comparison vaapi_h264.c @ 10732:a3af696f8499 libavcodec

Add H.264 bitstream decoding through VAAPI.
author gb
date Tue, 29 Dec 2009 08:48:18 +0000
parents
children 46a247e54d6e
comparison
equal deleted inserted replaced
10731:6e70cc7ddf96 10732:a3af696f8499
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 "vaapi_internal.h"
24 #include "h264.h"
25
26 /** @file
27 * This file implements the glue code between FFmpeg's and VA API's
28 * structures for H.264 decoding.
29 */
30
31 /** Reconstructs bitstream slice_type. */
32 static int get_slice_type(H264Context *h)
33 {
34 switch (h->slice_type) {
35 case FF_P_TYPE: return 0;
36 case FF_B_TYPE: return 1;
37 case FF_I_TYPE: return 2;
38 case FF_SP_TYPE: return 3;
39 case FF_SI_TYPE: return 4;
40 default: return -1;
41 }
42 }
43
44 /**
45 * Initializes an empty VA API picture.
46 *
47 * VA API requires a fixed-size reference picture array.
48 */
49 static void init_vaapi_pic(VAPictureH264 *va_pic)
50 {
51 va_pic->picture_id = VA_INVALID_ID;
52 va_pic->flags = VA_PICTURE_H264_INVALID;
53 va_pic->TopFieldOrderCnt = 0;
54 va_pic->BottomFieldOrderCnt = 0;
55 }
56
57 /**
58 * Translates an FFmpeg Picture into its VA API form.
59 *
60 * @param[out] va_pic A pointer to VA API's own picture struct
61 * @param[in] pic A pointer to the FFmpeg picture struct to convert
62 * @param[in] pic_structure The picture field type (as defined in mpegvideo.h),
63 * supersedes pic's field type if nonzero.
64 */
65 static void fill_vaapi_pic(VAPictureH264 *va_pic,
66 Picture *pic,
67 int pic_structure)
68 {
69 if (pic_structure == 0)
70 pic_structure = pic->reference;
71
72 va_pic->picture_id = ff_vaapi_get_surface_id(pic);
73 va_pic->frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
74
75 va_pic->flags = 0;
76 if (pic_structure != PICT_FRAME)
77 va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD;
78 if (pic->reference)
79 va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE;
80
81 va_pic->TopFieldOrderCnt = 0;
82 if (pic->field_poc[0] != INT_MAX)
83 va_pic->TopFieldOrderCnt = pic->field_poc[0];
84
85 va_pic->BottomFieldOrderCnt = 0;
86 if (pic->field_poc[1] != INT_MAX)
87 va_pic->BottomFieldOrderCnt = pic->field_poc[1];
88 }
89
90 /** Decoded Picture Buffer (DPB). */
91 typedef struct DPB {
92 int size; ///< Current number of reference frames in the DPB
93 int max_size; ///< Max number of reference frames. This is FF_ARRAY_ELEMS(VAPictureParameterBufferH264.ReferenceFrames)
94 VAPictureH264 *va_pics; ///< Pointer to VAPictureParameterBufferH264.ReferenceFrames array
95 } DPB;
96
97 /**
98 * Appends picture to the decoded picture buffer, in a VA API form that
99 * merges the second field picture attributes with the first, if
100 * available. The decoded picture buffer's size must be large enough
101 * to receive the new VA API picture object.
102 */
103 static int dpb_add(DPB *dpb, Picture *pic)
104 {
105 int i;
106
107 if (dpb->size >= dpb->max_size)
108 return -1;
109
110 for (i = 0; i < dpb->size; i++) {
111 VAPictureH264 * const va_pic = &dpb->va_pics[i];
112 if (va_pic->picture_id == ff_vaapi_get_surface_id(pic)) {
113 VAPictureH264 temp_va_pic;
114 fill_vaapi_pic(&temp_va_pic, pic, 0);
115
116 if ((temp_va_pic.flags ^ va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) {
117 va_pic->flags |= temp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD);
118 /* Merge second field */
119 if (temp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) {
120 va_pic->TopFieldOrderCnt = temp_va_pic.TopFieldOrderCnt;
121 } else {
122 va_pic->BottomFieldOrderCnt = temp_va_pic.BottomFieldOrderCnt;
123 }
124 }
125 return 0;
126 }
127 }
128
129 fill_vaapi_pic(&dpb->va_pics[dpb->size++], pic, 0);
130 return 0;
131 }
132
133 /** Fills in VA API reference frames array. */
134 static int fill_vaapi_ReferenceFrames(VAPictureParameterBufferH264 *pic_param,
135 H264Context *h)
136 {
137 DPB dpb;
138 int i;
139
140 dpb.size = 0;
141 dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames);
142 dpb.va_pics = pic_param->ReferenceFrames;
143 for (i = 0; i < dpb.max_size; i++)
144 init_vaapi_pic(&dpb.va_pics[i]);
145
146 for (i = 0; i < h->short_ref_count; i++) {
147 Picture * const pic = h->short_ref[i];
148 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
149 return -1;
150 }
151
152 for (i = 0; i < h->long_ref_count; i++) {
153 Picture * const pic = h->long_ref[i];
154 if (pic && pic->reference && dpb_add(&dpb, pic) < 0)
155 return -1;
156 }
157 return 0;
158 }
159
160 /**
161 * Fills in VA API reference picture lists from the FFmpeg reference
162 * picture list.
163 *
164 * @param[out] RefPicList VA API internal reference picture list
165 * @param[in] ref_list A pointer to the FFmpeg reference list
166 * @param[in] ref_count The number of reference pictures in ref_list
167 */
168 static void fill_vaapi_RefPicList(VAPictureH264 RefPicList[32],
169 Picture *ref_list,
170 unsigned int ref_count)
171 {
172 unsigned int i, n = 0;
173 for (i = 0; i < ref_count; i++)
174 if (ref_list[i].reference)
175 fill_vaapi_pic(&RefPicList[n++], &ref_list[i], 0);
176
177 for (; n < 32; n++)
178 init_vaapi_pic(&RefPicList[n]);
179 }
180
181 /**
182 * Fills 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(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 unsigned int i, j;
206
207 *luma_weight_flag = h->luma_weight_flag[list];
208 *chroma_weight_flag = h->chroma_weight_flag[list];
209
210 for (i = 0; i < h->ref_count[list]; i++) {
211 /* VA API also wants the inferred (default) values, not
212 only what is available in the bitstream (7.4.3.2). */
213 if (h->luma_weight_flag[list]) {
214 luma_weight[i] = h->luma_weight[list][i];
215 luma_offset[i] = h->luma_offset[list][i];
216 } else {
217 luma_weight[i] = 1 << h->luma_log2_weight_denom;
218 luma_offset[i] = 0;
219 }
220 for (j = 0; j < 2; j++) {
221 if (h->chroma_weight_flag[list]) {
222 chroma_weight[i][j] = h->chroma_weight[list][i][j];
223 chroma_offset[i][j] = h->chroma_offset[list][i][j];
224 } else {
225 chroma_weight[i][j] = 1 << h->chroma_log2_weight_denom;
226 chroma_offset[i][j] = 0;
227 }
228 }
229 }
230 }
231
232 /** Initializes and starts decoding a frame with VA API. */
233 static int start_frame(AVCodecContext *avctx,
234 av_unused const uint8_t *buffer,
235 av_unused uint32_t size)
236 {
237 H264Context * const h = avctx->priv_data;
238 MpegEncContext * const s = &h->s;
239 struct vaapi_context * const vactx = avctx->hwaccel_context;
240 VAPictureParameterBufferH264 *pic_param;
241 VAIQMatrixBufferH264 *iq_matrix;
242
243 dprintf(avctx, "start_frame()\n");
244
245 vactx->slice_param_size = sizeof(VASliceParameterBufferH264);
246
247 /* Fill in VAPictureParameterBufferH264. */
248 pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferH264));
249 if (!pic_param)
250 return -1;
251 fill_vaapi_pic(&pic_param->CurrPic, s->current_picture_ptr, s->picture_structure);
252 if (fill_vaapi_ReferenceFrames(pic_param, h) < 0)
253 return -1;
254 pic_param->picture_width_in_mbs_minus1 = s->mb_width - 1;
255 pic_param->picture_height_in_mbs_minus1 = s->mb_height - 1;
256 pic_param->bit_depth_luma_minus8 = h->sps.bit_depth_luma - 8;
257 pic_param->bit_depth_chroma_minus8 = h->sps.bit_depth_chroma - 8;
258 pic_param->num_ref_frames = h->sps.ref_frame_count;
259 pic_param->seq_fields.value = 0; /* reset all bits */
260 pic_param->seq_fields.bits.chroma_format_idc = h->sps.chroma_format_idc;
261 pic_param->seq_fields.bits.residual_colour_transform_flag = h->sps.residual_color_transform_flag; /* XXX: only for 4:4:4 high profile? */
262 pic_param->seq_fields.bits.gaps_in_frame_num_value_allowed_flag = h->sps.gaps_in_frame_num_allowed_flag;
263 pic_param->seq_fields.bits.frame_mbs_only_flag = h->sps.frame_mbs_only_flag;
264 pic_param->seq_fields.bits.mb_adaptive_frame_field_flag = h->sps.mb_aff;
265 pic_param->seq_fields.bits.direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag;
266 pic_param->seq_fields.bits.MinLumaBiPredSize8x8 = h->sps.level_idc >= 31; /* A.3.3.2 */
267 pic_param->seq_fields.bits.log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4;
268 pic_param->seq_fields.bits.pic_order_cnt_type = h->sps.poc_type;
269 pic_param->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4;
270 pic_param->seq_fields.bits.delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
271 pic_param->num_slice_groups_minus1 = h->pps.slice_group_count - 1;
272 pic_param->slice_group_map_type = h->pps.mb_slice_group_map_type;
273 pic_param->slice_group_change_rate_minus1 = 0; /* XXX: unimplemented in FFmpeg */
274 pic_param->pic_init_qp_minus26 = h->pps.init_qp - 26;
275 pic_param->pic_init_qs_minus26 = h->pps.init_qs - 26;
276 pic_param->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0];
277 pic_param->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1];
278 pic_param->pic_fields.value = 0; /* reset all bits */
279 pic_param->pic_fields.bits.entropy_coding_mode_flag = h->pps.cabac;
280 pic_param->pic_fields.bits.weighted_pred_flag = h->pps.weighted_pred;
281 pic_param->pic_fields.bits.weighted_bipred_idc = h->pps.weighted_bipred_idc;
282 pic_param->pic_fields.bits.transform_8x8_mode_flag = h->pps.transform_8x8_mode;
283 pic_param->pic_fields.bits.field_pic_flag = s->picture_structure != PICT_FRAME;
284 pic_param->pic_fields.bits.constrained_intra_pred_flag = h->pps.constrained_intra_pred;
285 pic_param->pic_fields.bits.pic_order_present_flag = h->pps.pic_order_present;
286 pic_param->pic_fields.bits.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
287 pic_param->pic_fields.bits.redundant_pic_cnt_present_flag = h->pps.redundant_pic_cnt_present;
288 pic_param->pic_fields.bits.reference_pic_flag = h->nal_ref_idc != 0;
289 pic_param->frame_num = h->frame_num;
290
291 /* Fill in VAIQMatrixBufferH264. */
292 iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferH264));
293 if (!iq_matrix)
294 return -1;
295 memcpy(iq_matrix->ScalingList4x4, h->pps.scaling_matrix4, sizeof(iq_matrix->ScalingList4x4));
296 memcpy(iq_matrix->ScalingList8x8, h->pps.scaling_matrix8, sizeof(iq_matrix->ScalingList8x8));
297 return 0;
298 }
299
300 /** Ends a hardware decoding based frame. */
301 static int end_frame(AVCodecContext *avctx)
302 {
303 H264Context * const h = avctx->priv_data;
304
305 dprintf(avctx, "end_frame()\n");
306 return ff_vaapi_common_end_frame(&h->s);
307 }
308
309 /** Decodes the given H.264 slice with VA API. */
310 static int decode_slice(AVCodecContext *avctx,
311 const uint8_t *buffer,
312 uint32_t size)
313 {
314 H264Context * const h = avctx->priv_data;
315 MpegEncContext * const s = &h->s;
316 VASliceParameterBufferH264 *slice_param;
317
318 dprintf(avctx, "decode_slice(): buffer %p, size %d\n", buffer, size);
319
320 /* Fill in VASliceParameterBufferH264. */
321 slice_param = (VASliceParameterBufferH264 *)ff_vaapi_alloc_slice(avctx->hwaccel_context, buffer, size);
322 if (!slice_param)
323 return -1;
324 slice_param->slice_data_bit_offset = get_bits_count(&h->s.gb) + 8; /* bit buffer started beyond nal_unit_type */
325 slice_param->first_mb_in_slice = (s->mb_y >> FIELD_OR_MBAFF_PICTURE) * s->mb_width + s->mb_x;
326 slice_param->slice_type = get_slice_type(h);
327 slice_param->direct_spatial_mv_pred_flag = h->slice_type == FF_B_TYPE ? h->direct_spatial_mv_pred : 0;
328 slice_param->num_ref_idx_l0_active_minus1 = h->list_count > 0 ? h->ref_count[0] - 1 : 0;
329 slice_param->num_ref_idx_l1_active_minus1 = h->list_count > 1 ? h->ref_count[1] - 1 : 0;
330 slice_param->cabac_init_idc = h->cabac_init_idc;
331 slice_param->slice_qp_delta = s->qscale - h->pps.init_qp;
332 slice_param->disable_deblocking_filter_idc = h->deblocking_filter < 2 ? !h->deblocking_filter : h->deblocking_filter;
333 slice_param->slice_alpha_c0_offset_div2 = h->slice_alpha_c0_offset / 2;
334 slice_param->slice_beta_offset_div2 = h->slice_beta_offset / 2;
335 slice_param->luma_log2_weight_denom = h->luma_log2_weight_denom;
336 slice_param->chroma_log2_weight_denom = h->chroma_log2_weight_denom;
337
338 fill_vaapi_RefPicList(slice_param->RefPicList0, h->ref_list[0], h->list_count > 0 ? h->ref_count[0] : 0);
339 fill_vaapi_RefPicList(slice_param->RefPicList1, h->ref_list[1], h->list_count > 1 ? h->ref_count[1] : 0);
340
341 fill_vaapi_plain_pred_weight_table(h, 0,
342 &slice_param->luma_weight_l0_flag, slice_param->luma_weight_l0, slice_param->luma_offset_l0,
343 &slice_param->chroma_weight_l0_flag, slice_param->chroma_weight_l0, slice_param->chroma_offset_l0);
344 fill_vaapi_plain_pred_weight_table(h, 1,
345 &slice_param->luma_weight_l1_flag, slice_param->luma_weight_l1, slice_param->luma_offset_l1,
346 &slice_param->chroma_weight_l1_flag, slice_param->chroma_weight_l1, slice_param->chroma_offset_l1);
347 return 0;
348 }
349
350 AVHWAccel h264_vaapi_hwaccel = {
351 .name = "h264_vaapi",
352 .type = CODEC_TYPE_VIDEO,
353 .id = CODEC_ID_H264,
354 .pix_fmt = PIX_FMT_VAAPI_VLD,
355 .capabilities = 0,
356 .start_frame = start_frame,
357 .end_frame = end_frame,
358 .decode_slice = decode_slice,
359 .priv_data_size = 0,
360 };