FFmpeg coverage

Line	Branch	Exec	Source
1			/*
2			* H.264 HW decode acceleration through VA API
3			*
4			* Copyright (C) 2008-2009 Splitted-Desktop Systems
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 "h264dec.h"
24			#include "h264_ps.h"
25			#include "hwaccel_internal.h"
26			#include "vaapi_decode.h"
27
28			/**
29			* @file
30			* This file implements the glue code between FFmpeg's and VA API's
31			* structures for H.264 decoding.
32			*/
33
34			/**
35			* Initialize an empty VA API picture.
36			*
37			* VA API requires a fixed-size reference picture array.
38			*/
39		✗	static void init_vaapi_pic(VAPictureH264 *va_pic)
40			{
41		✗	va_pic->picture_id = VA_INVALID_ID;
42		✗	va_pic->flags = VA_PICTURE_H264_INVALID;
43		✗	va_pic->TopFieldOrderCnt = 0;
44		✗	va_pic->BottomFieldOrderCnt = 0;
45		✗	}
46
47			/**
48			* Translate an FFmpeg Picture into its VA API form.
49			*
50			* @param[out] va_pic A pointer to VA API's own picture struct
51			* @param[in] pic A pointer to the FFmpeg picture struct to convert
52			* @param[in] pic_structure The picture field type (as defined in mpegvideo.h),
53			* supersedes pic's field type if nonzero.
54			*/
55		✗	static void fill_vaapi_pic(VAPictureH264 *va_pic,
56			const H264Picture *pic,
57			int pic_structure)
58			{
59		✗	if (pic_structure == 0)
60		✗	pic_structure = pic->reference;
61		✗	pic_structure &= PICT_FRAME; /* PICT_TOP_FIELD|PICT_BOTTOM_FIELD */
62
63		✗	va_pic->picture_id = ff_vaapi_get_surface_id(pic->f);
64		✗	va_pic->frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
65
66		✗	va_pic->flags = 0;
67		✗	if (pic_structure != PICT_FRAME)
68		✗	va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD;
69		✗	if (pic->reference)
70		✗	va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE;
71
72		✗	va_pic->TopFieldOrderCnt = 0;
73		✗	if (pic->field_poc[0] != INT_MAX)
74		✗	va_pic->TopFieldOrderCnt = pic->field_poc[0];
75
76		✗	va_pic->BottomFieldOrderCnt = 0;
77		✗	if (pic->field_poc[1] != INT_MAX)
78		✗	va_pic->BottomFieldOrderCnt = pic->field_poc[1];
79		✗	}
80
81			/** Decoded Picture Buffer (DPB). */
82			typedef struct DPB {
83			int size; ///< Current number of reference frames in the DPB
84			int max_size; ///< Max number of reference frames. This is FF_ARRAY_ELEMS(VAPictureParameterBufferH264.ReferenceFrames)
85			VAPictureH264 *va_pics; ///< Pointer to VAPictureParameterBufferH264.ReferenceFrames array
86			} DPB;
87
88			/**
89			* Append picture to the decoded picture buffer, in a VA API form that
90			* merges the second field picture attributes with the first, if
91			* available. The decoded picture buffer's size must be large enough
92			* to receive the new VA API picture object.
93			*/
94		✗	static int dpb_add(DPB *dpb, const H264Picture *pic)
95			{
96		✗	int i, pic_frame_idx, merged = 0;
97
98		✗	if (dpb->size >= dpb->max_size)
99		✗	return -1;
100
101		✗	pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
102
103		✗	for (i = 0; i < dpb->size; i++) {
104		✗	VAPictureH264 * const va_pic = &dpb->va_pics[i];
105		✗	int va_pic_long_ref = !!(va_pic->flags & VA_PICTURE_H264_LONG_TERM_REFERENCE);
106		✗	if (va_pic->picture_id == ff_vaapi_get_surface_id(pic->f) &&
107		✗	va_pic_long_ref == pic->long_ref &&
108		✗	va_pic->frame_idx == pic_frame_idx) {
109			VAPictureH264 temp_va_pic;
110		✗	fill_vaapi_pic(&temp_va_pic, pic, 0);
111
112		✗	if ((temp_va_pic.flags ^ va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) {
113		✗	va_pic->flags |= temp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD);
114			/* Merge second field */
115		✗	if (temp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) {
116		✗	va_pic->TopFieldOrderCnt = temp_va_pic.TopFieldOrderCnt;
117			} else {
118		✗	va_pic->BottomFieldOrderCnt = temp_va_pic.BottomFieldOrderCnt;
119			}
120		✗	merged = 1;
121			}
122			}
123			}
124
125		✗	if (merged)
126		✗	return 0;
127
128		✗	fill_vaapi_pic(&dpb->va_pics[dpb->size++], pic, 0);
129		✗	return 0;
130			}
131
132			/** Fill in VA API reference frames array. */
133		✗	static int fill_vaapi_ReferenceFrames(VAPictureParameterBufferH264 *pic_param,
134			const H264Context *h)
135			{
136			DPB dpb;
137			int i;
138
139		✗	dpb.size = 0;
140		✗	dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames);
141		✗	dpb.va_pics = pic_param->ReferenceFrames;
142		✗	for (i = 0; i < dpb.max_size; i++)
143		✗	init_vaapi_pic(&dpb.va_pics[i]);
144
145		✗	for (i = 0; i < h->short_ref_count; i++) {
146		✗	const H264Picture *pic = h->short_ref[i];
147		✗	if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
148		✗	return -1;
149			}
150
151		✗	for (i = 0; i < 16; i++) {
152		✗	const H264Picture *pic = h->long_ref[i];
153		✗	if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
154		✗	return -1;
155			}
156		✗	return 0;
157			}
158
159			/**
160			* Fill in VA API reference picture lists from the FFmpeg reference
161			* picture list.
162			*
163			* @param[out] RefPicList VA API internal reference picture list
164			* @param[in] ref_list A pointer to the FFmpeg reference list
165			* @param[in] ref_count The number of reference pictures in ref_list
166			*/
167		✗	static void fill_vaapi_RefPicList(VAPictureH264 RefPicList[32],
168			const H264Ref *ref_list,
169			unsigned int ref_count)
170			{
171		✗	unsigned int i, n = 0;
172		✗	for (i = 0; i < ref_count; i++)
173		✗	if (ref_list[i].reference)
174		✗	fill_vaapi_pic(&RefPicList[n++], ref_list[i].parent,
175		✗	ref_list[i].reference);
176
177		✗	for (; n < 32; n++)
178		✗	init_vaapi_pic(&RefPicList[n]);
179		✗	}
180
181			/**
182			* Fill in prediction weight table.
183			*
184			* VA API requires a plain prediction weight table as it does not infer
185			* any value.
186			*
187			* @param[in] h A pointer to the current H.264 context
188			* @param[in] list The reference frame list index to use
189			* @param[out] luma_weight_flag VA API plain luma weight flag
190			* @param[out] luma_weight VA API plain luma weight table
191			* @param[out] luma_offset VA API plain luma offset table
192			* @param[out] chroma_weight_flag VA API plain chroma weight flag
193			* @param[out] chroma_weight VA API plain chroma weight table
194			* @param[out] chroma_offset VA API plain chroma offset table
195			*/
196		✗	static void fill_vaapi_plain_pred_weight_table(const H264Context *h,
197			int list,
198			unsigned char *luma_weight_flag,
199			short luma_weight[32],
200			short luma_offset[32],
201			unsigned char *chroma_weight_flag,
202			short chroma_weight[32][2],
203			short chroma_offset[32][2])
204			{
205		✗	const H264SliceContext *sl = &h->slice_ctx[0];
206			unsigned int i, j;
207
208		✗	*luma_weight_flag = sl->pwt.luma_weight_flag[list];
209		✗	*chroma_weight_flag = sl->pwt.chroma_weight_flag[list];
210
211		✗	for (i = 0; i < sl->ref_count[list]; i++) {
212			/* VA API also wants the inferred (default) values, not
213			only what is available in the bitstream (7.4.3.2). */
214		✗	if (sl->pwt.luma_weight_flag[list]) {
215		✗	luma_weight[i] = sl->pwt.luma_weight[i][list][0];
216		✗	luma_offset[i] = sl->pwt.luma_weight[i][list][1];
217			} else {
218		✗	luma_weight[i] = 1 << sl->pwt.luma_log2_weight_denom;
219		✗	luma_offset[i] = 0;
220			}
221		✗	for (j = 0; j < 2; j++) {
222		✗	if (sl->pwt.chroma_weight_flag[list]) {
223		✗	chroma_weight[i][j] = sl->pwt.chroma_weight[i][list][j][0];
224		✗	chroma_offset[i][j] = sl->pwt.chroma_weight[i][list][j][1];
225			} else {
226		✗	chroma_weight[i][j] = 1 << sl->pwt.chroma_log2_weight_denom;
227		✗	chroma_offset[i][j] = 0;
228			}
229			}
230			}
231		✗	}
232
233			/** Initialize and start decoding a frame with VA API. */
234		✗	static int vaapi_h264_start_frame(AVCodecContext *avctx,
235			av_unused const AVBufferRef *buffer_ref,
236			av_unused const uint8_t *buffer,
237			av_unused uint32_t size)
238			{
239		✗	const H264Context *h = avctx->priv_data;
240		✗	VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
241		✗	const PPS *pps = h->ps.pps;
242		✗	const SPS *sps = h->ps.sps;
243			VAPictureParameterBufferH264 pic_param;
244			VAIQMatrixBufferH264 iq_matrix;
245			int err;
246
247		✗	pic->output_surface = ff_vaapi_get_surface_id(h->cur_pic_ptr->f);
248
249		✗	pic_param = (VAPictureParameterBufferH264) {
250		✗	.picture_width_in_mbs_minus1 = h->mb_width - 1,
251		✗	.picture_height_in_mbs_minus1 = h->mb_height - 1,
252		✗	.bit_depth_luma_minus8 = sps->bit_depth_luma - 8,
253		✗	.bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8,
254		✗	.num_ref_frames = sps->ref_frame_count,
255			.seq_fields.bits = {
256		✗	.chroma_format_idc = sps->chroma_format_idc,
257		✗	.residual_colour_transform_flag = sps->residual_color_transform_flag,
258		✗	.gaps_in_frame_num_value_allowed_flag = sps->gaps_in_frame_num_allowed_flag,
259		✗	.frame_mbs_only_flag = sps->frame_mbs_only_flag,
260		✗	.mb_adaptive_frame_field_flag = sps->mb_aff,
261		✗	.direct_8x8_inference_flag = sps->direct_8x8_inference_flag,
262		✗	.MinLumaBiPredSize8x8 = sps->level_idc >= 31, /* A.3.3.2 */
263		✗	.log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4,
264		✗	.pic_order_cnt_type = sps->poc_type,
265		✗	.log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4,
266		✗	.delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag,
267			},
268		✗	.pic_init_qp_minus26 = pps->init_qp - 26,
269		✗	.pic_init_qs_minus26 = pps->init_qs - 26,
270		✗	.chroma_qp_index_offset = pps->chroma_qp_index_offset[0],
271		✗	.second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1],
272			.pic_fields.bits = {
273		✗	.entropy_coding_mode_flag = pps->cabac,
274		✗	.weighted_pred_flag = pps->weighted_pred,
275		✗	.weighted_bipred_idc = pps->weighted_bipred_idc,
276		✗	.transform_8x8_mode_flag = pps->transform_8x8_mode,
277		✗	.field_pic_flag = h->picture_structure != PICT_FRAME,
278		✗	.constrained_intra_pred_flag = pps->constrained_intra_pred,
279		✗	.pic_order_present_flag = pps->pic_order_present,
280		✗	.deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present,
281		✗	.redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present,
282		✗	.reference_pic_flag = h->nal_ref_idc != 0,
283			},
284		✗	.frame_num = h->poc.frame_num,
285			};
286
287		✗	fill_vaapi_pic(&pic_param.CurrPic, h->cur_pic_ptr, h->picture_structure);
288		✗	err = fill_vaapi_ReferenceFrames(&pic_param, h);
289		✗	if (err < 0)
290		✗	goto fail;
291
292		✗	err = ff_vaapi_decode_make_param_buffer(avctx, pic,
293			VAPictureParameterBufferType,
294			&pic_param, sizeof(pic_param));
295		✗	if (err < 0)
296		✗	goto fail;
297
298		✗	memcpy(iq_matrix.ScalingList4x4,
299		✗	pps->scaling_matrix4, sizeof(iq_matrix.ScalingList4x4));
300		✗	memcpy(iq_matrix.ScalingList8x8[0],
301		✗	pps->scaling_matrix8[0], sizeof(iq_matrix.ScalingList8x8[0]));
302		✗	memcpy(iq_matrix.ScalingList8x8[1],
303		✗	pps->scaling_matrix8[3], sizeof(iq_matrix.ScalingList8x8[0]));
304
305		✗	err = ff_vaapi_decode_make_param_buffer(avctx, pic,
306			VAIQMatrixBufferType,
307			&iq_matrix, sizeof(iq_matrix));
308		✗	if (err < 0)
309		✗	goto fail;
310
311		✗	return 0;
312
313		✗	fail:
314		✗	ff_vaapi_decode_cancel(avctx, pic);
315		✗	return err;
316			}
317
318			/** End a hardware decoding based frame. */
319		✗	static int vaapi_h264_end_frame(AVCodecContext *avctx)
320			{
321		✗	const H264Context *h = avctx->priv_data;
322		✗	VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
323		✗	H264SliceContext *sl = &h->slice_ctx[0];
324			int ret;
325
326		✗	ret = ff_vaapi_decode_issue(avctx, pic);
327		✗	if (ret < 0)
328		✗	goto finish;
329
330		✗	ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height);
331
332		✗	finish:
333		✗	return ret;
334			}
335
336			/** Decode the given H.264 slice with VA API. */
337		✗	static int vaapi_h264_decode_slice(AVCodecContext *avctx,
338			const uint8_t *buffer,
339			uint32_t size)
340			{
341		✗	const H264Context *h = avctx->priv_data;
342		✗	VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
343		✗	const H264SliceContext *sl = &h->slice_ctx[0];
344			VASliceParameterBufferH264 slice_param;
345			int err;
346
347		✗	slice_param = (VASliceParameterBufferH264) {
348			.slice_data_size = size,
349			.slice_data_offset = 0,
350			.slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
351		✗	.slice_data_bit_offset = get_bits_count(&sl->gb),
352		✗	.first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x,
353		✗	.slice_type = ff_h264_get_slice_type(sl),
354		✗	.direct_spatial_mv_pred_flag = sl->slice_type == AV_PICTURE_TYPE_B ? sl->direct_spatial_mv_pred : 0,
355		✗	.num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0,
356		✗	.num_ref_idx_l1_active_minus1 = sl->list_count > 1 ? sl->ref_count[1] - 1 : 0,
357		✗	.cabac_init_idc = sl->cabac_init_idc,
358		✗	.slice_qp_delta = sl->qscale - h->ps.pps->init_qp,
359		✗	.disable_deblocking_filter_idc = sl->deblocking_filter < 2 ? !sl->deblocking_filter : sl->deblocking_filter,
360		✗	.slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2,
361		✗	.slice_beta_offset_div2 = sl->slice_beta_offset / 2,
362		✗	.luma_log2_weight_denom = sl->pwt.luma_log2_weight_denom,
363		✗	.chroma_log2_weight_denom = sl->pwt.chroma_log2_weight_denom,
364			};
365
366		✗	fill_vaapi_RefPicList(slice_param.RefPicList0, sl->ref_list[0],
367		✗	sl->list_count > 0 ? sl->ref_count[0] : 0);
368		✗	fill_vaapi_RefPicList(slice_param.RefPicList1, sl->ref_list[1],
369		✗	sl->list_count > 1 ? sl->ref_count[1] : 0);
370
371		✗	fill_vaapi_plain_pred_weight_table(h, 0,
372			&slice_param.luma_weight_l0_flag,
373			slice_param.luma_weight_l0,
374			slice_param.luma_offset_l0,
375			&slice_param.chroma_weight_l0_flag,
376			slice_param.chroma_weight_l0,
377			slice_param.chroma_offset_l0);
378		✗	fill_vaapi_plain_pred_weight_table(h, 1,
379			&slice_param.luma_weight_l1_flag,
380			slice_param.luma_weight_l1,
381			slice_param.luma_offset_l1,
382			&slice_param.chroma_weight_l1_flag,
383			slice_param.chroma_weight_l1,
384			slice_param.chroma_offset_l1);
385
386		✗	err = ff_vaapi_decode_make_slice_buffer(avctx, pic,
387			&slice_param, 1, sizeof(slice_param),
388			buffer, size);
389		✗	if (err) {
390		✗	ff_vaapi_decode_cancel(avctx, pic);
391		✗	return err;
392			}
393
394		✗	return 0;
395			}
396
397			const FFHWAccel ff_h264_vaapi_hwaccel = {
398			.p.name = "h264_vaapi",
399			.p.type = AVMEDIA_TYPE_VIDEO,
400			.p.id = AV_CODEC_ID_H264,
401			.p.pix_fmt = AV_PIX_FMT_VAAPI,
402			.start_frame = &vaapi_h264_start_frame,
403			.end_frame = &vaapi_h264_end_frame,
404			.decode_slice = &vaapi_h264_decode_slice,
405			.frame_priv_data_size = sizeof(VAAPIDecodePicture),
406			.init = &ff_vaapi_decode_init,
407			.uninit = &ff_vaapi_decode_uninit,
408			.frame_params = &ff_vaapi_common_frame_params,
409			.priv_data_size = sizeof(VAAPIDecodeContext),
410			.caps_internal = HWACCEL_CAP_ASYNC_SAFE,
411			};
412