlag = IS_IDR(h), |
196 |
|
|
.IntraPicFlag = IS_IRAP(h), |
197 |
|
|
}, |
198 |
|
|
}; |
199 |
|
|
|
200 |
|
|
fill_vaapi_pic(&pic_param->CurrPic, h->ref, 0); |
201 |
|
|
fill_vaapi_reference_frames(h, pic_param); |
202 |
|
|
|
203 |
|
|
if (pps->tiles_enabled_flag) { |
204 |
|
|
pic_param->num_tile_columns_minus1 = pps->num_tile_columns - 1; |
205 |
|
|
pic_param->num_tile_rows_minus1 = pps->num_tile_rows - 1; |
206 |
|
|
|
207 |
|
|
for (i = 0; i < pps->num_tile_columns; i++) |
208 |
|
|
pic_param->column_width_minus1[i] = pps->column_width[i] - 1; |
209 |
|
|
|
210 |
|
|
for (i = 0; i < pps->num_tile_rows; i++) |
211 |
|
|
pic_param->row_height_minus1[i] = pps->row_height[i] - 1; |
212 |
|
|
} |
213 |
|
|
|
214 |
|
|
if (h->sh.short_term_ref_pic_set_sps_flag == 0 && h->sh.short_term_rps) { |
215 |
|
|
pic_param->st_rps_bits = h->sh.short_term_ref_pic_set_size; |
216 |
|
|
} else { |
217 |
|
|
pic_param->st_rps_bits = 0; |
218 |
|
|
} |
219 |
|
|
|
220 |
|
|
#if VA_CHECK_VERSION(1, 2, 0) |
221 |
|
|
if (avctx->profile == FF_PROFILE_HEVC_REXT) { |
222 |
|
|
pic->pic_param.rext = (VAPictureParameterBufferHEVCRext) { |
223 |
|
|
.range_extension_pic_fields.bits = { |
224 |
|
|
.transform_skip_rotation_enabled_flag = sps->transform_skip_rotation_enabled_flag, |
225 |
|
|
.transform_skip_context_enabled_flag = sps->transform_skip_context_enabled_flag, |
226 |
|
|
.implicit_rdpcm_enabled_flag = sps->implicit_rdpcm_enabled_flag, |
227 |
|
|
.explicit_rdpcm_enabled_flag = sps->explicit_rdpcm_enabled_flag, |
228 |
|
|
.extended_precision_processing_flag = sps->extended_precision_processing_flag, |
229 |
|
|
.intra_smoothing_disabled_flag = sps->intra_smoothing_disabled_flag, |
230 |
|
|
.high_precision_offsets_enabled_flag = sps->high_precision_offsets_enabled_flag, |
231 |
|
|
.persistent_rice_adaptation_enabled_flag = sps->persistent_rice_adaptation_enabled_flag, |
232 |
|
|
.cabac_bypass_alignment_enabled_flag = sps->cabac_bypass_alignment_enabled_flag, |
233 |
|
|
.cross_component_prediction_enabled_flag = pps->cross_component_prediction_enabled_flag, |
234 |
|
|
.chroma_qp_offset_list_enabled_flag = pps->chroma_qp_offset_list_enabled_flag, |
235 |
|
|
}, |
236 |
|
|
.diff_cu_chroma_qp_offset_depth = pps->diff_cu_chroma_qp_offset_depth, |
237 |
|
|
.chroma_qp_offset_list_len_minus1 = pps->chroma_qp_offset_list_len_minus1, |
238 |
|
|
.log2_sao_offset_scale_luma = pps->log2_sao_offset_scale_luma, |
239 |
|
|
.log2_sao_offset_scale_chroma = pps->log2_sao_offset_scale_chroma, |
240 |
|
|
.log2_max_transform_skip_block_size_minus2 = pps->log2_max_transform_skip_block_size - 2, |
241 |
|
|
}; |
242 |
|
|
|
243 |
|
|
for (i = 0; i < 6; i++) |
244 |
|
|
pic->pic_param.rext.cb_qp_offset_list[i] = pps->cb_qp_offset_list[i]; |
245 |
|
|
for (i = 0; i < 6; i++) |
246 |
|
|
pic->pic_param.rext.cr_qp_offset_list[i] = pps->cr_qp_offset_list[i]; |
247 |
|
|
} |
248 |
|
|
#endif |
249 |
|
|
pic_param_size = avctx->profile == FF_PROFILE_HEVC_REXT ? |
250 |
|
|
sizeof(pic->pic_param) : sizeof(VAPictureParameterBufferHEVC); |
251 |
|
|
|
252 |
|
|
err = ff_vaapi_decode_make_param_buffer(avctx, &pic->pic, |
253 |
|
|
VAPictureParameterBufferType, |
254 |
|
|
&pic->pic_param, pic_param_size); |
255 |
|
|
if (err < 0) |
256 |
|
|
goto fail; |
257 |
|
|
|
258 |
|
|
if (pps->scaling_list_data_present_flag) |
259 |
|
|
scaling_list = &pps->scaling_list; |
260 |
|
|
else if (sps->scaling_list_enable_flag) |
261 |
|
|
scaling_list = &sps->scaling_list; |
262 |
|
|
|
263 |
|
|
if (scaling_list) { |
264 |
|
|
VAIQMatrixBufferHEVC iq_matrix; |
265 |
|
|
int j; |
266 |
|
|
|
267 |
|
|
for (i = 0; i < 6; i++) { |
268 |
|
|
for (j = 0; j < 16; j++) |
269 |
|
|
iq_matrix.ScalingList4x4[i][j] = scaling_list->sl[0][i][j]; |
270 |
|
|
for (j = 0; j < 64; j++) { |
271 |
|
|
iq_matrix.ScalingList8x8[i][j] = scaling_list->sl[1][i][j]; |
272 |
|
|
iq_matrix.ScalingList16x16[i][j] = scaling_list->sl[2][i][j]; |
273 |
|
|
if (i < 2) |
274 |
|
|
iq_matrix.ScalingList32x32[i][j] = scaling_list->sl[3][i * 3][j]; |
275 |
|
|
} |
276 |
|
|
iq_matrix.ScalingListDC16x16[i] = scaling_list->sl_dc[0][i]; |
277 |
|
|
if (i < 2) |
278 |
|
|
iq_matrix.ScalingListDC32x32[i] = scaling_list->sl_dc[1][i * 3]; |
279 |
|
|
} |
280 |
|
|
|
281 |
|
|
err = ff_vaapi_decode_make_param_buffer(avctx, &pic->pic, |
282 |
|
|
VAIQMatrixBufferType, |
283 |
|
|
&iq_matrix, sizeof(iq_matrix)); |
284 |
|
|
if (err < 0) |
285 |
|
|
goto fail; |
286 |
|
|
} |
287 |
|
|
|
288 |
|
|
return 0; |
289 |
|
|
|
290 |
|
|
fail: |
291 |
|
|
ff_vaapi_decode_cancel(avctx, &pic->pic); |
292 |
|
|
return err; |
293 |
|
|
} |
294 |
|
|
|
295 |
|
|
static int vaapi_hevc_end_frame(AVCodecContext *avctx) |
296 |
|
|
{ |
297 |
|
|
const HEVCContext *h = avctx->priv_data; |
298 |
|
|
VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private; |
299 |
|
|
VASliceParameterBufferHEVC *last_slice_param = (VASliceParameterBufferHEVC *)&pic->last_slice_param; |
300 |
|
|
int ret; |
301 |
|
|
|
302 |
|
|
int slice_param_size = avctx->profile == FF_PROFILE_HEVC_REXT ? |
303 |
|
|
sizeof(pic->last_slice_param) : sizeof(VASliceParameterBufferHEVC); |
304 |
|
|
|
305 |
|
|
if (pic->last_size) { |
306 |
|
|
last_slice_param->LongSliceFlags.fields.LastSliceOfPic = 1; |
307 |
|
|
ret = ff_vaapi_decode_make_slice_buffer(avctx, &pic->pic, |
308 |
|
|
&pic->last_slice_param, slice_param_size, |
309 |
|
|
pic->last_buffer, pic->last_size); |
310 |
|
|
if (ret < 0) |
311 |
|
|
goto fail; |
312 |
|
|
} |
313 |
|
|
|
314 |
|
|
|
315 |
|
|
ret = ff_vaapi_decode_issue(avctx, &pic->pic); |
316 |
|
|
if (ret < 0) |
317 |
|
|
goto fail; |
318 |
|
|
|
319 |
|
|
return 0; |
320 |
|
|
fail: |
321 |
|
|
ff_vaapi_decode_cancel(avctx, &pic->pic); |
322 |
|
|
return ret; |
323 |
|
|
} |
324 |
|
|
|
325 |
|
|
static void fill_pred_weight_table(const HEVCContext *h, |
326 |
|
|
const SliceHeader *sh, |
327 |
|
|
VASliceParameterBufferHEVC *slice_param) |
328 |
|
|
{ |
329 |
|
|
int i; |
330 |
|
|
|
331 |
|
|
memset(slice_param->delta_luma_weight_l0, 0, sizeof(slice_param->delta_luma_weight_l0)); |
332 |
|
|
memset(slice_param->delta_luma_weight_l1, 0, sizeof(slice_param->delta_luma_weight_l1)); |
333 |
|
|
memset(slice_param->luma_offset_l0, 0, sizeof(slice_param->luma_offset_l0)); |
334 |
|
|
memset(slice_param->luma_offset_l1, 0, sizeof(slice_param->luma_offset_l1)); |
335 |
|
|
memset(slice_param->delta_chroma_weight_l0, 0, sizeof(slice_param->delta_chroma_weight_l0)); |
336 |
|
|
memset(slice_param->delta_chroma_weight_l1, 0, sizeof(slice_param->delta_chroma_weight_l1)); |
337 |
|
|
memset(slice_param->ChromaOffsetL0, 0, sizeof(slice_param->ChromaOffsetL0)); |
338 |
|
|
memset(slice_param->ChromaOffsetL1, 0, sizeof(slice_param->ChromaOffsetL1)); |
339 |
|
|
|
340 |
|
|
slice_param->delta_chroma_log2_weight_denom = 0; |
341 |
|
|
slice_param->luma_log2_weight_denom = 0; |
342 |
|
|
|
343 |
|
|
if (sh->slice_type == HEVC_SLICE_I || |
344 |
|
|
(sh->slice_type == HEVC_SLICE_P && !h->ps.pps->weighted_pred_flag) || |
345 |
|
|
(sh->slice_type == HEVC_SLICE_B && !h->ps.pps->weighted_bipred_flag)) |
346 |
|
|
return; |
347 |
|
|
|
348 |
|
|
slice_param->luma_log2_weight_denom = sh->luma_log2_weight_denom; |
349 |
|
|
|
350 |
|
|
if (h->ps.sps->chroma_format_idc) { |
351 |
|
|
slice_param->delta_chroma_log2_weight_denom = sh->chroma_log2_weight_denom - sh->luma_log2_weight_denom; |
352 |
|
|
} |
353 |
|
|
|
354 |
|
|
for (i = 0; i < 15 && i < sh->nb_refs[L0]; i++) { |
355 |
|
|
slice_param->delta_luma_weight_l0[i] = sh->luma_weight_l0[i] - (1 << sh->luma_log2_weight_denom); |
356 |
|
|
slice_param->luma_offset_l0[i] = sh->luma_offset_l0[i]; |
357 |
|
|
slice_param->delta_chroma_weight_l0[i][0] = sh->chroma_weight_l0[i][0] - (1 << sh->chroma_log2_weight_denom); |
358 |
|
|
slice_param->delta_chroma_weight_l0[i][1] = sh->chroma_weight_l0[i][1] - (1 << sh->chroma_log2_weight_denom); |
359 |
|
|
slice_param->ChromaOffsetL0[i][0] = sh->chroma_offset_l0[i][0]; |
360 |
|
|
slice_param->ChromaOffsetL0[i][1] = sh->chroma_offset_l0[i][1]; |
361 |
|
|
} |
362 |
|
|
|
363 |
|
|
if (sh->slice_type == HEVC_SLICE_B) { |
364 |
|
|
for (i = 0; i < 15 && i < sh->nb_refs[L1]; i++) { |
365 |
|
|
slice_param->delta_luma_weight_l1[i] = sh->luma_weight_l1[i] - (1 << sh->luma_log2_weight_denom); |
366 |
|
|
slice_param->luma_offset_l1[i] = sh->luma_offset_l1[i]; |
367 |
|
|
slice_param->delta_chroma_weight_l1[i][0] = sh->chroma_weight_l1[i][0] - (1 << sh->chroma_log2_weight_denom); |
368 |
|
|
slice_param->delta_chroma_weight_l1[i][1] = sh->chroma_weight_l1[i][1] - (1 << sh->chroma_log2_weight_denom); |
369 |
|
|
slice_param->ChromaOffsetL1[i][0] = sh->chroma_offset_l1[i][0]; |
370 |
|
|
slice_param->ChromaOffsetL1[i][1] = sh->chroma_offset_l1[i][1]; |
371 |
|
|
} |
372 |
|
|
} |
373 |
|
|
} |
374 |
|
|
|
375 |
|
|
static uint8_t get_ref_pic_index(const HEVCContext *h, const HEVCFrame *frame) |
376 |
|
|
{ |
377 |
|
|
VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private; |
378 |
|
|
VAPictureParameterBufferHEVC *pp = (VAPictureParameterBufferHEVC *)&pic->pic_param; |
379 |
|
|
uint8_t i; |
380 |
|
|
|
381 |
|
|
if (!frame) |
382 |
|
|
return 0xff; |
383 |
|
|
|
384 |
|
|
for (i = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); i++) { |
385 |
|
|
VASurfaceID pid = pp->ReferenceFrames[i].picture_id; |
386 |
|
|
int poc = pp->ReferenceFrames[i].pic_order_cnt; |
387 |
|
|
if (pid != VA_INVALID_ID && pid == ff_vaapi_get_surface_id(frame->frame) && poc == frame->poc) |
388 |
|
|
return i; |
389 |
|
|
} |
390 |
|
|
|
391 |
|
|
return 0xff; |
392 |
|
|
} |
393 |
|
|
|
394 |
|
|
static int vaapi_hevc_decode_slice(AVCodecContext *avctx, |
395 |
|
|
const uint8_t *buffer, |
396 |
|
|
uint32_t size) |
397 |
|
|
{ |
398 |
|
|
const HEVCContext *h = avctx->priv_data; |
399 |
|
|
const SliceHeader *sh = &h->sh; |
400 |
|
|
VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private; |
401 |
|
|
VASliceParameterBufferHEVC *last_slice_param = (VASliceParameterBufferHEVC *)&pic->last_slice_param; |
402 |
|
|
|
403 |
|
|
int slice_param_size = avctx->profile == FF_PROFILE_HEVC_REXT ? |
404 |
|
|
sizeof(pic->last_slice_param) : sizeof(VASliceParameterBufferHEVC); |
405 |
|
|
|
406 |
|
|
int nb_list = (sh->slice_type == HEVC_SLICE_B) ? |
407 |
|
|
2 : (sh->slice_type == HEVC_SLICE_I ? 0 : 1); |
408 |
|
|
|
409 |
|
|
int err, i, list_idx; |
410 |
|
|
|
411 |
|
|
if (!sh->first_slice_in_pic_flag) { |
412 |
|
|
err = ff_vaapi_decode_make_slice_buffer(avctx, &pic->pic, |
413 |
|
|
&pic->last_slice_param, slice_param_size, |
414 |
|
|
pic->last_buffer, pic->last_size); |
415 |
|
|
pic->last_buffer = NULL; |
416 |
|
|
pic->last_size = 0; |
417 |
|
|
if (err) { |
418 |
|
|
ff_vaapi_decode_cancel(avctx, &pic->pic); |
419 |
|
|
return err; |
420 |
|
|
} |
421 |
|
|
} |
422 |
|
|
|
423 |
|
|
*last_slice_param = (VASliceParameterBufferHEVC) { |
424 |
|
|
.slice_data_size = size, |
425 |
|
|
.slice_data_offset = 0, |
426 |
|
|
.slice_data_flag = VA_SLICE_DATA_FLAG_ALL, |
427 |
|
|
/* Add 1 to the bits count here to account for the byte_alignment bit, which |
428 |
|
|
* always is at least one bit and not accounted for otherwise. */ |
429 |
|
|
.slice_data_byte_offset = (get_bits_count(&h->HEVClc->gb) + 1 + 7) / 8, |
430 |
|
|
.slice_segment_address = sh->slice_segment_addr, |
431 |
|
|
.slice_qp_delta = sh->slice_qp_delta, |
432 |
|
|
.slice_cb_qp_offset = sh->slice_cb_qp_offset, |
433 |
|
|
.slice_cr_qp_offset = sh->slice_cr_qp_offset, |
434 |
|
|
.slice_beta_offset_div2 = sh->beta_offset / 2, |
435 |
|
|
.slice_tc_offset_div2 = sh->tc_offset / 2, |
436 |
|
|
.collocated_ref_idx = sh->slice_temporal_mvp_enabled_flag ? sh->collocated_ref_idx : 0xFF, |
437 |
|
|
.five_minus_max_num_merge_cand = sh->slice_type == HEVC_SLICE_I ? 0 : 5 - sh->max_num_merge_cand, |
438 |
|
|
.num_ref_idx_l0_active_minus1 = sh->nb_refs[L0] ? sh->nb_refs[L0] - 1 : 0, |
439 |
|
|
.num_ref_idx_l1_active_minus1 = sh->nb_refs[L1] ? sh->nb_refs[L1] - 1 : 0, |
440 |
|
|
|
441 |
|
|
.LongSliceFlags.fields = { |
442 |
|
|
.dependent_slice_segment_flag = sh->dependent_slice_segment_flag, |
443 |
|
|
.slice_type = sh->slice_type, |
444 |
|
|
.color_plane_id = sh->colour_plane_id, |
445 |
|
|
.mvd_l1_zero_flag = sh->mvd_l1_zero_flag, |
446 |
|
|
.cabac_init_flag = sh->cabac_init_flag, |
447 |
|
|
.slice_temporal_mvp_enabled_flag = sh->slice_temporal_mvp_enabled_flag, |
448 |
|
|
.slice_deblocking_filter_disabled_flag = sh->disable_deblocking_filter_flag, |
449 |
|
|
.collocated_from_l0_flag = sh->collocated_list == L0 ? 1 : 0, |
450 |
|
|
.slice_loop_filter_across_slices_enabled_flag = sh->slice_loop_filter_across_slices_enabled_flag, |
451 |
|
|
.slice_sao_luma_flag = sh->slice_sample_adaptive_offset_flag[0], |
452 |
|
|
.slice_sao_chroma_flag = sh->slice_sample_adaptive_offset_flag[1], |
453 |
|
|
}, |
454 |
|
|
}; |
455 |
|
|
|
456 |
|
|
memset(last_slice_param->RefPicList, 0xFF, sizeof(last_slice_param->RefPicList)); |
457 |
|
|
|
458 |
|
|
for (list_idx = 0; list_idx < nb_list; list_idx++) { |
459 |
|
|
RefPicList *rpl = &h->ref->refPicList[list_idx]; |
460 |
|
|
|
461 |
|
|
for (i = 0; i < rpl->nb_refs; i++) |
462 |
|
|
last_slice_param->RefPicList[list_idx][i] = get_ref_pic_index(h, rpl->ref[i]); |
463 |
|
|
} |
464 |
|
|
|
465 |
|
|
fill_pred_weight_table(h, sh, last_slice_param); |
466 |
|
|
|
467 |
|
|
#if VA_CHECK_VERSION(1, 2, 0) |
468 |
|
|
if (avctx->profile == FF_PROFILE_HEVC_REXT) { |
469 |
|
|
pic->last_slice_param.rext = (VASliceParameterBufferHEVCRext) { |
470 |
|
|
.slice_ext_flags.bits = { |
471 |
|
|
.cu_chroma_qp_offset_enabled_flag = sh->cu_chroma_qp_offset_enabled_flag, |
472 |
|
|
}, |
473 |
|
|
}; |
474 |
|
|
|
475 |
|
|
memcpy(pic->last_slice_param.rext.luma_offset_l0, pic->last_slice_param.base.luma_offset_l0, |
476 |
|
|
sizeof(pic->last_slice_param.base.luma_offset_l0)); |
477 |
|
|
memcpy(pic->last_slice_param.rext.luma_offset_l1, pic->last_slice_param.base.luma_offset_l1, |
478 |
|
|
sizeof(pic->last_slice_param.base.luma_offset_l1)); |
479 |
|
|
memcpy(pic->last_slice_param.rext.ChromaOffsetL0, pic->last_slice_param.base.ChromaOffsetL0, |
480 |
|
|
sizeof(pic->last_slice_param.base.ChromaOffsetL0)); |
481 |
|
|
memcpy(pic->last_slice_param.rext.ChromaOffsetL1, pic->last_slice_param.base.ChromaOffsetL1, |
482 |
|
|
sizeof(pic->last_slice_param.base.ChromaOffsetL1)); |
483 |
|
|
} |
484 |
|
|
#endif |
485 |
|
|
|
486 |
|
|
pic->last_buffer = buffer; |
487 |
|
|
pic->last_size = size; |
488 |
|
|
|
489 |
|
|
return 0; |
490 |
|
|
} |
491 |
|
|
|
492 |
|
|
static int ptl_convert(const PTLCommon *general_ptl, H265RawProfileTierLevel *h265_raw_ptl) |
493 |
|
|
{ |
494 |
|
|
h265_raw_ptl->general_profile_space = general_ptl->profile_space; |
495 |
|
|
h265_raw_ptl->general_tier_flag = general_ptl->tier_flag; |
496 |
|
|
h265_raw_ptl->general_profile_idc = general_ptl->profile_idc; |
497 |
|
|
|
498 |
|
|
memcpy(h265_raw_ptl->general_profile_compatibility_flag, |
499 |
|
|
general_ptl->profile_compatibility_flag, 32 * sizeof(uint8_t)); |
500 |
|
|
|
501 |
|
|
#define copy_field(name) h265_raw_ptl->general_ ## name = general_ptl->name |
502 |
|
|
copy_field(progressive_source_flag); |
503 |
|
|
copy_field(interlaced_source_flag); |
504 |
|
|
copy_field(non_packed_constraint_flag); |
505 |
|
|
copy_field(frame_only_constraint_flag); |
506 |
|
|
copy_field(max_12bit_constraint_flag); |
507 |
|
|
copy_field(max_10bit_constraint_flag); |
508 |
|
|
copy_field(max_8bit_constraint_flag); |
509 |
|
|
copy_field(max_422chroma_constraint_flag); |
510 |
|
|
copy_field(max_420chroma_constraint_flag); |
511 |
|
|
copy_field(max_monochrome_constraint_flag); |
512 |
|
|
copy_field(intra_constraint_flag); |
513 |
|
|
copy_field(one_picture_only_constraint_flag); |
514 |
|
|
copy_field(lower_bit_rate_constraint_flag); |
515 |
|
|
copy_field(max_14bit_constraint_flag); |
516 |
|
|
copy_field(inbld_flag); |
517 |
|
|
copy_field(level_idc); |
518 |
|
|
#undef copy_field |
519 |
|
|
|
520 |
|
|
return 0; |
521 |
|
|
} |
522 |
|
|
|
523 |
|
|
/* |
524 |
|
|
* Find exact va_profile for HEVC Range Extension |
525 |
|
|
*/ |
526 |
|
|
VAProfile ff_vaapi_parse_hevc_rext_profile(AVCodecContext *avctx) |
527 |
|
|
{ |
528 |
|
|
const HEVCContext *h = avctx->priv_data; |
529 |
|
|
const HEVCSPS *sps = h->ps.sps; |
530 |
|
|
const PTL *ptl = &sps->ptl; |
531 |
|
|
const PTLCommon *general_ptl = &ptl->general_ptl; |
532 |
|
|
const H265ProfileDescriptor *profile; |
533 |
|
|
H265RawProfileTierLevel h265_raw_ptl = {0}; |
534 |
|
|
|
535 |
|
|
/* convert PTLCommon to H265RawProfileTierLevel */ |
536 |
|
|
ptl_convert(general_ptl, &h265_raw_ptl); |
537 |
|
|
|
538 |
|
|
profile = ff_h265_get_profile(&h265_raw_ptl); |
539 |
|
|
if (!profile) { |
540 |
|
|
av_log(avctx, AV_LOG_WARNING, "HEVC profile is not found.\n"); |
541 |
|
|
goto end; |
542 |
|
|
} else { |
543 |
|
|
av_log(avctx, AV_LOG_VERBOSE, "HEVC profile %s is found.\n", profile->name); |
544 |
|
|
} |
545 |
|
|
|
546 |
|
|
#if VA_CHECK_VERSION(1, 2, 0) |
547 |
|
|
if (!strcmp(profile->name, "Main 4:2:2 10") || |
548 |
|
|
!strcmp(profile->name, "Main 4:2:2 10 Intra")) |
549 |
|
|
return VAProfileHEVCMain422_10; |
550 |
|
|
else if (!strcmp(profile->name, "Main 4:4:4") || |
551 |
|
|
!strcmp(profile->name, "Main 4:4:4 Intra")) |
552 |
|
|
return VAProfileHEVCMain444; |
553 |
|
|
else if (!strcmp(profile->name, "Main 4:4:4 10") || |
554 |
|
|
!strcmp(profile->name, "Main 4:4:4 10 Intra")) |
555 |
|
|
return VAProfileHEVCMain444_10; |
556 |
|
|
#else |
557 |
|
|
av_log(avctx, AV_LOG_WARNING, "HEVC profile %s is " |
558 |
|
|
"not supported with this VA version.\n", profile->name); |
559 |
|
|
#endif |
560 |
|
|
|
561 |
|
|
end: |
562 |
|
|
if (avctx->hwaccel_flags & AV_HWACCEL_FLAG_ALLOW_PROFILE_MISMATCH) { |
563 |
|
|
// Default to selecting Main profile if profile mismatch is allowed |
564 |
|
|
return VAProfileHEVCMain; |
565 |
|
|
} else |
566 |
|
|
return VAProfileNone; |
567 |
|
|
} |
568 |
|
|
|
569 |
|
|
const AVHWAccel ff_hevc_vaapi_hwaccel = { |
570 |
|
|
.name = "hevc_vaapi", |
571 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
572 |
|
|
.id = AV_CODEC_ID_HEVC, |
573 |
|
|
.pix_fmt = AV_PIX_FMT_VAAPI, |
574 |
|
|
.start_frame = vaapi_hevc_start_frame, |
575 |
|
|
.end_frame = vaapi_hevc_end_frame, |
576 |
|
|
.decode_slice = vaapi_hevc_decode_slice, |
577 |
|
|
.frame_priv_data_size = sizeof(VAAPIDecodePictureHEVC), |
578 |
|
|
.init = ff_vaapi_decode_init, |
579 |
|
|
.uninit = ff_vaapi_decode_uninit, |
580 |
|
|
.frame_params = ff_vaapi_common_frame_params, |
581 |
|
|
.priv_data_size = sizeof(VAAPIDecodeContext), |
582 |
|
|
.caps_internal = HWACCEL_CAP_ASYNC_SAFE, |
583 |
|
|
}; |