66c7f59f27d958da7f106c44dd4d0495cd86f372
[libav.git] / libavcodec / vdpau.c
1 /*
2 * Video Decode and Presentation API for UNIX (VDPAU) is used for
3 * HW decode acceleration for MPEG-1/2, MPEG-4 ASP, H.264 and VC-1.
4 *
5 * Copyright (c) 2008 NVIDIA
6 *
7 * This file is part of Libav.
8 *
9 * Libav is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * Libav is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with Libav; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 #include <limits.h>
25 #include "avcodec.h"
26 #include "h264.h"
27 #include "vc1.h"
28
29 #undef NDEBUG
30 #include <assert.h>
31
32 #include "vdpau.h"
33 #include "vdpau_internal.h"
34
35 /**
36 * @addtogroup VDPAU_Decoding
37 *
38 * @{
39 */
40
41 int ff_vdpau_common_start_frame(AVCodecContext *avctx,
42 av_unused const uint8_t *buffer,
43 av_unused uint32_t size)
44 {
45 AVVDPAUContext *hwctx = avctx->hwaccel_context;
46
47 hwctx->bitstream_buffers_used = 0;
48 return 0;
49 }
50
51 int ff_vdpau_mpeg_end_frame(AVCodecContext *avctx)
52 {
53 AVVDPAUContext *hwctx = avctx->hwaccel_context;
54 MpegEncContext *s = avctx->priv_data;
55 VdpVideoSurface surf = ff_vdpau_get_surface_id(s->current_picture_ptr);
56
57 hwctx->render(hwctx->decoder, surf, (void *)&hwctx->info,
58 hwctx->bitstream_buffers_used, hwctx->bitstream_buffers);
59
60 ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
61 hwctx->bitstream_buffers_used = 0;
62
63 return 0;
64 }
65
66 int ff_vdpau_add_buffer(AVCodecContext *avctx,
67 const uint8_t *buf, uint32_t size)
68 {
69 AVVDPAUContext *hwctx = avctx->hwaccel_context;
70 VdpBitstreamBuffer *buffers = hwctx->bitstream_buffers;
71
72 buffers = av_fast_realloc(buffers, &hwctx->bitstream_buffers_allocated,
73 (hwctx->bitstream_buffers_used + 1) * sizeof(*buffers));
74 if (!buffers)
75 return AVERROR(ENOMEM);
76
77 hwctx->bitstream_buffers = buffers;
78 buffers += hwctx->bitstream_buffers_used++;
79
80 buffers->struct_version = VDP_BITSTREAM_BUFFER_VERSION;
81 buffers->bitstream = buf;
82 buffers->bitstream_bytes = size;
83 return 0;
84 }
85
86 /* Obsolete non-hwaccel VDPAU support below... */
87
88 void ff_vdpau_h264_set_reference_frames(H264Context *h)
89 {
90 struct vdpau_render_state *render, *render_ref;
91 VdpReferenceFrameH264 *rf, *rf2;
92 Picture *pic;
93 int i, list, pic_frame_idx;
94
95 render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
96 assert(render);
97
98 rf = &render->info.h264.referenceFrames[0];
99 #define H264_RF_COUNT FF_ARRAY_ELEMS(render->info.h264.referenceFrames)
100
101 for (list = 0; list < 2; ++list) {
102 Picture **lp = list ? h->long_ref : h->short_ref;
103 int ls = list ? 16 : h->short_ref_count;
104
105 for (i = 0; i < ls; ++i) {
106 pic = lp[i];
107 if (!pic || !pic->reference)
108 continue;
109 pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
110
111 render_ref = (struct vdpau_render_state *)pic->f.data[0];
112 assert(render_ref);
113
114 rf2 = &render->info.h264.referenceFrames[0];
115 while (rf2 != rf) {
116 if (
117 (rf2->surface == render_ref->surface)
118 && (rf2->is_long_term == pic->long_ref)
119 && (rf2->frame_idx == pic_frame_idx)
120 )
121 break;
122 ++rf2;
123 }
124 if (rf2 != rf) {
125 rf2->top_is_reference |= (pic->reference & PICT_TOP_FIELD) ? VDP_TRUE : VDP_FALSE;
126 rf2->bottom_is_reference |= (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
127 continue;
128 }
129
130 if (rf >= &render->info.h264.referenceFrames[H264_RF_COUNT])
131 continue;
132
133 rf->surface = render_ref->surface;
134 rf->is_long_term = pic->long_ref;
135 rf->top_is_reference = (pic->reference & PICT_TOP_FIELD) ? VDP_TRUE : VDP_FALSE;
136 rf->bottom_is_reference = (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
137 rf->field_order_cnt[0] = pic->field_poc[0];
138 rf->field_order_cnt[1] = pic->field_poc[1];
139 rf->frame_idx = pic_frame_idx;
140
141 ++rf;
142 }
143 }
144
145 for (; rf < &render->info.h264.referenceFrames[H264_RF_COUNT]; ++rf) {
146 rf->surface = VDP_INVALID_HANDLE;
147 rf->is_long_term = 0;
148 rf->top_is_reference = 0;
149 rf->bottom_is_reference = 0;
150 rf->field_order_cnt[0] = 0;
151 rf->field_order_cnt[1] = 0;
152 rf->frame_idx = 0;
153 }
154 }
155
156 void ff_vdpau_add_data_chunk(uint8_t *data, const uint8_t *buf, int buf_size)
157 {
158 struct vdpau_render_state *render = (struct vdpau_render_state*)data;
159 assert(render);
160
161 render->bitstream_buffers= av_fast_realloc(
162 render->bitstream_buffers,
163 &render->bitstream_buffers_allocated,
164 sizeof(*render->bitstream_buffers)*(render->bitstream_buffers_used + 1)
165 );
166
167 render->bitstream_buffers[render->bitstream_buffers_used].struct_version = VDP_BITSTREAM_BUFFER_VERSION;
168 render->bitstream_buffers[render->bitstream_buffers_used].bitstream = buf;
169 render->bitstream_buffers[render->bitstream_buffers_used].bitstream_bytes = buf_size;
170 render->bitstream_buffers_used++;
171 }
172
173 void ff_vdpau_h264_picture_start(H264Context *h)
174 {
175 struct vdpau_render_state *render;
176 int i;
177
178 render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
179 assert(render);
180
181 for (i = 0; i < 2; ++i) {
182 int foc = h->cur_pic_ptr->field_poc[i];
183 if (foc == INT_MAX)
184 foc = 0;
185 render->info.h264.field_order_cnt[i] = foc;
186 }
187
188 render->info.h264.frame_num = h->frame_num;
189 }
190
191 void ff_vdpau_h264_picture_complete(H264Context *h)
192 {
193 struct vdpau_render_state *render;
194
195 render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
196 assert(render);
197
198 render->info.h264.slice_count = h->slice_num;
199 if (render->info.h264.slice_count < 1)
200 return;
201
202 render->info.h264.is_reference = (h->cur_pic_ptr->reference & 3) ? VDP_TRUE : VDP_FALSE;
203 render->info.h264.field_pic_flag = h->picture_structure != PICT_FRAME;
204 render->info.h264.bottom_field_flag = h->picture_structure == PICT_BOTTOM_FIELD;
205 render->info.h264.num_ref_frames = h->sps.ref_frame_count;
206 render->info.h264.mb_adaptive_frame_field_flag = h->sps.mb_aff && !render->info.h264.field_pic_flag;
207 render->info.h264.constrained_intra_pred_flag = h->pps.constrained_intra_pred;
208 render->info.h264.weighted_pred_flag = h->pps.weighted_pred;
209 render->info.h264.weighted_bipred_idc = h->pps.weighted_bipred_idc;
210 render->info.h264.frame_mbs_only_flag = h->sps.frame_mbs_only_flag;
211 render->info.h264.transform_8x8_mode_flag = h->pps.transform_8x8_mode;
212 render->info.h264.chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0];
213 render->info.h264.second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1];
214 render->info.h264.pic_init_qp_minus26 = h->pps.init_qp - 26;
215 render->info.h264.num_ref_idx_l0_active_minus1 = h->pps.ref_count[0] - 1;
216 render->info.h264.num_ref_idx_l1_active_minus1 = h->pps.ref_count[1] - 1;
217 render->info.h264.log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4;
218 render->info.h264.pic_order_cnt_type = h->sps.poc_type;
219 render->info.h264.log2_max_pic_order_cnt_lsb_minus4 = h->sps.poc_type ? 0 : h->sps.log2_max_poc_lsb - 4;
220 render->info.h264.delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
221 render->info.h264.direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag;
222 render->info.h264.entropy_coding_mode_flag = h->pps.cabac;
223 render->info.h264.pic_order_present_flag = h->pps.pic_order_present;
224 render->info.h264.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
225 render->info.h264.redundant_pic_cnt_present_flag = h->pps.redundant_pic_cnt_present;
226 memcpy(render->info.h264.scaling_lists_4x4, h->pps.scaling_matrix4, sizeof(render->info.h264.scaling_lists_4x4));
227 memcpy(render->info.h264.scaling_lists_8x8[0], h->pps.scaling_matrix8[0], sizeof(render->info.h264.scaling_lists_8x8[0]));
228 memcpy(render->info.h264.scaling_lists_8x8[1], h->pps.scaling_matrix8[3], sizeof(render->info.h264.scaling_lists_8x8[0]));
229
230 ff_h264_draw_horiz_band(h, 0, h->avctx->height);
231 render->bitstream_buffers_used = 0;
232 }
233
234 void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
235 int buf_size, int slice_count)
236 {
237 struct vdpau_render_state *render, *last, *next;
238 int i;
239
240 if (!s->current_picture_ptr) return;
241
242 render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
243 assert(render);
244
245 /* fill VdpPictureInfoMPEG1Or2 struct */
246 render->info.mpeg.picture_structure = s->picture_structure;
247 render->info.mpeg.picture_coding_type = s->pict_type;
248 render->info.mpeg.intra_dc_precision = s->intra_dc_precision;
249 render->info.mpeg.frame_pred_frame_dct = s->frame_pred_frame_dct;
250 render->info.mpeg.concealment_motion_vectors = s->concealment_motion_vectors;
251 render->info.mpeg.intra_vlc_format = s->intra_vlc_format;
252 render->info.mpeg.alternate_scan = s->alternate_scan;
253 render->info.mpeg.q_scale_type = s->q_scale_type;
254 render->info.mpeg.top_field_first = s->top_field_first;
255 render->info.mpeg.full_pel_forward_vector = s->full_pel[0]; // MPEG-1 only. Set 0 for MPEG-2
256 render->info.mpeg.full_pel_backward_vector = s->full_pel[1]; // MPEG-1 only. Set 0 for MPEG-2
257 render->info.mpeg.f_code[0][0] = s->mpeg_f_code[0][0]; // For MPEG-1 fill both horiz. & vert.
258 render->info.mpeg.f_code[0][1] = s->mpeg_f_code[0][1];
259 render->info.mpeg.f_code[1][0] = s->mpeg_f_code[1][0];
260 render->info.mpeg.f_code[1][1] = s->mpeg_f_code[1][1];
261 for (i = 0; i < 64; ++i) {
262 render->info.mpeg.intra_quantizer_matrix[i] = s->intra_matrix[i];
263 render->info.mpeg.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
264 }
265
266 render->info.mpeg.forward_reference = VDP_INVALID_HANDLE;
267 render->info.mpeg.backward_reference = VDP_INVALID_HANDLE;
268
269 switch(s->pict_type){
270 case AV_PICTURE_TYPE_B:
271 next = (struct vdpau_render_state *)s->next_picture.f.data[0];
272 assert(next);
273 render->info.mpeg.backward_reference = next->surface;
274 // no return here, going to set forward prediction
275 case AV_PICTURE_TYPE_P:
276 last = (struct vdpau_render_state *)s->last_picture.f.data[0];
277 if (!last) // FIXME: Does this test make sense?
278 last = render; // predict second field from the first
279 render->info.mpeg.forward_reference = last->surface;
280 }
281
282 ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
283
284 render->info.mpeg.slice_count = slice_count;
285
286 if (slice_count)
287 ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
288 render->bitstream_buffers_used = 0;
289 }
290
291 void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
292 int buf_size)
293 {
294 VC1Context *v = s->avctx->priv_data;
295 struct vdpau_render_state *render, *last, *next;
296
297 render = (struct vdpau_render_state *)s->current_picture.f.data[0];
298 assert(render);
299
300 /* fill LvPictureInfoVC1 struct */
301 render->info.vc1.frame_coding_mode = v->fcm;
302 render->info.vc1.postprocflag = v->postprocflag;
303 render->info.vc1.pulldown = v->broadcast;
304 render->info.vc1.interlace = v->interlace;
305 render->info.vc1.tfcntrflag = v->tfcntrflag;
306 render->info.vc1.finterpflag = v->finterpflag;
307 render->info.vc1.psf = v->psf;
308 render->info.vc1.dquant = v->dquant;
309 render->info.vc1.panscan_flag = v->panscanflag;
310 render->info.vc1.refdist_flag = v->refdist_flag;
311 render->info.vc1.quantizer = v->quantizer_mode;
312 render->info.vc1.extended_mv = v->extended_mv;
313 render->info.vc1.extended_dmv = v->extended_dmv;
314 render->info.vc1.overlap = v->overlap;
315 render->info.vc1.vstransform = v->vstransform;
316 render->info.vc1.loopfilter = v->s.loop_filter;
317 render->info.vc1.fastuvmc = v->fastuvmc;
318 render->info.vc1.range_mapy_flag = v->range_mapy_flag;
319 render->info.vc1.range_mapy = v->range_mapy;
320 render->info.vc1.range_mapuv_flag = v->range_mapuv_flag;
321 render->info.vc1.range_mapuv = v->range_mapuv;
322 /* Specific to simple/main profile only */
323 render->info.vc1.multires = v->multires;
324 render->info.vc1.syncmarker = v->s.resync_marker;
325 render->info.vc1.rangered = v->rangered | (v->rangeredfrm << 1);
326 render->info.vc1.maxbframes = v->s.max_b_frames;
327
328 render->info.vc1.deblockEnable = v->postprocflag & 1;
329 render->info.vc1.pquant = v->pq;
330
331 render->info.vc1.forward_reference = VDP_INVALID_HANDLE;
332 render->info.vc1.backward_reference = VDP_INVALID_HANDLE;
333
334 if (v->bi_type)
335 render->info.vc1.picture_type = 4;
336 else
337 render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
338
339 switch(s->pict_type){
340 case AV_PICTURE_TYPE_B:
341 next = (struct vdpau_render_state *)s->next_picture.f.data[0];
342 assert(next);
343 render->info.vc1.backward_reference = next->surface;
344 // no break here, going to set forward prediction
345 case AV_PICTURE_TYPE_P:
346 last = (struct vdpau_render_state *)s->last_picture.f.data[0];
347 if (!last) // FIXME: Does this test make sense?
348 last = render; // predict second field from the first
349 render->info.vc1.forward_reference = last->surface;
350 }
351
352 ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
353
354 render->info.vc1.slice_count = 1;
355
356 ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
357 render->bitstream_buffers_used = 0;
358 }
359
360 void ff_vdpau_mpeg4_decode_picture(MpegEncContext *s, const uint8_t *buf,
361 int buf_size)
362 {
363 struct vdpau_render_state *render, *last, *next;
364 int i;
365
366 if (!s->current_picture_ptr) return;
367
368 render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
369 assert(render);
370
371 /* fill VdpPictureInfoMPEG4Part2 struct */
372 render->info.mpeg4.trd[0] = s->pp_time;
373 render->info.mpeg4.trb[0] = s->pb_time;
374 render->info.mpeg4.trd[1] = s->pp_field_time >> 1;
375 render->info.mpeg4.trb[1] = s->pb_field_time >> 1;
376 render->info.mpeg4.vop_time_increment_resolution = s->avctx->time_base.den;
377 render->info.mpeg4.vop_coding_type = 0;
378 render->info.mpeg4.vop_fcode_forward = s->f_code;
379 render->info.mpeg4.vop_fcode_backward = s->b_code;
380 render->info.mpeg4.resync_marker_disable = !s->resync_marker;
381 render->info.mpeg4.interlaced = !s->progressive_sequence;
382 render->info.mpeg4.quant_type = s->mpeg_quant;
383 render->info.mpeg4.quarter_sample = s->quarter_sample;
384 render->info.mpeg4.short_video_header = s->avctx->codec->id == AV_CODEC_ID_H263;
385 render->info.mpeg4.rounding_control = s->no_rounding;
386 render->info.mpeg4.alternate_vertical_scan_flag = s->alternate_scan;
387 render->info.mpeg4.top_field_first = s->top_field_first;
388 for (i = 0; i < 64; ++i) {
389 render->info.mpeg4.intra_quantizer_matrix[i] = s->intra_matrix[i];
390 render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
391 }
392 render->info.mpeg4.forward_reference = VDP_INVALID_HANDLE;
393 render->info.mpeg4.backward_reference = VDP_INVALID_HANDLE;
394
395 switch (s->pict_type) {
396 case AV_PICTURE_TYPE_B:
397 next = (struct vdpau_render_state *)s->next_picture.f.data[0];
398 assert(next);
399 render->info.mpeg4.backward_reference = next->surface;
400 render->info.mpeg4.vop_coding_type = 2;
401 // no break here, going to set forward prediction
402 case AV_PICTURE_TYPE_P:
403 last = (struct vdpau_render_state *)s->last_picture.f.data[0];
404 assert(last);
405 render->info.mpeg4.forward_reference = last->surface;
406 }
407
408 ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
409
410 ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
411 render->bitstream_buffers_used = 0;
412 }
413
414 /* @}*/