path: root/libavcodec/vdpau_hevc.c (plain)
blob: fcdf4b2747f26135b8d4735b2c3467b52b533b7a

1	/*
2	* MPEG-H Part 2 / HEVC / H.265 HW decode acceleration through VDPAU
3	*
4	* Copyright (c) 2013 Philip Langdale
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
25	#include "avcodec.h"
26	#include "internal.h"
27	#include "hevc_data.h"
28	#include "hevcdec.h"
29	#include "hwaccel.h"
30	#include "vdpau.h"
31	#include "vdpau_internal.h"
32
33	static int vdpau_hevc_start_frame(AVCodecContext *avctx,
34	const uint8_t *buffer, uint32_t size)
35	{
36	HEVCContext *h = avctx->priv_data;
37	HEVCFrame *pic = h->ref;
38	struct vdpau_picture_context *pic_ctx = pic->hwaccel_picture_private;
39
40	VdpPictureInfoHEVC *info = &pic_ctx->info.hevc;
41
42	const HEVCSPS *sps = h->ps.sps;
43	const HEVCPPS *pps = h->ps.pps;
44	const SliceHeader *sh = &h->sh;
45	const ScalingList *sl = pps->scaling_list_data_present_flag ?
46	&pps->scaling_list : &sps->scaling_list;
47
48	/* init VdpPictureInfoHEVC */
49
50	/* SPS */
51	info->chroma_format_idc = sps->chroma_format_idc;
52	info->separate_colour_plane_flag = sps->separate_colour_plane_flag;
53	info->pic_width_in_luma_samples = sps->width;
54	info->pic_height_in_luma_samples = sps->height;
55	info->bit_depth_luma_minus8 = sps->bit_depth - 8;
56	info->bit_depth_chroma_minus8 = sps->bit_depth - 8;
57	info->log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4;
58	/* Provide the value corresponding to the nuh_temporal_id of the frame
59	to be decoded. */
60	info->sps_max_dec_pic_buffering_minus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1;
61	info->log2_min_luma_coding_block_size_minus3 = sps->log2_min_cb_size - 3;
62	info->log2_diff_max_min_luma_coding_block_size = sps->log2_diff_max_min_coding_block_size;
63	info->log2_min_transform_block_size_minus2 = sps->log2_min_tb_size - 2;
64	info->log2_diff_max_min_transform_block_size = sps->log2_max_trafo_size - sps->log2_min_tb_size;
65	info->max_transform_hierarchy_depth_inter = sps->max_transform_hierarchy_depth_inter;
66	info->max_transform_hierarchy_depth_intra = sps->max_transform_hierarchy_depth_intra;
67	info->scaling_list_enabled_flag = sps->scaling_list_enable_flag;
68	/* Scaling lists, in diagonal order, to be used for this frame. */
69	for (size_t i = 0; i < 6; i++) {
70	for (size_t j = 0; j < 16; j++) {
71	/* Scaling List for 4x4 quantization matrix,
72	indexed as ScalingList4x4[matrixId][i]. */
73	uint8_t pos = 4 * ff_hevc_diag_scan4x4_y[j] + ff_hevc_diag_scan4x4_x[j];
74	info->ScalingList4x4[i][j] = sl->sl[0][i][pos];
75	}
76	for (size_t j = 0; j < 64; j++) {
77	uint8_t pos = 8 * ff_hevc_diag_scan8x8_y[j] + ff_hevc_diag_scan8x8_x[j];
78	/* Scaling List for 8x8 quantization matrix,
79	indexed as ScalingList8x8[matrixId][i]. */
80	info->ScalingList8x8[i][j] = sl->sl[1][i][pos];
81	/* Scaling List for 16x16 quantization matrix,
82	indexed as ScalingList16x16[matrixId][i]. */
83	info->ScalingList16x16[i][j] = sl->sl[2][i][pos];
84	if (i < 2) {
85	/* Scaling List for 32x32 quantization matrix,
86	indexed as ScalingList32x32[matrixId][i]. */
87	info->ScalingList32x32[i][j] = sl->sl[3][i * 3][pos];
88	}
89	}
90	/* Scaling List DC Coefficients for 16x16,
91	indexed as ScalingListDCCoeff16x16[matrixId]. */
92	info->ScalingListDCCoeff16x16[i] = sl->sl_dc[0][i];
93	if (i < 2) {
94	/* Scaling List DC Coefficients for 32x32,
95	indexed as ScalingListDCCoeff32x32[matrixId]. */
96	info->ScalingListDCCoeff32x32[i] = sl->sl_dc[1][i * 3];
97	}
98	}
99	info->amp_enabled_flag = sps->amp_enabled_flag;
100	info->sample_adaptive_offset_enabled_flag = sps->sao_enabled;
101	info->pcm_enabled_flag = sps->pcm_enabled_flag;
102	if (info->pcm_enabled_flag) {
103	/* Only needs to be set if pcm_enabled_flag is set. Ignored otherwise. */
104	info->pcm_sample_bit_depth_luma_minus1 = sps->pcm.bit_depth - 1;
105	/* Only needs to be set if pcm_enabled_flag is set. Ignored otherwise. */
106	info->pcm_sample_bit_depth_chroma_minus1 = sps->pcm.bit_depth_chroma - 1;
107	/* Only needs to be set if pcm_enabled_flag is set. Ignored otherwise. */
108	info->log2_min_pcm_luma_coding_block_size_minus3 = sps->pcm.log2_min_pcm_cb_size - 3;
109	/* Only needs to be set if pcm_enabled_flag is set. Ignored otherwise. */
110	info->log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size;
111	/* Only needs to be set if pcm_enabled_flag is set. Ignored otherwise. */
112	info->pcm_loop_filter_disabled_flag = sps->pcm.loop_filter_disable_flag;
113	}
114	/* Per spec, when zero, assume short_term_ref_pic_set_sps_flag
115	is also zero. */
116	info->num_short_term_ref_pic_sets = sps->nb_st_rps;
117	info->long_term_ref_pics_present_flag = sps->long_term_ref_pics_present_flag;
118	/* Only needed if long_term_ref_pics_present_flag is set. Ignored
119	otherwise. */
120	info->num_long_term_ref_pics_sps = sps->num_long_term_ref_pics_sps;
121	info->sps_temporal_mvp_enabled_flag = sps->sps_temporal_mvp_enabled_flag;
122	info->strong_intra_smoothing_enabled_flag = sps->sps_strong_intra_smoothing_enable_flag;
123
124	/* Copy the HEVC Picture Parameter Set bitstream fields. */
125	info->dependent_slice_segments_enabled_flag = pps->dependent_slice_segments_enabled_flag;
126	info->output_flag_present_flag = pps->output_flag_present_flag;
127	info->num_extra_slice_header_bits = pps->num_extra_slice_header_bits;
128	info->sign_data_hiding_enabled_flag = pps->sign_data_hiding_flag;
129	info->cabac_init_present_flag = pps->cabac_init_present_flag;
130	info->num_ref_idx_l0_default_active_minus1 = pps->num_ref_idx_l0_default_active - 1;
131	info->num_ref_idx_l1_default_active_minus1 = pps->num_ref_idx_l1_default_active - 1;
132	info->init_qp_minus26 = pps->pic_init_qp_minus26;
133	info->constrained_intra_pred_flag = pps->constrained_intra_pred_flag;
134	info->transform_skip_enabled_flag = pps->transform_skip_enabled_flag;
135	info->cu_qp_delta_enabled_flag = pps->cu_qp_delta_enabled_flag;
136	/* Only needed if cu_qp_delta_enabled_flag is set. Ignored otherwise. */
137	info->diff_cu_qp_delta_depth = pps->diff_cu_qp_delta_depth;
138	info->pps_cb_qp_offset = pps->cb_qp_offset;
139	info->pps_cr_qp_offset = pps->cr_qp_offset;
140	info->pps_slice_chroma_qp_offsets_present_flag = pps->pic_slice_level_chroma_qp_offsets_present_flag;
141	info->weighted_pred_flag = pps->weighted_pred_flag;
142	info->weighted_bipred_flag = pps->weighted_bipred_flag;
143	info->transquant_bypass_enabled_flag = pps->transquant_bypass_enable_flag;
144	info->tiles_enabled_flag = pps->tiles_enabled_flag;
145	info->entropy_coding_sync_enabled_flag = pps->entropy_coding_sync_enabled_flag;
146	if (info->tiles_enabled_flag) {
147	/* Only valid if tiles_enabled_flag is set. Ignored otherwise. */
148	info->num_tile_columns_minus1 = pps->num_tile_columns - 1;
149	/* Only valid if tiles_enabled_flag is set. Ignored otherwise. */
150	info->num_tile_rows_minus1 = pps->num_tile_rows - 1;
151	/* Only valid if tiles_enabled_flag is set. Ignored otherwise. */
152	info->uniform_spacing_flag = pps->uniform_spacing_flag;
153	/* Only need to set 0..num_tile_columns_minus1. The struct
154	definition reserves up to the maximum of 20. Invalid values are
155	ignored. */
156	for (ssize_t i = 0; i < pps->num_tile_columns; i++) {
157	info->column_width_minus1[i] = pps->column_width[i] - 1;
158	}
159	/* Only need to set 0..num_tile_rows_minus1. The struct
160	definition reserves up to the maximum of 22. Invalid values are
161	ignored.*/
162	for (ssize_t i = 0; i < pps->num_tile_rows; i++) {
163	info->row_height_minus1[i] = pps->row_height[i] - 1;
164	}
165	/* Only needed if tiles_enabled_flag is set. Invalid values are
166	ignored. */
167	info->loop_filter_across_tiles_enabled_flag = pps->loop_filter_across_tiles_enabled_flag;
168	}
169	info->pps_loop_filter_across_slices_enabled_flag = pps->seq_loop_filter_across_slices_enabled_flag;
170	info->deblocking_filter_control_present_flag = pps->deblocking_filter_control_present_flag;
171	/* Only valid if deblocking_filter_control_present_flag is set. Ignored
172	otherwise. */
173	info->deblocking_filter_override_enabled_flag = pps->deblocking_filter_override_enabled_flag;
174	/* Only valid if deblocking_filter_control_present_flag is set. Ignored
175	otherwise. */
176	info->pps_deblocking_filter_disabled_flag = pps->disable_dbf;
177	/* Only valid if deblocking_filter_control_present_flag is set and
178	pps_deblocking_filter_disabled_flag is not set. Ignored otherwise.*/
179	info->pps_beta_offset_div2 = pps->beta_offset / 2;
180	/* Only valid if deblocking_filter_control_present_flag is set and
181	pps_deblocking_filter_disabled_flag is not set. Ignored otherwise. */
182	info->pps_tc_offset_div2 = pps->tc_offset / 2;
183	info->lists_modification_present_flag = pps->lists_modification_present_flag;
184	info->log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level - 2;
185	info->slice_segment_header_extension_present_flag = pps->slice_header_extension_present_flag;
186
187	/* Set to 1 if nal_unit_type is equal to IDR_W_RADL or IDR_N_LP.
188	Set to zero otherwise. */
189	info->IDRPicFlag = IS_IDR(h);
190	/* Set to 1 if nal_unit_type in the range of BLA_W_LP to
191	RSV_IRAP_VCL23, inclusive. Set to zero otherwise.*/
192	info->RAPPicFlag = IS_IRAP(h);
193	/* See section 7.4.7.1 of the specification. */
194	info->CurrRpsIdx = sps->nb_st_rps;
195	if (sh->short_term_ref_pic_set_sps_flag == 1) {
196	for (size_t i = 0; i < sps->nb_st_rps; i++) {
197	if (sh->short_term_rps == &sps->st_rps[i]) {
198	info->CurrRpsIdx = i;
199	break;
200	}
201	}
202	}
203	/* See section 7.4.7.2 of the specification. */
204	info->NumPocTotalCurr = ff_hevc_frame_nb_refs(h);
205	if (sh->short_term_ref_pic_set_sps_flag == 0 && sh->short_term_rps) {
206	/* Corresponds to specification field, NumDeltaPocs[RefRpsIdx].
207	Only applicable when short_term_ref_pic_set_sps_flag == 0.
208	Implementations will ignore this value in other cases. See 7.4.8. */
209	info->NumDeltaPocsOfRefRpsIdx = sh->short_term_rps->rps_idx_num_delta_pocs;
210	}
211	/* Section 7.6.3.1 of the H.265/HEVC Specification defines the syntax of
212	the slice_segment_header. This header contains information that
213	some VDPAU implementations may choose to skip. The VDPAU API
214	requires client applications to track the number of bits used in the
215	slice header for structures associated with short term and long term
216	reference pictures. First, VDPAU requires the number of bits used by
217	the short_term_ref_pic_set array in the slice_segment_header. */
218	info->NumShortTermPictureSliceHeaderBits = sh->short_term_ref_pic_set_size;
219	/* Second, VDPAU requires the number of bits used for long term reference
220	pictures in the slice_segment_header. This is equal to the number
221	of bits used for the contents of the block beginning with
222	"if(long_term_ref_pics_present_flag)". */
223	info->NumLongTermPictureSliceHeaderBits = sh->long_term_ref_pic_set_size;
224
225	/* The value of PicOrderCntVal of the picture in the access unit
226	containing the SEI message. The picture being decoded. */
227	info->CurrPicOrderCntVal = h->poc;
228
229	/* Slice Decoding Process - Reference Picture Sets */
230	for (size_t i = 0; i < 16; i++) {
231	info->RefPics[i] = VDP_INVALID_HANDLE;
232	info->PicOrderCntVal[i] = 0;
233	info->IsLongTerm[i] = 0;
234	}
235	for (size_t i = 0, j = 0; i < FF_ARRAY_ELEMS(h->DPB); i++) {
236	const HEVCFrame *frame = &h->DPB[i];
237	if (frame != h->ref && (frame->flags & (HEVC_FRAME_FLAG_LONG_REF |
238	HEVC_FRAME_FLAG_SHORT_REF))) {
239	if (j > 15) {
240	av_log(avctx, AV_LOG_WARNING,
241	"VDPAU only supports up to 16 references in the DPB. "
242	"This frame may not be decoded correctly.\n");
243	break;
244	}
245	/* Array of video reference surfaces.
246	Set any unused positions to VDP_INVALID_HANDLE. */
247	info->RefPics[j] = ff_vdpau_get_surface_id(frame->frame);
248	/* Array of picture order counts. These correspond to positions
249	in the RefPics array. */
250	info->PicOrderCntVal[j] = frame->poc;
251	/* Array used to specify whether a particular RefPic is
252	a long term reference. A value of "1" indicates a long-term
253	reference. */
254	// XXX: Setting this caused glitches in the nvidia implementation
255	// Always setting it to zero, produces correct results
256	//info->IsLongTerm[j] = frame->flags & HEVC_FRAME_FLAG_LONG_REF;
257	info->IsLongTerm[j] = 0;
258	j++;
259	}
260	}
261	/* Copy of specification field, see Section 8.3.2 of the
262	H.265/HEVC Specification. */
263	info->NumPocStCurrBefore = h->rps[ST_CURR_BEF].nb_refs;
264	if (info->NumPocStCurrBefore > 8) {
265	av_log(avctx, AV_LOG_WARNING,
266	"VDPAU only supports up to 8 references in StCurrBefore. "
267	"This frame may not be decoded correctly.\n");
268	info->NumPocStCurrBefore = 8;
269	}
270	/* Copy of specification field, see Section 8.3.2 of the
271	H.265/HEVC Specification. */
272	info->NumPocStCurrAfter = h->rps[ST_CURR_AFT].nb_refs;
273	if (info->NumPocStCurrAfter > 8) {
274	av_log(avctx, AV_LOG_WARNING,
275	"VDPAU only supports up to 8 references in StCurrAfter. "
276	"This frame may not be decoded correctly.\n");
277	info->NumPocStCurrAfter = 8;
278	}
279	/* Copy of specification field, see Section 8.3.2 of the
280	H.265/HEVC Specification. */
281	info->NumPocLtCurr = h->rps[LT_CURR].nb_refs;
282	if (info->NumPocLtCurr > 8) {
283	av_log(avctx, AV_LOG_WARNING,
284	"VDPAU only supports up to 8 references in LtCurr. "
285	"This frame may not be decoded correctly.\n");
286	info->NumPocLtCurr = 8;
287	}
288	/* Reference Picture Set list, one of the short-term RPS. These
289	correspond to positions in the RefPics array. */
290	for (ssize_t i = 0, j = 0; i < h->rps[ST_CURR_BEF].nb_refs; i++) {
291	HEVCFrame *frame = h->rps[ST_CURR_BEF].ref[i];
292	if (frame) {
293	uint8_t found = 0;
294	uintptr_t id = ff_vdpau_get_surface_id(frame->frame);
295	for (size_t k = 0; k < 16; k++) {
296	if (id == info->RefPics[k]) {
297	info->RefPicSetStCurrBefore[j] = k;
298	j++;
299	found = 1;
300	break;
301	}
302	}
303	if (!found) {
304	av_log(avctx, AV_LOG_WARNING, "missing surface: %p\n",
305	(void *)id);
306	}
307	} else {
308	av_log(avctx, AV_LOG_WARNING, "missing STR Before frame: %zd\n", i);
309	}
310	}
311	/* Reference Picture Set list, one of the short-term RPS. These
312	correspond to positions in the RefPics array. */
313	for (ssize_t i = 0, j = 0; i < h->rps[ST_CURR_AFT].nb_refs; i++) {
314	HEVCFrame *frame = h->rps[ST_CURR_AFT].ref[i];
315	if (frame) {
316	uint8_t found = 0;
317	uintptr_t id = ff_vdpau_get_surface_id(frame->frame);
318	for (size_t k = 0; k < 16; k++) {
319	if (id == info->RefPics[k]) {
320	info->RefPicSetStCurrAfter[j] = k;
321	j++;
322	found = 1;
323	break;
324	}
325	}
326	if (!found) {
327	av_log(avctx, AV_LOG_WARNING, "missing surface: %p\n",
328	(void *)id);
329	}
330	} else {
331	av_log(avctx, AV_LOG_WARNING, "missing STR After frame: %zd\n", i);
332	}
333	}
334	/* Reference Picture Set list, one of the long-term RPS. These
335	correspond to positions in the RefPics array. */
336	for (ssize_t i = 0, j = 0; i < h->rps[LT_CURR].nb_refs; i++) {
337	HEVCFrame *frame = h->rps[LT_CURR].ref[i];
338	if (frame) {
339	uint8_t found = 0;
340	uintptr_t id = ff_vdpau_get_surface_id(frame->frame);
341	for (size_t k = 0; k < 16; k++) {
342	if (id == info->RefPics[k]) {
343	info->RefPicSetLtCurr[j] = k;
344	j++;
345	found = 1;
346	break;
347	}
348	}
349	if (!found) {
350	av_log(avctx, AV_LOG_WARNING, "missing surface: %p\n",
351	(void *)id);
352	}
353	} else {
354	av_log(avctx, AV_LOG_WARNING, "missing LTR frame: %zd\n", i);
355	}
356	}
357
358	return ff_vdpau_common_start_frame(pic_ctx, buffer, size);
359	}
360
361	static const uint8_t start_code_prefix[3] = { 0x00, 0x00, 0x01 };
362
363	static int vdpau_hevc_decode_slice(AVCodecContext *avctx,
364	const uint8_t *buffer, uint32_t size)
365	{
366	HEVCContext *h = avctx->priv_data;
367	struct vdpau_picture_context *pic_ctx = h->ref->hwaccel_picture_private;
368	int val;
369
370	val = ff_vdpau_add_buffer(pic_ctx, start_code_prefix, 3);
371	if (val)
372	return val;
373
374	val = ff_vdpau_add_buffer(pic_ctx, buffer, size);
375	if (val)
376	return val;
377
378	return 0;
379	}
380
381	static int vdpau_hevc_end_frame(AVCodecContext *avctx)
382	{
383	HEVCContext *h = avctx->priv_data;
384	struct vdpau_picture_context *pic_ctx = h->ref->hwaccel_picture_private;
385	int val;
386
387	val = ff_vdpau_common_end_frame(avctx, h->ref->frame, pic_ctx);
388	if (val < 0)
389	return val;
390
391	return 0;
392	}
393
394	static int vdpau_hevc_init(AVCodecContext *avctx)
395	{
396	VdpDecoderProfile profile;
397	uint32_t level = avctx->level;
398
399	switch (avctx->profile) {
400	case FF_PROFILE_HEVC_MAIN:
401	profile = VDP_DECODER_PROFILE_HEVC_MAIN;
402	break;
403	case FF_PROFILE_HEVC_MAIN_10:
404	profile = VDP_DECODER_PROFILE_HEVC_MAIN_10;
405	break;
406	case FF_PROFILE_HEVC_MAIN_STILL_PICTURE:
407	profile = VDP_DECODER_PROFILE_HEVC_MAIN_STILL;
408	break;
409	default:
410	return AVERROR(ENOTSUP);
411	}
412
413	return ff_vdpau_common_init(avctx, profile, level);
414	}
415
416	AVHWAccel ff_hevc_vdpau_hwaccel = {
417	.name = "hevc_vdpau",
418	.type = AVMEDIA_TYPE_VIDEO,
419	.id = AV_CODEC_ID_HEVC,
420	.pix_fmt = AV_PIX_FMT_VDPAU,
421	.start_frame = vdpau_hevc_start_frame,
422	.end_frame = vdpau_hevc_end_frame,
423	.decode_slice = vdpau_hevc_decode_slice,
424	.frame_priv_data_size = sizeof(struct vdpau_picture_context),
425	.init = vdpau_hevc_init,
426	.uninit = ff_vdpau_common_uninit,
427	.priv_data_size = sizeof(VDPAUContext),
428	.caps_internal = HWACCEL_CAP_ASYNC_SAFE,
429	};
430