Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
Timo Rothenpieler authored on 2015/08/26 21:55:40... | ... |
@@ -2419,6 +2419,8 @@ hevc_d3d11va_hwaccel_select="hevc_decoder" |
2419 | 2419 |
hevc_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_HEVC" |
2420 | 2420 |
hevc_dxva2_hwaccel_select="hevc_decoder" |
2421 | 2421 |
hevc_qsv_hwaccel_deps="libmfx" |
2422 |
+hevc_vaapi_hwaccel_deps="vaapi VAPictureParameterBufferHEVC" |
|
2423 |
+hevc_vaapi_hwaccel_select="hevc_decoder" |
|
2422 | 2424 |
hevc_vdpau_hwaccel_deps="vdpau VdpPictureInfoHEVC" |
2423 | 2425 |
hevc_vdpau_hwaccel_select="hevc_decoder" |
2424 | 2426 |
mpeg_vdpau_decoder_deps="vdpau" |
... | ... |
@@ -5148,6 +5150,8 @@ check_type "windows.h dxva.h" "DXVA_PicParams_HEVC" -DWINAPI_FAMILY=WINAPI_FAMIL |
5148 | 5148 |
check_type "windows.h d3d11.h" "ID3D11VideoDecoder" |
5149 | 5149 |
check_type "d3d9.h dxva2api.h" DXVA2_ConfigPictureDecode -D_WIN32_WINNT=0x0602 |
5150 | 5150 |
|
5151 |
+check_type "va/va.h" "VAPictureParameterBufferHEVC" |
|
5152 |
+ |
|
5151 | 5153 |
check_type "vdpau/vdpau.h" "VdpPictureInfoHEVC" |
5152 | 5154 |
|
5153 | 5155 |
if ! disabled w32threads && ! enabled pthreads; then |
... | ... |
@@ -715,6 +715,7 @@ OBJS-$(CONFIG_H264_VDPAU_HWACCEL) += vdpau_h264.o |
715 | 715 |
OBJS-$(CONFIG_H264_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o |
716 | 716 |
OBJS-$(CONFIG_HEVC_D3D11VA_HWACCEL) += dxva2_hevc.o |
717 | 717 |
OBJS-$(CONFIG_HEVC_DXVA2_HWACCEL) += dxva2_hevc.o |
718 |
+OBJS-$(CONFIG_HEVC_VAAPI_HWACCEL) += vaapi_hevc.o |
|
718 | 719 |
OBJS-$(CONFIG_HEVC_VDPAU_HWACCEL) += vdpau_hevc.o |
719 | 720 |
OBJS-$(CONFIG_MPEG1_VDPAU_HWACCEL) += vdpau_mpeg12.o |
720 | 721 |
OBJS-$(CONFIG_MPEG1_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o |
... | ... |
@@ -89,6 +89,7 @@ void avcodec_register_all(void) |
89 | 89 |
REGISTER_HWACCEL(HEVC_D3D11VA, hevc_d3d11va); |
90 | 90 |
REGISTER_HWACCEL(HEVC_DXVA2, hevc_dxva2); |
91 | 91 |
REGISTER_HWACCEL(HEVC_QSV, hevc_qsv); |
92 |
+ REGISTER_HWACCEL(HEVC_VAAPI, hevc_vaapi); |
|
92 | 93 |
REGISTER_HWACCEL(HEVC_VDPAU, hevc_vdpau); |
93 | 94 |
REGISTER_HWACCEL(MPEG1_XVMC, mpeg1_xvmc); |
94 | 95 |
REGISTER_HWACCEL(MPEG1_VDPAU, mpeg1_vdpau); |
... | ... |
@@ -328,7 +328,7 @@ static void export_stream_params(AVCodecContext *avctx, const HEVCParamSets *ps, |
328 | 328 |
|
329 | 329 |
static int set_sps(HEVCContext *s, const HEVCSPS *sps, enum AVPixelFormat pix_fmt) |
330 | 330 |
{ |
331 |
- #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL) |
|
331 |
+ #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL + CONFIG_HEVC_VAAPI_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL) |
|
332 | 332 |
enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts; |
333 | 333 |
int ret, i; |
334 | 334 |
|
... | ... |
@@ -352,6 +352,9 @@ static int set_sps(HEVCContext *s, const HEVCSPS *sps, enum AVPixelFormat pix_fm |
352 | 352 |
#if CONFIG_HEVC_D3D11VA_HWACCEL |
353 | 353 |
*fmt++ = AV_PIX_FMT_D3D11VA_VLD; |
354 | 354 |
#endif |
355 |
+#if CONFIG_HEVC_VAAPI_HWACCEL |
|
356 |
+ *fmt++ = AV_PIX_FMT_VAAPI; |
|
357 |
+#endif |
|
355 | 358 |
#if CONFIG_HEVC_VDPAU_HWACCEL |
356 | 359 |
*fmt++ = AV_PIX_FMT_VDPAU; |
357 | 360 |
#endif |
358 | 361 |
new file mode 100644 |
... | ... |
@@ -0,0 +1,490 @@ |
0 |
+/* |
|
1 |
+ * HEVC HW decode acceleration through VA API |
|
2 |
+ * |
|
3 |
+ * Copyright (C) 2015 Timo Rothenpieler <timo@rothenpieler.org> |
|
4 |
+ * |
|
5 |
+ * This file is part of FFmpeg. |
|
6 |
+ * |
|
7 |
+ * FFmpeg is free software; you can redistribute it and/or |
|
8 |
+ * modify it under the terms of the GNU Lesser General Public |
|
9 |
+ * License as published by the Free Software Foundation; either |
|
10 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
11 |
+ * |
|
12 |
+ * FFmpeg is distributed in the hope that it will be useful, |
|
13 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
14 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
15 |
+ * Lesser General Public License for more details. |
|
16 |
+ * |
|
17 |
+ * You should have received a copy of the GNU Lesser General Public |
|
18 |
+ * License along with FFmpeg; if not, write to the Free Software |
|
19 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
20 |
+ */ |
|
21 |
+ |
|
22 |
+#include "vaapi_internal.h" |
|
23 |
+#include "hevc.h" |
|
24 |
+#include "mpegutils.h" |
|
25 |
+ |
|
26 |
+/** |
|
27 |
+ * @file |
|
28 |
+ * This file implements the glue code between FFmpeg's and VA API's |
|
29 |
+ * structures for HEVC decoding. |
|
30 |
+ */ |
|
31 |
+ |
|
32 |
+typedef struct vaapi_hevc_frame_data { |
|
33 |
+ VAPictureParameterBufferHEVC *pic_param; |
|
34 |
+ VASliceParameterBufferHEVC *last_slice_param; |
|
35 |
+} vaapi_hevc_frame_data; |
|
36 |
+ |
|
37 |
+/** |
|
38 |
+ * Initialize an empty VA API picture. |
|
39 |
+ * |
|
40 |
+ * VA API requires a fixed-size reference picture array. |
|
41 |
+ */ |
|
42 |
+static void init_vaapi_pic(VAPictureHEVC *va_pic) |
|
43 |
+{ |
|
44 |
+ va_pic->picture_id = VA_INVALID_ID; |
|
45 |
+ va_pic->flags = VA_PICTURE_HEVC_INVALID; |
|
46 |
+ va_pic->pic_order_cnt = 0; |
|
47 |
+} |
|
48 |
+ |
|
49 |
+static void fill_vaapi_pic(VAPictureHEVC *va_pic, const HEVCFrame *pic, int rps_type) |
|
50 |
+{ |
|
51 |
+ va_pic->picture_id = ff_vaapi_get_surface_id(pic->frame); |
|
52 |
+ va_pic->pic_order_cnt = pic->poc; |
|
53 |
+ va_pic->flags = rps_type; |
|
54 |
+ |
|
55 |
+ if (pic->flags & HEVC_FRAME_FLAG_LONG_REF) |
|
56 |
+ va_pic->flags |= VA_PICTURE_HEVC_LONG_TERM_REFERENCE; |
|
57 |
+ |
|
58 |
+ if (pic->frame->interlaced_frame) { |
|
59 |
+ va_pic->flags |= VA_PICTURE_HEVC_FIELD_PIC; |
|
60 |
+ |
|
61 |
+ if (!pic->frame->top_field_first) { |
|
62 |
+ va_pic->flags |= VA_PICTURE_HEVC_BOTTOM_FIELD; |
|
63 |
+ } |
|
64 |
+ } |
|
65 |
+} |
|
66 |
+ |
|
67 |
+static int find_frame_rps_type(const HEVCContext *h, const HEVCFrame *pic) |
|
68 |
+{ |
|
69 |
+ VASurfaceID pic_surf = ff_vaapi_get_surface_id(pic->frame); |
|
70 |
+ int i; |
|
71 |
+ |
|
72 |
+ for (i = 0; i < h->rps[ST_CURR_BEF].nb_refs; ++i) { |
|
73 |
+ if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_BEF].ref[i]->frame)) |
|
74 |
+ return VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE; |
|
75 |
+ } |
|
76 |
+ |
|
77 |
+ for (i = 0; i < h->rps[ST_CURR_AFT].nb_refs; ++i) { |
|
78 |
+ if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_AFT].ref[i]->frame)) |
|
79 |
+ return VA_PICTURE_HEVC_RPS_ST_CURR_AFTER; |
|
80 |
+ } |
|
81 |
+ |
|
82 |
+ for (i = 0; i < h->rps[LT_CURR].nb_refs; ++i) { |
|
83 |
+ if (pic_surf == ff_vaapi_get_surface_id(h->rps[LT_CURR].ref[i]->frame)) |
|
84 |
+ return VA_PICTURE_HEVC_RPS_LT_CURR; |
|
85 |
+ } |
|
86 |
+ |
|
87 |
+ return 0; |
|
88 |
+} |
|
89 |
+ |
|
90 |
+static void fill_vaapi_ReferenceFrames(const HEVCContext *h, VAPictureParameterBufferHEVC *pp) |
|
91 |
+{ |
|
92 |
+ const HEVCFrame *current_picture = h->ref; |
|
93 |
+ int i, j, rps_type; |
|
94 |
+ |
|
95 |
+ for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); i++) { |
|
96 |
+ const HEVCFrame *frame = NULL; |
|
97 |
+ |
|
98 |
+ while (!frame && j < FF_ARRAY_ELEMS(h->DPB)) { |
|
99 |
+ if (&h->DPB[j] != current_picture && (h->DPB[j].flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF))) |
|
100 |
+ frame = &h->DPB[j]; |
|
101 |
+ j++; |
|
102 |
+ } |
|
103 |
+ |
|
104 |
+ init_vaapi_pic(&pp->ReferenceFrames[i]); |
|
105 |
+ |
|
106 |
+ if (frame) { |
|
107 |
+ rps_type = find_frame_rps_type(h, frame); |
|
108 |
+ fill_vaapi_pic(&pp->ReferenceFrames[i], frame, rps_type); |
|
109 |
+ } |
|
110 |
+ } |
|
111 |
+} |
|
112 |
+ |
|
113 |
+static uint8_t get_ref_pic_index(const HEVCContext *h, const HEVCFrame *frame) |
|
114 |
+{ |
|
115 |
+ vaapi_hevc_frame_data *frame_data = h->ref->hwaccel_picture_private; |
|
116 |
+ VAPictureParameterBufferHEVC *pp = frame_data->pic_param; |
|
117 |
+ uint8_t i; |
|
118 |
+ |
|
119 |
+ if (!frame) |
|
120 |
+ return 0xff; |
|
121 |
+ |
|
122 |
+ for (i = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); ++i) { |
|
123 |
+ VASurfaceID pid = pp->ReferenceFrames[i].picture_id; |
|
124 |
+ int poc = pp->ReferenceFrames[i].pic_order_cnt; |
|
125 |
+ if (pid != VA_INVALID_ID && pid == ff_vaapi_get_surface_id(frame->frame) && poc == frame->poc) |
|
126 |
+ return i; |
|
127 |
+ } |
|
128 |
+ |
|
129 |
+ return 0xff; |
|
130 |
+} |
|
131 |
+ |
|
132 |
+static void fill_picture_parameters(const HEVCContext *h, VAPictureParameterBufferHEVC *pp) |
|
133 |
+{ |
|
134 |
+ int i; |
|
135 |
+ |
|
136 |
+ pp->pic_fields.value = 0; |
|
137 |
+ pp->slice_parsing_fields.value = 0; |
|
138 |
+ |
|
139 |
+ fill_vaapi_pic(&pp->CurrPic, h->ref, 0); |
|
140 |
+ fill_vaapi_ReferenceFrames(h, pp); |
|
141 |
+ |
|
142 |
+ pp->pic_width_in_luma_samples = h->ps.sps->width; |
|
143 |
+ pp->pic_height_in_luma_samples = h->ps.sps->height; |
|
144 |
+ |
|
145 |
+ pp->log2_min_luma_coding_block_size_minus3 = h->ps.sps->log2_min_cb_size - 3; |
|
146 |
+ |
|
147 |
+ pp->pic_fields.bits.chroma_format_idc = h->ps.sps->chroma_format_idc; |
|
148 |
+ |
|
149 |
+ pp->sps_max_dec_pic_buffering_minus1 = h->ps.sps->temporal_layer[h->ps.sps->max_sub_layers - 1].max_dec_pic_buffering - 1; |
|
150 |
+ pp->log2_diff_max_min_luma_coding_block_size = h->ps.sps->log2_diff_max_min_coding_block_size; |
|
151 |
+ pp->log2_min_transform_block_size_minus2 = h->ps.sps->log2_min_tb_size - 2; |
|
152 |
+ pp->log2_diff_max_min_transform_block_size = h->ps.sps->log2_max_trafo_size - h->ps.sps->log2_min_tb_size; |
|
153 |
+ pp->max_transform_hierarchy_depth_inter = h->ps.sps->max_transform_hierarchy_depth_inter; |
|
154 |
+ pp->max_transform_hierarchy_depth_intra = h->ps.sps->max_transform_hierarchy_depth_intra; |
|
155 |
+ pp->num_short_term_ref_pic_sets = h->ps.sps->nb_st_rps; |
|
156 |
+ pp->num_long_term_ref_pic_sps = h->ps.sps->num_long_term_ref_pics_sps; |
|
157 |
+ |
|
158 |
+ pp->num_ref_idx_l0_default_active_minus1 = h->ps.pps->num_ref_idx_l0_default_active - 1; |
|
159 |
+ pp->num_ref_idx_l1_default_active_minus1 = h->ps.pps->num_ref_idx_l1_default_active - 1; |
|
160 |
+ pp->init_qp_minus26 = h->ps.pps->pic_init_qp_minus26; |
|
161 |
+ |
|
162 |
+ pp->pps_cb_qp_offset = h->ps.pps->cb_qp_offset; |
|
163 |
+ pp->pps_cr_qp_offset = h->ps.pps->cr_qp_offset; |
|
164 |
+ |
|
165 |
+ pp->pic_fields.bits.tiles_enabled_flag = h->ps.pps->tiles_enabled_flag; |
|
166 |
+ pp->pic_fields.bits.separate_colour_plane_flag = h->ps.sps->separate_colour_plane_flag; |
|
167 |
+ pp->pic_fields.bits.pcm_enabled_flag = h->ps.sps->pcm_enabled_flag; |
|
168 |
+ pp->pic_fields.bits.scaling_list_enabled_flag = h->ps.sps->scaling_list_enable_flag; |
|
169 |
+ pp->pic_fields.bits.transform_skip_enabled_flag = h->ps.pps->transform_skip_enabled_flag; |
|
170 |
+ pp->pic_fields.bits.amp_enabled_flag = h->ps.sps->amp_enabled_flag; |
|
171 |
+ pp->pic_fields.bits.strong_intra_smoothing_enabled_flag = h->ps.sps->sps_strong_intra_smoothing_enable_flag; |
|
172 |
+ pp->pic_fields.bits.sign_data_hiding_enabled_flag = h->ps.pps->sign_data_hiding_flag; |
|
173 |
+ pp->pic_fields.bits.constrained_intra_pred_flag = h->ps.pps->constrained_intra_pred_flag; |
|
174 |
+ pp->pic_fields.bits.cu_qp_delta_enabled_flag = h->ps.pps->cu_qp_delta_enabled_flag; |
|
175 |
+ pp->pic_fields.bits.weighted_pred_flag = h->ps.pps->weighted_pred_flag; |
|
176 |
+ pp->pic_fields.bits.weighted_bipred_flag = h->ps.pps->weighted_bipred_flag; |
|
177 |
+ pp->pic_fields.bits.transquant_bypass_enabled_flag = h->ps.pps->transquant_bypass_enable_flag; |
|
178 |
+ pp->pic_fields.bits.entropy_coding_sync_enabled_flag = h->ps.pps->entropy_coding_sync_enabled_flag; |
|
179 |
+ pp->pic_fields.bits.pps_loop_filter_across_slices_enabled_flag = h->ps.pps->seq_loop_filter_across_slices_enabled_flag; |
|
180 |
+ pp->pic_fields.bits.loop_filter_across_tiles_enabled_flag = h->ps.pps->loop_filter_across_tiles_enabled_flag; |
|
181 |
+ |
|
182 |
+ pp->pic_fields.bits.pcm_loop_filter_disabled_flag = h->ps.sps->pcm.loop_filter_disable_flag; |
|
183 |
+ pp->pcm_sample_bit_depth_luma_minus1 = h->ps.sps->pcm.bit_depth - 1; |
|
184 |
+ pp->pcm_sample_bit_depth_chroma_minus1 = h->ps.sps->pcm.bit_depth_chroma - 1; |
|
185 |
+ pp->log2_min_pcm_luma_coding_block_size_minus3 = h->ps.sps->pcm.log2_min_pcm_cb_size - 3; |
|
186 |
+ pp->log2_diff_max_min_pcm_luma_coding_block_size = h->ps.sps->pcm.log2_max_pcm_cb_size - h->ps.sps->pcm.log2_min_pcm_cb_size; |
|
187 |
+ |
|
188 |
+ memset(pp->column_width_minus1, 0, sizeof(pp->column_width_minus1)); |
|
189 |
+ memset(pp->row_height_minus1, 0, sizeof(pp->row_height_minus1)); |
|
190 |
+ |
|
191 |
+ if (h->ps.pps->tiles_enabled_flag) { |
|
192 |
+ pp->num_tile_columns_minus1 = h->ps.pps->num_tile_columns - 1; |
|
193 |
+ pp->num_tile_rows_minus1 = h->ps.pps->num_tile_rows - 1; |
|
194 |
+ |
|
195 |
+ for (i = 0; i < h->ps.pps->num_tile_columns; i++) |
|
196 |
+ pp->column_width_minus1[i] = h->ps.pps->column_width[i] - 1; |
|
197 |
+ |
|
198 |
+ for (i = 0; i < h->ps.pps->num_tile_rows; i++) |
|
199 |
+ pp->row_height_minus1[i] = h->ps.pps->row_height[i] - 1; |
|
200 |
+ } |
|
201 |
+ |
|
202 |
+ pp->diff_cu_qp_delta_depth = h->ps.pps->diff_cu_qp_delta_depth; |
|
203 |
+ pp->pps_beta_offset_div2 = h->ps.pps->beta_offset / 2; |
|
204 |
+ pp->pps_tc_offset_div2 = h->ps.pps->tc_offset / 2; |
|
205 |
+ pp->log2_parallel_merge_level_minus2 = h->ps.pps->log2_parallel_merge_level - 2; |
|
206 |
+ |
|
207 |
+ /* Diffrent chroma/luma bit depths are currently not supported by ffmpeg. */ |
|
208 |
+ pp->bit_depth_luma_minus8 = h->ps.sps->bit_depth - 8; |
|
209 |
+ pp->bit_depth_chroma_minus8 = h->ps.sps->bit_depth - 8; |
|
210 |
+ |
|
211 |
+ pp->slice_parsing_fields.bits.lists_modification_present_flag = h->ps.pps->lists_modification_present_flag; |
|
212 |
+ pp->slice_parsing_fields.bits.long_term_ref_pics_present_flag = h->ps.sps->long_term_ref_pics_present_flag; |
|
213 |
+ pp->slice_parsing_fields.bits.sps_temporal_mvp_enabled_flag = h->ps.sps->sps_temporal_mvp_enabled_flag; |
|
214 |
+ pp->slice_parsing_fields.bits.cabac_init_present_flag = h->ps.pps->cabac_init_present_flag; |
|
215 |
+ pp->slice_parsing_fields.bits.output_flag_present_flag = h->ps.pps->output_flag_present_flag; |
|
216 |
+ pp->slice_parsing_fields.bits.dependent_slice_segments_enabled_flag = h->ps.pps->dependent_slice_segments_enabled_flag; |
|
217 |
+ pp->slice_parsing_fields.bits.pps_slice_chroma_qp_offsets_present_flag = h->ps.pps->pic_slice_level_chroma_qp_offsets_present_flag; |
|
218 |
+ pp->slice_parsing_fields.bits.sample_adaptive_offset_enabled_flag = h->ps.sps->sao_enabled; |
|
219 |
+ pp->slice_parsing_fields.bits.deblocking_filter_override_enabled_flag = h->ps.pps->deblocking_filter_override_enabled_flag; |
|
220 |
+ pp->slice_parsing_fields.bits.pps_disable_deblocking_filter_flag = h->ps.pps->disable_dbf; |
|
221 |
+ pp->slice_parsing_fields.bits.slice_segment_header_extension_present_flag = h->ps.pps->slice_header_extension_present_flag; |
|
222 |
+ |
|
223 |
+ pp->log2_max_pic_order_cnt_lsb_minus4 = h->ps.sps->log2_max_poc_lsb - 4; |
|
224 |
+ pp->num_extra_slice_header_bits = h->ps.pps->num_extra_slice_header_bits; |
|
225 |
+ |
|
226 |
+ if (h->nal_unit_type >= NAL_BLA_W_LP && h->nal_unit_type <= NAL_CRA_NUT) { |
|
227 |
+ pp->slice_parsing_fields.bits.RapPicFlag = 1; |
|
228 |
+ } else { |
|
229 |
+ pp->slice_parsing_fields.bits.RapPicFlag = 0; |
|
230 |
+ } |
|
231 |
+ |
|
232 |
+ if (IS_IDR(h)) { |
|
233 |
+ pp->slice_parsing_fields.bits.IdrPicFlag = 1; |
|
234 |
+ } else { |
|
235 |
+ pp->slice_parsing_fields.bits.IdrPicFlag = 0; |
|
236 |
+ } |
|
237 |
+ |
|
238 |
+ if (IS_IRAP(h)) { |
|
239 |
+ pp->slice_parsing_fields.bits.IntraPicFlag = 1; |
|
240 |
+ } else { |
|
241 |
+ pp->slice_parsing_fields.bits.IntraPicFlag = 0; |
|
242 |
+ } |
|
243 |
+ |
|
244 |
+ if (h->sh.short_term_ref_pic_set_sps_flag == 0 && h->sh.short_term_rps) { |
|
245 |
+ pp->st_rps_bits = h->sh.short_term_ref_pic_set_size; |
|
246 |
+ } else { |
|
247 |
+ pp->st_rps_bits = 0; |
|
248 |
+ } |
|
249 |
+ |
|
250 |
+ /* TODO */ |
|
251 |
+ pp->pic_fields.bits.NoPicReorderingFlag = 0; |
|
252 |
+ pp->pic_fields.bits.NoBiPredFlag = 0; |
|
253 |
+} |
|
254 |
+ |
|
255 |
+ |
|
256 |
+/** Initialize and start decoding a frame with VA API. */ |
|
257 |
+static int vaapi_hevc_start_frame(AVCodecContext *avctx, |
|
258 |
+ av_unused const uint8_t *buffer, |
|
259 |
+ av_unused uint32_t size) |
|
260 |
+{ |
|
261 |
+ HEVCContext * const h = avctx->priv_data; |
|
262 |
+ FFVAContext * const vactx = ff_vaapi_get_context(avctx); |
|
263 |
+ vaapi_hevc_frame_data *frame_data = h->ref->hwaccel_picture_private; |
|
264 |
+ VAPictureParameterBufferHEVC *pic_param; |
|
265 |
+ VAIQMatrixBufferHEVC *iq_matrix; |
|
266 |
+ ScalingList const * scaling_list; |
|
267 |
+ int i, j, pos; |
|
268 |
+ |
|
269 |
+ ff_dlog(avctx, "vaapi_hevc_start_frame()\n"); |
|
270 |
+ |
|
271 |
+ vactx->slice_param_size = sizeof(VASliceParameterBufferHEVC); |
|
272 |
+ |
|
273 |
+ /* Fill in VAPictureParameterBufferHEVC. */ |
|
274 |
+ pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferHEVC)); |
|
275 |
+ if (!pic_param) |
|
276 |
+ return -1; |
|
277 |
+ fill_picture_parameters(h, pic_param); |
|
278 |
+ frame_data->pic_param = pic_param; |
|
279 |
+ |
|
280 |
+ /* Fill in VAIQMatrixBufferHEVC. */ |
|
281 |
+ if (h->ps.pps->scaling_list_data_present_flag) { |
|
282 |
+ scaling_list = &h->ps.pps->scaling_list; |
|
283 |
+ } else if (h->ps.sps->scaling_list_enable_flag) { |
|
284 |
+ scaling_list = &h->ps.sps->scaling_list; |
|
285 |
+ } else { |
|
286 |
+ return 0; |
|
287 |
+ } |
|
288 |
+ |
|
289 |
+ iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferHEVC)); |
|
290 |
+ if (!iq_matrix) |
|
291 |
+ return -1; |
|
292 |
+ |
|
293 |
+ for (i = 0; i < 6; ++i) { |
|
294 |
+ for (j = 0; j < 16; ++j) { |
|
295 |
+ pos = 4 * ff_hevc_diag_scan4x4_y[j] + ff_hevc_diag_scan4x4_x[j]; |
|
296 |
+ iq_matrix->ScalingList4x4[i][j] = scaling_list->sl[0][i][pos]; |
|
297 |
+ } |
|
298 |
+ for (j = 0; j < 64; ++j) { |
|
299 |
+ pos = 8 * ff_hevc_diag_scan8x8_y[j] + ff_hevc_diag_scan8x8_x[j]; |
|
300 |
+ iq_matrix->ScalingList8x8[i][j] = scaling_list->sl[1][i][pos]; |
|
301 |
+ iq_matrix->ScalingList16x16[i][j] = scaling_list->sl[2][i][pos]; |
|
302 |
+ if (i < 2) { |
|
303 |
+ iq_matrix->ScalingList32x32[i][j] = scaling_list->sl[3][i * 3][pos]; |
|
304 |
+ } |
|
305 |
+ } |
|
306 |
+ iq_matrix->ScalingListDC16x16[i] = scaling_list->sl_dc[0][i]; |
|
307 |
+ if (i < 2) { |
|
308 |
+ iq_matrix->ScalingListDC32x32[i] = scaling_list->sl_dc[1][i * 3]; |
|
309 |
+ } |
|
310 |
+ } |
|
311 |
+ |
|
312 |
+ return 0; |
|
313 |
+} |
|
314 |
+ |
|
315 |
+/** End a hardware decoding based frame. */ |
|
316 |
+static int vaapi_hevc_end_frame(AVCodecContext *avctx) |
|
317 |
+{ |
|
318 |
+ FFVAContext * const vactx = ff_vaapi_get_context(avctx); |
|
319 |
+ HEVCContext * const h = avctx->priv_data; |
|
320 |
+ vaapi_hevc_frame_data *frame_data = h->ref->hwaccel_picture_private; |
|
321 |
+ int ret; |
|
322 |
+ |
|
323 |
+ ff_dlog(avctx, "vaapi_hevc_end_frame()\n"); |
|
324 |
+ |
|
325 |
+ frame_data->last_slice_param->LongSliceFlags.fields.LastSliceOfPic = 1; |
|
326 |
+ |
|
327 |
+ ret = ff_vaapi_commit_slices(vactx); |
|
328 |
+ if (ret < 0) |
|
329 |
+ goto finish; |
|
330 |
+ |
|
331 |
+ ret = ff_vaapi_render_picture(vactx, ff_vaapi_get_surface_id(h->ref->frame)); |
|
332 |
+ if (ret < 0) |
|
333 |
+ goto finish; |
|
334 |
+ |
|
335 |
+finish: |
|
336 |
+ ff_vaapi_common_end_frame(avctx); |
|
337 |
+ return ret; |
|
338 |
+} |
|
339 |
+ |
|
340 |
+static int fill_pred_weight_table(HEVCContext * const h, |
|
341 |
+ VASliceParameterBufferHEVC *slice_param, |
|
342 |
+ SliceHeader * const sh) |
|
343 |
+{ |
|
344 |
+ int i; |
|
345 |
+ |
|
346 |
+ memset(slice_param->delta_luma_weight_l0, 0, sizeof(slice_param->delta_luma_weight_l0)); |
|
347 |
+ memset(slice_param->delta_luma_weight_l1, 0, sizeof(slice_param->delta_luma_weight_l1)); |
|
348 |
+ memset(slice_param->luma_offset_l0, 0, sizeof(slice_param->luma_offset_l0)); |
|
349 |
+ memset(slice_param->luma_offset_l1, 0, sizeof(slice_param->luma_offset_l1)); |
|
350 |
+ memset(slice_param->delta_chroma_weight_l0, 0, sizeof(slice_param->delta_chroma_weight_l0)); |
|
351 |
+ memset(slice_param->delta_chroma_weight_l1, 0, sizeof(slice_param->delta_chroma_weight_l1)); |
|
352 |
+ memset(slice_param->ChromaOffsetL0, 0, sizeof(slice_param->ChromaOffsetL0)); |
|
353 |
+ memset(slice_param->ChromaOffsetL1, 0, sizeof(slice_param->ChromaOffsetL1)); |
|
354 |
+ |
|
355 |
+ slice_param->delta_chroma_log2_weight_denom = 0; |
|
356 |
+ slice_param->luma_log2_weight_denom = 0; |
|
357 |
+ |
|
358 |
+ if ( sh->slice_type == I_SLICE |
|
359 |
+ || (sh->slice_type == P_SLICE && !h->ps.pps->weighted_pred_flag) |
|
360 |
+ || (sh->slice_type == B_SLICE && !h->ps.pps->weighted_bipred_flag)) { |
|
361 |
+ return 0; |
|
362 |
+ } |
|
363 |
+ |
|
364 |
+ slice_param->luma_log2_weight_denom = sh->luma_log2_weight_denom; |
|
365 |
+ |
|
366 |
+ if (h->ps.sps->chroma_format_idc) { |
|
367 |
+ slice_param->delta_chroma_log2_weight_denom = sh->chroma_log2_weight_denom - sh->luma_log2_weight_denom; |
|
368 |
+ } |
|
369 |
+ |
|
370 |
+ for (i = 0; i < 15 && i < sh->nb_refs[L0]; ++i) { |
|
371 |
+ slice_param->delta_luma_weight_l0[i] = sh->luma_weight_l0[i] - (1 << sh->luma_log2_weight_denom); |
|
372 |
+ slice_param->luma_offset_l0[i] = sh->luma_offset_l0[i]; |
|
373 |
+ slice_param->delta_chroma_weight_l0[i][0] = sh->chroma_weight_l0[i][0] - (1 << sh->chroma_log2_weight_denom); |
|
374 |
+ slice_param->delta_chroma_weight_l0[i][1] = sh->chroma_weight_l0[i][1] - (1 << sh->chroma_log2_weight_denom); |
|
375 |
+ slice_param->ChromaOffsetL0[i][0] = sh->chroma_offset_l0[i][0]; |
|
376 |
+ slice_param->ChromaOffsetL0[i][1] = sh->chroma_offset_l0[i][1]; |
|
377 |
+ } |
|
378 |
+ |
|
379 |
+ if (sh->slice_type == B_SLICE) { |
|
380 |
+ for (i = 0; i < 15 && i < sh->nb_refs[L1]; ++i) { |
|
381 |
+ slice_param->delta_luma_weight_l1[i] = sh->luma_weight_l1[i] - (1 << sh->luma_log2_weight_denom); |
|
382 |
+ slice_param->luma_offset_l1[i] = sh->luma_offset_l1[i]; |
|
383 |
+ slice_param->delta_chroma_weight_l1[i][0] = sh->chroma_weight_l1[i][0] - (1 << sh->chroma_log2_weight_denom); |
|
384 |
+ slice_param->delta_chroma_weight_l1[i][1] = sh->chroma_weight_l1[i][1] - (1 << sh->chroma_log2_weight_denom); |
|
385 |
+ slice_param->ChromaOffsetL1[i][0] = sh->chroma_offset_l1[i][0]; |
|
386 |
+ slice_param->ChromaOffsetL1[i][1] = sh->chroma_offset_l1[i][1]; |
|
387 |
+ } |
|
388 |
+ } |
|
389 |
+ |
|
390 |
+ return 0; |
|
391 |
+} |
|
392 |
+ |
|
393 |
+/** Decode the given hevc slice with VA API. */ |
|
394 |
+static int vaapi_hevc_decode_slice(AVCodecContext *avctx, |
|
395 |
+ const uint8_t *buffer, |
|
396 |
+ uint32_t size) |
|
397 |
+{ |
|
398 |
+ FFVAContext * const vactx = ff_vaapi_get_context(avctx); |
|
399 |
+ HEVCContext * const h = avctx->priv_data; |
|
400 |
+ vaapi_hevc_frame_data *frame_data = h->ref->hwaccel_picture_private; |
|
401 |
+ SliceHeader * const sh = &h->sh; |
|
402 |
+ VASliceParameterBufferHEVC *slice_param; |
|
403 |
+ int i, list_idx; |
|
404 |
+ uint8_t nb_list = sh->slice_type == B_SLICE ? 2 : 1; |
|
405 |
+ |
|
406 |
+ if (sh->slice_type == I_SLICE) |
|
407 |
+ nb_list = 0; |
|
408 |
+ |
|
409 |
+ ff_dlog(avctx, "vaapi_hevc_decode_slice(): buffer %p, size %d\n", buffer, size); |
|
410 |
+ |
|
411 |
+ /* Fill in VASliceParameterBufferH264. */ |
|
412 |
+ slice_param = (VASliceParameterBufferHEVC *)ff_vaapi_alloc_slice(vactx, buffer, size); |
|
413 |
+ if (!slice_param) |
|
414 |
+ return -1; |
|
415 |
+ |
|
416 |
+ frame_data->last_slice_param = slice_param; |
|
417 |
+ |
|
418 |
+ /* The base structure changed, so this has to be re-set in order to be valid on every byte order. */ |
|
419 |
+ slice_param->slice_data_flag = VA_SLICE_DATA_FLAG_ALL; |
|
420 |
+ |
|
421 |
+ /* Add 1 to the bits count here to account for the byte_alignment bit, which allways is at least one bit and not accounted for otherwise. */ |
|
422 |
+ slice_param->slice_data_byte_offset = (get_bits_count(&h->HEVClc->gb) + 1 + 7) / 8; |
|
423 |
+ |
|
424 |
+ slice_param->slice_segment_address = sh->slice_segment_addr; |
|
425 |
+ |
|
426 |
+ slice_param->LongSliceFlags.value = 0; |
|
427 |
+ slice_param->LongSliceFlags.fields.dependent_slice_segment_flag = sh->dependent_slice_segment_flag; |
|
428 |
+ slice_param->LongSliceFlags.fields.slice_type = sh->slice_type; |
|
429 |
+ slice_param->LongSliceFlags.fields.color_plane_id = sh->colour_plane_id; |
|
430 |
+ slice_param->LongSliceFlags.fields.mvd_l1_zero_flag = sh->mvd_l1_zero_flag; |
|
431 |
+ slice_param->LongSliceFlags.fields.cabac_init_flag = sh->cabac_init_flag; |
|
432 |
+ slice_param->LongSliceFlags.fields.slice_temporal_mvp_enabled_flag = sh->slice_temporal_mvp_enabled_flag; |
|
433 |
+ slice_param->LongSliceFlags.fields.slice_deblocking_filter_disabled_flag = sh->disable_deblocking_filter_flag; |
|
434 |
+ slice_param->LongSliceFlags.fields.collocated_from_l0_flag = sh->collocated_list == L0 ? 1 : 0; |
|
435 |
+ slice_param->LongSliceFlags.fields.slice_loop_filter_across_slices_enabled_flag = sh->slice_loop_filter_across_slices_enabled_flag; |
|
436 |
+ |
|
437 |
+ slice_param->LongSliceFlags.fields.slice_sao_luma_flag = sh->slice_sample_adaptive_offset_flag[0]; |
|
438 |
+ if (h->ps.sps->chroma_format_idc) { |
|
439 |
+ slice_param->LongSliceFlags.fields.slice_sao_chroma_flag = sh->slice_sample_adaptive_offset_flag[1]; |
|
440 |
+ } |
|
441 |
+ |
|
442 |
+ if (sh->slice_temporal_mvp_enabled_flag) { |
|
443 |
+ slice_param->collocated_ref_idx = sh->collocated_ref_idx; |
|
444 |
+ } else { |
|
445 |
+ slice_param->collocated_ref_idx = 0xFF; |
|
446 |
+ } |
|
447 |
+ |
|
448 |
+ slice_param->slice_qp_delta = sh->slice_qp_delta; |
|
449 |
+ slice_param->slice_cb_qp_offset = sh->slice_cb_qp_offset; |
|
450 |
+ slice_param->slice_cr_qp_offset = sh->slice_cr_qp_offset; |
|
451 |
+ slice_param->slice_beta_offset_div2 = sh->beta_offset / 2; |
|
452 |
+ slice_param->slice_tc_offset_div2 = sh->tc_offset / 2; |
|
453 |
+ |
|
454 |
+ if (sh->slice_type == I_SLICE) { |
|
455 |
+ slice_param->five_minus_max_num_merge_cand = 0; |
|
456 |
+ } else { |
|
457 |
+ slice_param->five_minus_max_num_merge_cand = 5 - sh->max_num_merge_cand; |
|
458 |
+ } |
|
459 |
+ |
|
460 |
+ slice_param->num_ref_idx_l0_active_minus1 = sh->nb_refs[L0] ? sh->nb_refs[L0] - 1 : 0; |
|
461 |
+ slice_param->num_ref_idx_l1_active_minus1 = sh->nb_refs[L1] ? sh->nb_refs[L1] - 1 : 0; |
|
462 |
+ |
|
463 |
+ memset(slice_param->RefPicList, 0xFF, sizeof(slice_param->RefPicList)); |
|
464 |
+ |
|
465 |
+ /* h->ref->refPicList is updated befor calling each slice */ |
|
466 |
+ for (list_idx = 0; list_idx < nb_list; ++list_idx) { |
|
467 |
+ RefPicList *rpl = &h->ref->refPicList[list_idx]; |
|
468 |
+ |
|
469 |
+ for (i = 0; i < rpl->nb_refs; ++i) { |
|
470 |
+ slice_param->RefPicList[list_idx][i] = get_ref_pic_index(h, rpl->ref[i]); |
|
471 |
+ } |
|
472 |
+ } |
|
473 |
+ |
|
474 |
+ return fill_pred_weight_table(h, slice_param, sh); |
|
475 |
+} |
|
476 |
+ |
|
477 |
+AVHWAccel ff_hevc_vaapi_hwaccel = { |
|
478 |
+ .name = "hevc_vaapi", |
|
479 |
+ .type = AVMEDIA_TYPE_VIDEO, |
|
480 |
+ .id = AV_CODEC_ID_HEVC, |
|
481 |
+ .pix_fmt = AV_PIX_FMT_VAAPI, |
|
482 |
+ .start_frame = vaapi_hevc_start_frame, |
|
483 |
+ .end_frame = vaapi_hevc_end_frame, |
|
484 |
+ .decode_slice = vaapi_hevc_decode_slice, |
|
485 |
+ .init = ff_vaapi_context_init, |
|
486 |
+ .uninit = ff_vaapi_context_fini, |
|
487 |
+ .priv_data_size = sizeof(FFVAContext), |
|
488 |
+ .frame_priv_data_size = sizeof(vaapi_hevc_frame_data), |
|
489 |
+}; |
... | ... |
@@ -29,7 +29,7 @@ |
29 | 29 |
#include "libavutil/version.h" |
30 | 30 |
|
31 | 31 |
#define LIBAVCODEC_VERSION_MAJOR 56 |
32 |
-#define LIBAVCODEC_VERSION_MINOR 58 |
|
32 |
+#define LIBAVCODEC_VERSION_MINOR 59 |
|
33 | 33 |
#define LIBAVCODEC_VERSION_MICRO 100 |
34 | 34 |
|
35 | 35 |
#define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \ |