vaapi_h264: Convert to use the new VAAPI hwaccel code
[libav.git] / libavcodec / vaapi_h264.c
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 Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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 "vaapi_decode.h"
26
27 /**
28 * @file
29 * This file implements the glue code between Libav's and VA API's
30 * structures for H.264 decoding.
31 */
32
33 /**
34 * Initialize an empty VA API picture.
35 *
36 * VA API requires a fixed-size reference picture array.
37 */
38 static void init_vaapi_pic(VAPictureH264 *va_pic)
39 {
40 va_pic->picture_id = VA_INVALID_ID;
41 va_pic->flags = VA_PICTURE_H264_INVALID;
42 va_pic->TopFieldOrderCnt = 0;
43 va_pic->BottomFieldOrderCnt = 0;
44 }
45
46 /**
47 * Translate an Libav Picture into its VA API form.
48 *
49 * @param[out] va_pic A pointer to VA API's own picture struct
50 * @param[in] pic A pointer to the Libav picture struct to convert
51 * @param[in] pic_structure The picture field type (as defined in mpegvideo.h),
52 * supersedes pic's field type if nonzero.
53 */
54 static void fill_vaapi_pic(VAPictureH264 *va_pic,
55 const H264Picture *pic,
56 int pic_structure)
57 {
58 if (pic_structure == 0)
59 pic_structure = pic->reference;
60 pic_structure &= PICT_FRAME; /* PICT_TOP_FIELD|PICT_BOTTOM_FIELD */
61
62 va_pic->picture_id = ff_vaapi_get_surface_id(pic->f);
63 va_pic->frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
64
65 va_pic->flags = 0;
66 if (pic_structure != PICT_FRAME)
67 va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD;
68 if (pic->reference)
69 va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE;
70
71 va_pic->TopFieldOrderCnt = 0;
72 if (pic->field_poc[0] != INT_MAX)
73 va_pic->TopFieldOrderCnt = pic->field_poc[0];
74
75 va_pic->BottomFieldOrderCnt = 0;
76 if (pic->field_poc[1] != INT_MAX)
77 va_pic->BottomFieldOrderCnt = pic->field_poc[1];
78 }
79
80 /** Decoded Picture Buffer (DPB). */
81 typedef struct DPB {
82 int size; ///< Current number of reference frames in the DPB
83 int max_size; ///< Max number of reference frames. This is FF_ARRAY_ELEMS(VAPictureParameterBufferH264.ReferenceFrames)
84 VAPictureH264 *va_pics; ///< Pointer to VAPictureParameterBufferH264.ReferenceFrames array
85 } DPB;
86
87 /**
88 * Append picture to the decoded picture buffer, in a VA API form that
89 * merges the second field picture attributes with the first, if
90 * available. The decoded picture buffer's size must be large enough
91 * to receive the new VA API picture object.
92 */
93 static int dpb_add(DPB *dpb, const H264Picture *pic)
94 {
95 int i;
96
97 if (dpb->size >= dpb->max_size)
98 return -1;
99
100 for (i = 0; i < dpb->size; i++) {
101 VAPictureH264 * const va_pic = &dpb->va_pics[i];
102 if (va_pic->picture_id == ff_vaapi_get_surface_id(pic->f)) {
103 VAPictureH264 temp_va_pic;
104 fill_vaapi_pic(&temp_va_pic, pic, 0);
105
106 if ((temp_va_pic.flags ^ va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) {
107 va_pic->flags |= temp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD);
108 /* Merge second field */
109 if (temp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) {
110 va_pic->TopFieldOrderCnt = temp_va_pic.TopFieldOrderCnt;
111 } else {
112 va_pic->BottomFieldOrderCnt = temp_va_pic.BottomFieldOrderCnt;
113 }
114 }
115 return 0;
116 }
117 }
118
119 fill_vaapi_pic(&dpb->va_pics[dpb->size++], pic, 0);
120 return 0;
121 }
122
123 /** Fill in VA API reference frames array. */
124 static int fill_vaapi_ReferenceFrames(VAPictureParameterBufferH264 *pic_param,
125 const H264Context *h)
126 {
127 DPB dpb;
128 int i;
129
130 dpb.size = 0;
131 dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames);
132 dpb.va_pics = pic_param->ReferenceFrames;
133 for (i = 0; i < dpb.max_size; i++)
134 init_vaapi_pic(&dpb.va_pics[i]);
135
136 for (i = 0; i < h->short_ref_count; i++) {
137 const H264Picture *pic = h->short_ref[i];
138 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
139 return -1;
140 }
141
142 for (i = 0; i < 16; i++) {
143 const H264Picture *pic = h->long_ref[i];
144 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
145 return -1;
146 }
147 return 0;
148 }
149
150 /**
151 * Fill in VA API reference picture lists from the Libav reference
152 * picture list.
153 *
154 * @param[out] RefPicList VA API internal reference picture list
155 * @param[in] ref_list A pointer to the Libav reference list
156 * @param[in] ref_count The number of reference pictures in ref_list
157 */
158 static void fill_vaapi_RefPicList(VAPictureH264 RefPicList[32],
159 const H264Ref *ref_list,
160 unsigned int ref_count)
161 {
162 unsigned int i, n = 0;
163 for (i = 0; i < ref_count; i++)
164 if (ref_list[i].reference)
165 fill_vaapi_pic(&RefPicList[n++], ref_list[i].parent, 0);
166
167 for (; n < 32; n++)
168 init_vaapi_pic(&RefPicList[n]);
169 }
170
171 /**
172 * Fill in prediction weight table.
173 *
174 * VA API requires a plain prediction weight table as it does not infer
175 * any value.
176 *
177 * @param[in] h A pointer to the current H.264 context
178 * @param[in] list The reference frame list index to use
179 * @param[out] luma_weight_flag VA API plain luma weight flag
180 * @param[out] luma_weight VA API plain luma weight table
181 * @param[out] luma_offset VA API plain luma offset table
182 * @param[out] chroma_weight_flag VA API plain chroma weight flag
183 * @param[out] chroma_weight VA API plain chroma weight table
184 * @param[out] chroma_offset VA API plain chroma offset table
185 */
186 static void fill_vaapi_plain_pred_weight_table(const H264Context *h,
187 int list,
188 unsigned char *luma_weight_flag,
189 short luma_weight[32],
190 short luma_offset[32],
191 unsigned char *chroma_weight_flag,
192 short chroma_weight[32][2],
193 short chroma_offset[32][2])
194 {
195 const H264SliceContext *sl = &h->slice_ctx[0];
196 unsigned int i, j;
197
198 *luma_weight_flag = sl->pwt.luma_weight_flag[list];
199 *chroma_weight_flag = sl->pwt.chroma_weight_flag[list];
200
201 for (i = 0; i < sl->ref_count[list]; i++) {
202 /* VA API also wants the inferred (default) values, not
203 only what is available in the bitstream (7.4.3.2). */
204 if (sl->pwt.luma_weight_flag[list]) {
205 luma_weight[i] = sl->pwt.luma_weight[i][list][0];
206 luma_offset[i] = sl->pwt.luma_weight[i][list][1];
207 } else {
208 luma_weight[i] = 1 << sl->pwt.luma_log2_weight_denom;
209 luma_offset[i] = 0;
210 }
211 for (j = 0; j < 2; j++) {
212 if (sl->pwt.chroma_weight_flag[list]) {
213 chroma_weight[i][j] = sl->pwt.chroma_weight[i][list][j][0];
214 chroma_offset[i][j] = sl->pwt.chroma_weight[i][list][j][1];
215 } else {
216 chroma_weight[i][j] = 1 << sl->pwt.chroma_log2_weight_denom;
217 chroma_offset[i][j] = 0;
218 }
219 }
220 }
221 }
222
223 /** Initialize and start decoding a frame with VA API. */
224 static int vaapi_h264_start_frame(AVCodecContext *avctx,
225 av_unused const uint8_t *buffer,
226 av_unused uint32_t size)
227 {
228 const H264Context *h = avctx->priv_data;
229 VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
230 const PPS *pps = h->ps.pps;
231 const SPS *sps = h->ps.sps;
232 VAPictureParameterBufferH264 pic_param;
233 VAIQMatrixBufferH264 iq_matrix;
234 int err;
235
236 pic->output_surface = ff_vaapi_get_surface_id(h->cur_pic_ptr->f);
237
238 pic_param = (VAPictureParameterBufferH264) {
239 .picture_width_in_mbs_minus1 = h->mb_width - 1,
240 .picture_height_in_mbs_minus1 = h->mb_height - 1,
241 .bit_depth_luma_minus8 = sps->bit_depth_luma - 8,
242 .bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8,
243 .num_ref_frames = sps->ref_frame_count,
244 .seq_fields.bits = {
245 .chroma_format_idc = sps->chroma_format_idc,
246 .residual_colour_transform_flag = sps->residual_color_transform_flag,
247 .gaps_in_frame_num_value_allowed_flag = sps->gaps_in_frame_num_allowed_flag,
248 .frame_mbs_only_flag = sps->frame_mbs_only_flag,
249 .mb_adaptive_frame_field_flag = sps->mb_aff,
250 .direct_8x8_inference_flag = sps->direct_8x8_inference_flag,
251 .MinLumaBiPredSize8x8 = sps->level_idc >= 31, /* A.3.3.2 */
252 .log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4,
253 .pic_order_cnt_type = sps->poc_type,
254 .log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4,
255 .delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag,
256 },
257 .num_slice_groups_minus1 = pps->slice_group_count - 1,
258 .slice_group_map_type = pps->mb_slice_group_map_type,
259 .slice_group_change_rate_minus1 = 0, /* FMO is not implemented */
260 .pic_init_qp_minus26 = pps->init_qp - 26,
261 .pic_init_qs_minus26 = pps->init_qs - 26,
262 .chroma_qp_index_offset = pps->chroma_qp_index_offset[0],
263 .second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1],
264 .pic_fields.bits = {
265 .entropy_coding_mode_flag = pps->cabac,
266 .weighted_pred_flag = pps->weighted_pred,
267 .weighted_bipred_idc = pps->weighted_bipred_idc,
268 .transform_8x8_mode_flag = pps->transform_8x8_mode,
269 .field_pic_flag = h->picture_structure != PICT_FRAME,
270 .constrained_intra_pred_flag = pps->constrained_intra_pred,
271 .pic_order_present_flag = pps->pic_order_present,
272 .deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present,
273 .redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present,
274 .reference_pic_flag = h->nal_ref_idc != 0,
275 },
276 .frame_num = h->poc.frame_num,
277 };
278
279 fill_vaapi_pic(&pic_param.CurrPic, h->cur_pic_ptr, h->picture_structure);
280 err = fill_vaapi_ReferenceFrames(&pic_param, h);
281 if (err < 0)
282 goto fail;
283
284 err = ff_vaapi_decode_make_param_buffer(avctx, pic,
285 VAPictureParameterBufferType,
286 &pic_param, sizeof(pic_param));
287 if (err < 0)
288 goto fail;
289
290 memcpy(iq_matrix.ScalingList4x4,
291 pps->scaling_matrix4, sizeof(iq_matrix.ScalingList4x4));
292 memcpy(iq_matrix.ScalingList8x8[0],
293 pps->scaling_matrix8[0], sizeof(iq_matrix.ScalingList8x8[0]));
294 memcpy(iq_matrix.ScalingList8x8[1],
295 pps->scaling_matrix8[3], sizeof(iq_matrix.ScalingList8x8[0]));
296
297 err = ff_vaapi_decode_make_param_buffer(avctx, pic,
298 VAIQMatrixBufferType,
299 &iq_matrix, sizeof(iq_matrix));
300 if (err < 0)
301 goto fail;
302
303 return 0;
304
305 fail:
306 ff_vaapi_decode_cancel(avctx, pic);
307 return err;
308 }
309
310 /** End a hardware decoding based frame. */
311 static int vaapi_h264_end_frame(AVCodecContext *avctx)
312 {
313 const H264Context *h = avctx->priv_data;
314 VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
315 H264SliceContext *sl = &h->slice_ctx[0];
316 int ret;
317
318 ret = ff_vaapi_decode_issue(avctx, pic);
319 if (ret < 0)
320 goto finish;
321
322 ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height);
323
324 finish:
325 return ret;
326 }
327
328 /** Decode the given H.264 slice with VA API. */
329 static int vaapi_h264_decode_slice(AVCodecContext *avctx,
330 const uint8_t *buffer,
331 uint32_t size)
332 {
333 const H264Context *h = avctx->priv_data;
334 VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
335 const H264SliceContext *sl = &h->slice_ctx[0];
336 VASliceParameterBufferH264 slice_param;
337 int err;
338
339 slice_param = (VASliceParameterBufferH264) {
340 .slice_data_size = size,
341 .slice_data_offset = 0,
342 .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
343 .slice_data_bit_offset = get_bits_count(&sl->gb),
344 .first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x,
345 .slice_type = ff_h264_get_slice_type(sl),
346 .direct_spatial_mv_pred_flag = sl->slice_type == AV_PICTURE_TYPE_B ? sl->direct_spatial_mv_pred : 0,
347 .num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0,
348 .num_ref_idx_l1_active_minus1 = sl->list_count > 1 ? sl->ref_count[1] - 1 : 0,
349 .cabac_init_idc = sl->cabac_init_idc,
350 .slice_qp_delta = sl->qscale - h->ps.pps->init_qp,
351 .disable_deblocking_filter_idc = sl->deblocking_filter < 2 ? !sl->deblocking_filter : sl->deblocking_filter,
352 .slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2,
353 .slice_beta_offset_div2 = sl->slice_beta_offset / 2,
354 .luma_log2_weight_denom = sl->pwt.luma_log2_weight_denom,
355 .chroma_log2_weight_denom = sl->pwt.chroma_log2_weight_denom,
356 };
357
358 fill_vaapi_RefPicList(slice_param.RefPicList0, sl->ref_list[0],
359 sl->list_count > 0 ? sl->ref_count[0] : 0);
360 fill_vaapi_RefPicList(slice_param.RefPicList1, sl->ref_list[1],
361 sl->list_count > 1 ? sl->ref_count[1] : 0);
362
363 fill_vaapi_plain_pred_weight_table(h, 0,
364 &slice_param.luma_weight_l0_flag,
365 slice_param.luma_weight_l0,
366 slice_param.luma_offset_l0,
367 &slice_param.chroma_weight_l0_flag,
368 slice_param.chroma_weight_l0,
369 slice_param.chroma_offset_l0);
370 fill_vaapi_plain_pred_weight_table(h, 1,
371 &slice_param.luma_weight_l1_flag,
372 slice_param.luma_weight_l1,
373 slice_param.luma_offset_l1,
374 &slice_param.chroma_weight_l1_flag,
375 slice_param.chroma_weight_l1,
376 slice_param.chroma_offset_l1);
377
378 err = ff_vaapi_decode_make_slice_buffer(avctx, pic,
379 &slice_param, sizeof(slice_param),
380 buffer, size);
381 if (err) {
382 ff_vaapi_decode_cancel(avctx, pic);
383 return err;
384 }
385
386 return 0;
387 }
388
389 AVHWAccel ff_h264_vaapi_hwaccel = {
390 .name = "h264_vaapi",
391 .type = AVMEDIA_TYPE_VIDEO,
392 .id = AV_CODEC_ID_H264,
393 .pix_fmt = AV_PIX_FMT_VAAPI,
394 .start_frame = &vaapi_h264_start_frame,
395 .end_frame = &vaapi_h264_end_frame,
396 .decode_slice = &vaapi_h264_decode_slice,
397 .frame_priv_data_size = sizeof(VAAPIDecodePicture),
398 .init = &ff_vaapi_decode_init,
399 .uninit = &ff_vaapi_decode_uninit,
400 .priv_data_size = sizeof(VAAPIDecodeContext),
401 };