]> git.sesse.net Git - ffmpeg/blob - libavcodec/vdpau.c
rv40: don't always do the full prev_type search
[ffmpeg] / 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 void ff_vdpau_h264_set_reference_frames(MpegEncContext *s)
42 {
43     H264Context *h = s->avctx->priv_data;
44     struct vdpau_render_state *render, *render_ref;
45     VdpReferenceFrameH264 *rf, *rf2;
46     Picture *pic;
47     int i, list, pic_frame_idx;
48
49     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
50     assert(render);
51
52     rf = &render->info.h264.referenceFrames[0];
53 #define H264_RF_COUNT FF_ARRAY_ELEMS(render->info.h264.referenceFrames)
54
55     for (list = 0; list < 2; ++list) {
56         Picture **lp = list ? h->long_ref : h->short_ref;
57         int ls = list ? 16 : h->short_ref_count;
58
59         for (i = 0; i < ls; ++i) {
60             pic = lp[i];
61             if (!pic || !pic->f.reference)
62                 continue;
63             pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
64
65             render_ref = (struct vdpau_render_state *)pic->f.data[0];
66             assert(render_ref);
67
68             rf2 = &render->info.h264.referenceFrames[0];
69             while (rf2 != rf) {
70                 if (
71                     (rf2->surface == render_ref->surface)
72                     && (rf2->is_long_term == pic->long_ref)
73                     && (rf2->frame_idx == pic_frame_idx)
74                 )
75                     break;
76                 ++rf2;
77             }
78             if (rf2 != rf) {
79                 rf2->top_is_reference    |= (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
80                 rf2->bottom_is_reference |= (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
81                 continue;
82             }
83
84             if (rf >= &render->info.h264.referenceFrames[H264_RF_COUNT])
85                 continue;
86
87             rf->surface             = render_ref->surface;
88             rf->is_long_term        = pic->long_ref;
89             rf->top_is_reference    = (pic->f.reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
90             rf->bottom_is_reference = (pic->f.reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
91             rf->field_order_cnt[0]  = pic->field_poc[0];
92             rf->field_order_cnt[1]  = pic->field_poc[1];
93             rf->frame_idx           = pic_frame_idx;
94
95             ++rf;
96         }
97     }
98
99     for (; rf < &render->info.h264.referenceFrames[H264_RF_COUNT]; ++rf) {
100         rf->surface             = VDP_INVALID_HANDLE;
101         rf->is_long_term        = 0;
102         rf->top_is_reference    = 0;
103         rf->bottom_is_reference = 0;
104         rf->field_order_cnt[0]  = 0;
105         rf->field_order_cnt[1]  = 0;
106         rf->frame_idx           = 0;
107     }
108 }
109
110 void ff_vdpau_add_data_chunk(MpegEncContext *s,
111                              const uint8_t *buf, int buf_size)
112 {
113     struct vdpau_render_state *render;
114
115     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
116     assert(render);
117
118     render->bitstream_buffers= av_fast_realloc(
119         render->bitstream_buffers,
120         &render->bitstream_buffers_allocated,
121         sizeof(*render->bitstream_buffers)*(render->bitstream_buffers_used + 1)
122     );
123
124     render->bitstream_buffers[render->bitstream_buffers_used].struct_version  = VDP_BITSTREAM_BUFFER_VERSION;
125     render->bitstream_buffers[render->bitstream_buffers_used].bitstream       = buf;
126     render->bitstream_buffers[render->bitstream_buffers_used].bitstream_bytes = buf_size;
127     render->bitstream_buffers_used++;
128 }
129
130 void ff_vdpau_h264_picture_start(MpegEncContext *s)
131 {
132     H264Context *h = s->avctx->priv_data;
133     struct vdpau_render_state *render;
134     int i;
135
136     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
137     assert(render);
138
139     for (i = 0; i < 2; ++i) {
140         int foc = s->current_picture_ptr->field_poc[i];
141         if (foc == INT_MAX)
142             foc = 0;
143         render->info.h264.field_order_cnt[i] = foc;
144     }
145
146     render->info.h264.frame_num = h->frame_num;
147 }
148
149 void ff_vdpau_h264_picture_complete(MpegEncContext *s)
150 {
151     H264Context *h = s->avctx->priv_data;
152     struct vdpau_render_state *render;
153
154     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
155     assert(render);
156
157     render->info.h264.slice_count = h->slice_num;
158     if (render->info.h264.slice_count < 1)
159         return;
160
161     render->info.h264.is_reference                           = (s->current_picture_ptr->f.reference & 3) ? VDP_TRUE : VDP_FALSE;
162     render->info.h264.field_pic_flag                         = s->picture_structure != PICT_FRAME;
163     render->info.h264.bottom_field_flag                      = s->picture_structure == PICT_BOTTOM_FIELD;
164     render->info.h264.num_ref_frames                         = h->sps.ref_frame_count;
165     render->info.h264.mb_adaptive_frame_field_flag           = h->sps.mb_aff && !render->info.h264.field_pic_flag;
166     render->info.h264.constrained_intra_pred_flag            = h->pps.constrained_intra_pred;
167     render->info.h264.weighted_pred_flag                     = h->pps.weighted_pred;
168     render->info.h264.weighted_bipred_idc                    = h->pps.weighted_bipred_idc;
169     render->info.h264.frame_mbs_only_flag                    = h->sps.frame_mbs_only_flag;
170     render->info.h264.transform_8x8_mode_flag                = h->pps.transform_8x8_mode;
171     render->info.h264.chroma_qp_index_offset                 = h->pps.chroma_qp_index_offset[0];
172     render->info.h264.second_chroma_qp_index_offset          = h->pps.chroma_qp_index_offset[1];
173     render->info.h264.pic_init_qp_minus26                    = h->pps.init_qp - 26;
174     render->info.h264.num_ref_idx_l0_active_minus1           = h->pps.ref_count[0] - 1;
175     render->info.h264.num_ref_idx_l1_active_minus1           = h->pps.ref_count[1] - 1;
176     render->info.h264.log2_max_frame_num_minus4              = h->sps.log2_max_frame_num - 4;
177     render->info.h264.pic_order_cnt_type                     = h->sps.poc_type;
178     render->info.h264.log2_max_pic_order_cnt_lsb_minus4      = h->sps.poc_type ? 0 : h->sps.log2_max_poc_lsb - 4;
179     render->info.h264.delta_pic_order_always_zero_flag       = h->sps.delta_pic_order_always_zero_flag;
180     render->info.h264.direct_8x8_inference_flag              = h->sps.direct_8x8_inference_flag;
181     render->info.h264.entropy_coding_mode_flag               = h->pps.cabac;
182     render->info.h264.pic_order_present_flag                 = h->pps.pic_order_present;
183     render->info.h264.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
184     render->info.h264.redundant_pic_cnt_present_flag         = h->pps.redundant_pic_cnt_present;
185     memcpy(render->info.h264.scaling_lists_4x4, h->pps.scaling_matrix4, sizeof(render->info.h264.scaling_lists_4x4));
186     memcpy(render->info.h264.scaling_lists_8x8[0], h->pps.scaling_matrix8[0], sizeof(render->info.h264.scaling_lists_8x8[0]));
187     memcpy(render->info.h264.scaling_lists_8x8[1], h->pps.scaling_matrix8[3], sizeof(render->info.h264.scaling_lists_8x8[0]));
188
189     ff_draw_horiz_band(s, 0, s->avctx->height);
190     render->bitstream_buffers_used = 0;
191 }
192
193 void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
194                                     int buf_size, int slice_count)
195 {
196     struct vdpau_render_state *render, *last, *next;
197     int i;
198
199     if (!s->current_picture_ptr) return;
200
201     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
202     assert(render);
203
204     /* fill VdpPictureInfoMPEG1Or2 struct */
205     render->info.mpeg.picture_structure          = s->picture_structure;
206     render->info.mpeg.picture_coding_type        = s->pict_type;
207     render->info.mpeg.intra_dc_precision         = s->intra_dc_precision;
208     render->info.mpeg.frame_pred_frame_dct       = s->frame_pred_frame_dct;
209     render->info.mpeg.concealment_motion_vectors = s->concealment_motion_vectors;
210     render->info.mpeg.intra_vlc_format           = s->intra_vlc_format;
211     render->info.mpeg.alternate_scan             = s->alternate_scan;
212     render->info.mpeg.q_scale_type               = s->q_scale_type;
213     render->info.mpeg.top_field_first            = s->top_field_first;
214     render->info.mpeg.full_pel_forward_vector    = s->full_pel[0]; // MPEG-1 only.  Set 0 for MPEG-2
215     render->info.mpeg.full_pel_backward_vector   = s->full_pel[1]; // MPEG-1 only.  Set 0 for MPEG-2
216     render->info.mpeg.f_code[0][0]               = s->mpeg_f_code[0][0]; // For MPEG-1 fill both horiz. & vert.
217     render->info.mpeg.f_code[0][1]               = s->mpeg_f_code[0][1];
218     render->info.mpeg.f_code[1][0]               = s->mpeg_f_code[1][0];
219     render->info.mpeg.f_code[1][1]               = s->mpeg_f_code[1][1];
220     for (i = 0; i < 64; ++i) {
221         render->info.mpeg.intra_quantizer_matrix[i]     = s->intra_matrix[i];
222         render->info.mpeg.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
223     }
224
225     render->info.mpeg.forward_reference          = VDP_INVALID_HANDLE;
226     render->info.mpeg.backward_reference         = VDP_INVALID_HANDLE;
227
228     switch(s->pict_type){
229     case  AV_PICTURE_TYPE_B:
230         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
231         assert(next);
232         render->info.mpeg.backward_reference     = next->surface;
233         // no return here, going to set forward prediction
234     case  AV_PICTURE_TYPE_P:
235         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
236         if (!last) // FIXME: Does this test make sense?
237             last = render; // predict second field from the first
238         render->info.mpeg.forward_reference      = last->surface;
239     }
240
241     ff_vdpau_add_data_chunk(s, buf, buf_size);
242
243     render->info.mpeg.slice_count                = slice_count;
244
245     if (slice_count)
246         ff_draw_horiz_band(s, 0, s->avctx->height);
247     render->bitstream_buffers_used               = 0;
248 }
249
250 void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
251                                  int buf_size)
252 {
253     VC1Context *v = s->avctx->priv_data;
254     struct vdpau_render_state *render, *last, *next;
255
256     render = (struct vdpau_render_state *)s->current_picture.f.data[0];
257     assert(render);
258
259     /*  fill LvPictureInfoVC1 struct */
260     render->info.vc1.frame_coding_mode  = v->fcm;
261     render->info.vc1.postprocflag       = v->postprocflag;
262     render->info.vc1.pulldown           = v->broadcast;
263     render->info.vc1.interlace          = v->interlace;
264     render->info.vc1.tfcntrflag         = v->tfcntrflag;
265     render->info.vc1.finterpflag        = v->finterpflag;
266     render->info.vc1.psf                = v->psf;
267     render->info.vc1.dquant             = v->dquant;
268     render->info.vc1.panscan_flag       = v->panscanflag;
269     render->info.vc1.refdist_flag       = v->refdist_flag;
270     render->info.vc1.quantizer          = v->quantizer_mode;
271     render->info.vc1.extended_mv        = v->extended_mv;
272     render->info.vc1.extended_dmv       = v->extended_dmv;
273     render->info.vc1.overlap            = v->overlap;
274     render->info.vc1.vstransform        = v->vstransform;
275     render->info.vc1.loopfilter         = v->s.loop_filter;
276     render->info.vc1.fastuvmc           = v->fastuvmc;
277     render->info.vc1.range_mapy_flag    = v->range_mapy_flag;
278     render->info.vc1.range_mapy         = v->range_mapy;
279     render->info.vc1.range_mapuv_flag   = v->range_mapuv_flag;
280     render->info.vc1.range_mapuv        = v->range_mapuv;
281     /* Specific to simple/main profile only */
282     render->info.vc1.multires           = v->multires;
283     render->info.vc1.syncmarker         = v->s.resync_marker;
284     render->info.vc1.rangered           = v->rangered | (v->rangeredfrm << 1);
285     render->info.vc1.maxbframes         = v->s.max_b_frames;
286
287     render->info.vc1.deblockEnable      = v->postprocflag & 1;
288     render->info.vc1.pquant             = v->pq;
289
290     render->info.vc1.forward_reference  = VDP_INVALID_HANDLE;
291     render->info.vc1.backward_reference = VDP_INVALID_HANDLE;
292
293     if (v->bi_type)
294         render->info.vc1.picture_type = 4;
295     else
296         render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
297
298     switch(s->pict_type){
299     case  AV_PICTURE_TYPE_B:
300         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
301         assert(next);
302         render->info.vc1.backward_reference = next->surface;
303         // no break here, going to set forward prediction
304     case  AV_PICTURE_TYPE_P:
305         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
306         if (!last) // FIXME: Does this test make sense?
307             last = render; // predict second field from the first
308         render->info.vc1.forward_reference = last->surface;
309     }
310
311     ff_vdpau_add_data_chunk(s, buf, buf_size);
312
313     render->info.vc1.slice_count          = 1;
314
315     ff_draw_horiz_band(s, 0, s->avctx->height);
316     render->bitstream_buffers_used        = 0;
317 }
318
319 void ff_vdpau_mpeg4_decode_picture(MpegEncContext *s, const uint8_t *buf,
320                                    int buf_size)
321 {
322     struct vdpau_render_state *render, *last, *next;
323     int i;
324
325     if (!s->current_picture_ptr) return;
326
327     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
328     assert(render);
329
330     /* fill VdpPictureInfoMPEG4Part2 struct */
331     render->info.mpeg4.trd[0]                            = s->pp_time;
332     render->info.mpeg4.trb[0]                            = s->pb_time;
333     render->info.mpeg4.trd[1]                            = s->pp_field_time >> 1;
334     render->info.mpeg4.trb[1]                            = s->pb_field_time >> 1;
335     render->info.mpeg4.vop_time_increment_resolution     = s->avctx->time_base.den;
336     render->info.mpeg4.vop_coding_type                   = 0;
337     render->info.mpeg4.vop_fcode_forward                 = s->f_code;
338     render->info.mpeg4.vop_fcode_backward                = s->b_code;
339     render->info.mpeg4.resync_marker_disable             = !s->resync_marker;
340     render->info.mpeg4.interlaced                        = !s->progressive_sequence;
341     render->info.mpeg4.quant_type                        = s->mpeg_quant;
342     render->info.mpeg4.quarter_sample                    = s->quarter_sample;
343     render->info.mpeg4.short_video_header                = s->avctx->codec->id == CODEC_ID_H263;
344     render->info.mpeg4.rounding_control                  = s->no_rounding;
345     render->info.mpeg4.alternate_vertical_scan_flag      = s->alternate_scan;
346     render->info.mpeg4.top_field_first                   = s->top_field_first;
347     for (i = 0; i < 64; ++i) {
348         render->info.mpeg4.intra_quantizer_matrix[i]     = s->intra_matrix[i];
349         render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
350     }
351     render->info.mpeg4.forward_reference                 = VDP_INVALID_HANDLE;
352     render->info.mpeg4.backward_reference                = VDP_INVALID_HANDLE;
353
354     switch (s->pict_type) {
355     case AV_PICTURE_TYPE_B:
356         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
357         assert(next);
358         render->info.mpeg4.backward_reference     = next->surface;
359         render->info.mpeg4.vop_coding_type        = 2;
360         // no break here, going to set forward prediction
361     case AV_PICTURE_TYPE_P:
362         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
363         assert(last);
364         render->info.mpeg4.forward_reference      = last->surface;
365     }
366
367     ff_vdpau_add_data_chunk(s, buf, buf_size);
368
369     ff_draw_horiz_band(s, 0, s->avctx->height);
370     render->bitstream_buffers_used = 0;
371 }
372
373 /* @}*/