2 * H.264 HW decode acceleration through VA API
4 * Copyright (C) 2008-2009 Splitted-Desktop Systems
6 * This file is part of Libav.
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.
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.
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
23 #include "vaapi_internal.h"
27 * This file implements the glue code between Libav's and VA API's
28 * structures for H.264 decoding.
32 * Initialize an empty VA API picture.
34 * VA API requires a fixed-size reference picture array.
36 static void init_vaapi_pic(VAPictureH264 *va_pic)
38 va_pic->picture_id = VA_INVALID_ID;
39 va_pic->flags = VA_PICTURE_H264_INVALID;
40 va_pic->TopFieldOrderCnt = 0;
41 va_pic->BottomFieldOrderCnt = 0;
45 * Translate an Libav Picture into its VA API form.
47 * @param[out] va_pic A pointer to VA API's own picture struct
48 * @param[in] pic A pointer to the Libav picture struct to convert
49 * @param[in] pic_structure The picture field type (as defined in mpegvideo.h),
50 * supersedes pic's field type if nonzero.
52 static void fill_vaapi_pic(VAPictureH264 *va_pic,
56 if (pic_structure == 0)
57 pic_structure = pic->reference;
58 pic_structure &= PICT_FRAME; /* PICT_TOP_FIELD|PICT_BOTTOM_FIELD */
60 va_pic->picture_id = ff_vaapi_get_surface_id(pic);
61 va_pic->frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
64 if (pic_structure != PICT_FRAME)
65 va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD;
67 va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE;
69 va_pic->TopFieldOrderCnt = 0;
70 if (pic->field_poc[0] != INT_MAX)
71 va_pic->TopFieldOrderCnt = pic->field_poc[0];
73 va_pic->BottomFieldOrderCnt = 0;
74 if (pic->field_poc[1] != INT_MAX)
75 va_pic->BottomFieldOrderCnt = pic->field_poc[1];
78 /** Decoded Picture Buffer (DPB). */
80 int size; ///< Current number of reference frames in the DPB
81 int max_size; ///< Max number of reference frames. This is FF_ARRAY_ELEMS(VAPictureParameterBufferH264.ReferenceFrames)
82 VAPictureH264 *va_pics; ///< Pointer to VAPictureParameterBufferH264.ReferenceFrames array
86 * Append picture to the decoded picture buffer, in a VA API form that
87 * merges the second field picture attributes with the first, if
88 * available. The decoded picture buffer's size must be large enough
89 * to receive the new VA API picture object.
91 static int dpb_add(DPB *dpb, Picture *pic)
95 if (dpb->size >= dpb->max_size)
98 for (i = 0; i < dpb->size; i++) {
99 VAPictureH264 * const va_pic = &dpb->va_pics[i];
100 if (va_pic->picture_id == ff_vaapi_get_surface_id(pic)) {
101 VAPictureH264 temp_va_pic;
102 fill_vaapi_pic(&temp_va_pic, pic, 0);
104 if ((temp_va_pic.flags ^ va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) {
105 va_pic->flags |= temp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD);
106 /* Merge second field */
107 if (temp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) {
108 va_pic->TopFieldOrderCnt = temp_va_pic.TopFieldOrderCnt;
110 va_pic->BottomFieldOrderCnt = temp_va_pic.BottomFieldOrderCnt;
117 fill_vaapi_pic(&dpb->va_pics[dpb->size++], pic, 0);
121 /** Fill in VA API reference frames array. */
122 static int fill_vaapi_ReferenceFrames(VAPictureParameterBufferH264 *pic_param,
129 dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames);
130 dpb.va_pics = pic_param->ReferenceFrames;
131 for (i = 0; i < dpb.max_size; i++)
132 init_vaapi_pic(&dpb.va_pics[i]);
134 for (i = 0; i < h->short_ref_count; i++) {
135 Picture * const pic = h->short_ref[i];
136 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
140 for (i = 0; i < 16; i++) {
141 Picture * const pic = h->long_ref[i];
142 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
149 * Fill in VA API reference picture lists from the Libav reference
152 * @param[out] RefPicList VA API internal reference picture list
153 * @param[in] ref_list A pointer to the Libav reference list
154 * @param[in] ref_count The number of reference pictures in ref_list
156 static void fill_vaapi_RefPicList(VAPictureH264 RefPicList[32],
158 unsigned int ref_count)
160 unsigned int i, n = 0;
161 for (i = 0; i < ref_count; i++)
162 if (ref_list[i].reference)
163 fill_vaapi_pic(&RefPicList[n++], &ref_list[i], 0);
166 init_vaapi_pic(&RefPicList[n]);
170 * Fill in prediction weight table.
172 * VA API requires a plain prediction weight table as it does not infer
175 * @param[in] h A pointer to the current H.264 context
176 * @param[in] list The reference frame list index to use
177 * @param[out] luma_weight_flag VA API plain luma weight flag
178 * @param[out] luma_weight VA API plain luma weight table
179 * @param[out] luma_offset VA API plain luma offset table
180 * @param[out] chroma_weight_flag VA API plain chroma weight flag
181 * @param[out] chroma_weight VA API plain chroma weight table
182 * @param[out] chroma_offset VA API plain chroma offset table
184 static void fill_vaapi_plain_pred_weight_table(H264Context *h,
186 unsigned char *luma_weight_flag,
187 short luma_weight[32],
188 short luma_offset[32],
189 unsigned char *chroma_weight_flag,
190 short chroma_weight[32][2],
191 short chroma_offset[32][2])
195 *luma_weight_flag = h->luma_weight_flag[list];
196 *chroma_weight_flag = h->chroma_weight_flag[list];
198 for (i = 0; i < h->ref_count[list]; i++) {
199 /* VA API also wants the inferred (default) values, not
200 only what is available in the bitstream (7.4.3.2). */
201 if (h->luma_weight_flag[list]) {
202 luma_weight[i] = h->luma_weight[i][list][0];
203 luma_offset[i] = h->luma_weight[i][list][1];
205 luma_weight[i] = 1 << h->luma_log2_weight_denom;
208 for (j = 0; j < 2; j++) {
209 if (h->chroma_weight_flag[list]) {
210 chroma_weight[i][j] = h->chroma_weight[i][list][j][0];
211 chroma_offset[i][j] = h->chroma_weight[i][list][j][1];
213 chroma_weight[i][j] = 1 << h->chroma_log2_weight_denom;
214 chroma_offset[i][j] = 0;
220 /** Initialize and start decoding a frame with VA API. */
221 static int start_frame(AVCodecContext *avctx,
222 av_unused const uint8_t *buffer,
223 av_unused uint32_t size)
225 H264Context * const h = avctx->priv_data;
226 MpegEncContext * const s = &h->s;
227 struct vaapi_context * const vactx = avctx->hwaccel_context;
228 VAPictureParameterBufferH264 *pic_param;
229 VAIQMatrixBufferH264 *iq_matrix;
231 av_dlog(avctx, "start_frame()\n");
233 vactx->slice_param_size = sizeof(VASliceParameterBufferH264);
235 /* Fill in VAPictureParameterBufferH264. */
236 pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferH264));
239 fill_vaapi_pic(&pic_param->CurrPic, s->current_picture_ptr, s->picture_structure);
240 if (fill_vaapi_ReferenceFrames(pic_param, h) < 0)
242 pic_param->picture_width_in_mbs_minus1 = s->mb_width - 1;
243 pic_param->picture_height_in_mbs_minus1 = s->mb_height - 1;
244 pic_param->bit_depth_luma_minus8 = h->sps.bit_depth_luma - 8;
245 pic_param->bit_depth_chroma_minus8 = h->sps.bit_depth_chroma - 8;
246 pic_param->num_ref_frames = h->sps.ref_frame_count;
247 pic_param->seq_fields.value = 0; /* reset all bits */
248 pic_param->seq_fields.bits.chroma_format_idc = h->sps.chroma_format_idc;
249 pic_param->seq_fields.bits.residual_colour_transform_flag = h->sps.residual_color_transform_flag; /* XXX: only for 4:4:4 high profile? */
250 pic_param->seq_fields.bits.gaps_in_frame_num_value_allowed_flag = h->sps.gaps_in_frame_num_allowed_flag;
251 pic_param->seq_fields.bits.frame_mbs_only_flag = h->sps.frame_mbs_only_flag;
252 pic_param->seq_fields.bits.mb_adaptive_frame_field_flag = h->sps.mb_aff;
253 pic_param->seq_fields.bits.direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag;
254 pic_param->seq_fields.bits.MinLumaBiPredSize8x8 = h->sps.level_idc >= 31; /* A.3.3.2 */
255 pic_param->seq_fields.bits.log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4;
256 pic_param->seq_fields.bits.pic_order_cnt_type = h->sps.poc_type;
257 pic_param->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4;
258 pic_param->seq_fields.bits.delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
259 pic_param->num_slice_groups_minus1 = h->pps.slice_group_count - 1;
260 pic_param->slice_group_map_type = h->pps.mb_slice_group_map_type;
261 pic_param->slice_group_change_rate_minus1 = 0; /* XXX: unimplemented in Libav */
262 pic_param->pic_init_qp_minus26 = h->pps.init_qp - 26;
263 pic_param->pic_init_qs_minus26 = h->pps.init_qs - 26;
264 pic_param->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0];
265 pic_param->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1];
266 pic_param->pic_fields.value = 0; /* reset all bits */
267 pic_param->pic_fields.bits.entropy_coding_mode_flag = h->pps.cabac;
268 pic_param->pic_fields.bits.weighted_pred_flag = h->pps.weighted_pred;
269 pic_param->pic_fields.bits.weighted_bipred_idc = h->pps.weighted_bipred_idc;
270 pic_param->pic_fields.bits.transform_8x8_mode_flag = h->pps.transform_8x8_mode;
271 pic_param->pic_fields.bits.field_pic_flag = s->picture_structure != PICT_FRAME;
272 pic_param->pic_fields.bits.constrained_intra_pred_flag = h->pps.constrained_intra_pred;
273 pic_param->pic_fields.bits.pic_order_present_flag = h->pps.pic_order_present;
274 pic_param->pic_fields.bits.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
275 pic_param->pic_fields.bits.redundant_pic_cnt_present_flag = h->pps.redundant_pic_cnt_present;
276 pic_param->pic_fields.bits.reference_pic_flag = h->nal_ref_idc != 0;
277 pic_param->frame_num = h->frame_num;
279 /* Fill in VAIQMatrixBufferH264. */
280 iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferH264));
283 memcpy(iq_matrix->ScalingList4x4, h->pps.scaling_matrix4, sizeof(iq_matrix->ScalingList4x4));
284 memcpy(iq_matrix->ScalingList8x8, h->pps.scaling_matrix8, sizeof(iq_matrix->ScalingList8x8));
288 /** End a hardware decoding based frame. */
289 static int end_frame(AVCodecContext *avctx)
291 H264Context * const h = avctx->priv_data;
293 av_dlog(avctx, "end_frame()\n");
294 return ff_vaapi_common_end_frame(&h->s);
297 /** Decode the given H.264 slice with VA API. */
298 static int decode_slice(AVCodecContext *avctx,
299 const uint8_t *buffer,
302 H264Context * const h = avctx->priv_data;
303 MpegEncContext * const s = &h->s;
304 VASliceParameterBufferH264 *slice_param;
306 av_dlog(avctx, "decode_slice(): buffer %p, size %d\n", buffer, size);
308 /* Fill in VASliceParameterBufferH264. */
309 slice_param = (VASliceParameterBufferH264 *)ff_vaapi_alloc_slice(avctx->hwaccel_context, buffer, size);
312 slice_param->slice_data_bit_offset = get_bits_count(&h->s.gb) + 8; /* bit buffer started beyond nal_unit_type */
313 slice_param->first_mb_in_slice = (s->mb_y >> FIELD_OR_MBAFF_PICTURE) * s->mb_width + s->mb_x;
314 slice_param->slice_type = ff_h264_get_slice_type(h);
315 slice_param->direct_spatial_mv_pred_flag = h->slice_type == AV_PICTURE_TYPE_B ? h->direct_spatial_mv_pred : 0;
316 slice_param->num_ref_idx_l0_active_minus1 = h->list_count > 0 ? h->ref_count[0] - 1 : 0;
317 slice_param->num_ref_idx_l1_active_minus1 = h->list_count > 1 ? h->ref_count[1] - 1 : 0;
318 slice_param->cabac_init_idc = h->cabac_init_idc;
319 slice_param->slice_qp_delta = s->qscale - h->pps.init_qp;
320 slice_param->disable_deblocking_filter_idc = h->deblocking_filter < 2 ? !h->deblocking_filter : h->deblocking_filter;
321 slice_param->slice_alpha_c0_offset_div2 = h->slice_alpha_c0_offset / 2 - 26;
322 slice_param->slice_beta_offset_div2 = h->slice_beta_offset / 2 - 26;
323 slice_param->luma_log2_weight_denom = h->luma_log2_weight_denom;
324 slice_param->chroma_log2_weight_denom = h->chroma_log2_weight_denom;
326 fill_vaapi_RefPicList(slice_param->RefPicList0, h->ref_list[0], h->list_count > 0 ? h->ref_count[0] : 0);
327 fill_vaapi_RefPicList(slice_param->RefPicList1, h->ref_list[1], h->list_count > 1 ? h->ref_count[1] : 0);
329 fill_vaapi_plain_pred_weight_table(h, 0,
330 &slice_param->luma_weight_l0_flag, slice_param->luma_weight_l0, slice_param->luma_offset_l0,
331 &slice_param->chroma_weight_l0_flag, slice_param->chroma_weight_l0, slice_param->chroma_offset_l0);
332 fill_vaapi_plain_pred_weight_table(h, 1,
333 &slice_param->luma_weight_l1_flag, slice_param->luma_weight_l1, slice_param->luma_offset_l1,
334 &slice_param->chroma_weight_l1_flag, slice_param->chroma_weight_l1, slice_param->chroma_offset_l1);
338 AVHWAccel ff_h264_vaapi_hwaccel = {
339 .name = "h264_vaapi",
340 .type = AVMEDIA_TYPE_VIDEO,
342 .pix_fmt = PIX_FMT_VAAPI_VLD,
344 .start_frame = start_frame,
345 .end_frame = end_frame,
346 .decode_slice = decode_slice,