Optimize decodeplane32().
[ffmpeg.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 FFmpeg.
8  *
9  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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->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->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->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->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
80                 rf2->bottom_is_reference |= (pic->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->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
90             rf->bottom_is_reference = (pic->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->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->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->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->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, h->pps.scaling_matrix8, sizeof(render->info.h264.scaling_lists_8x8));
187
188     ff_draw_horiz_band(s, 0, s->avctx->height);
189     render->bitstream_buffers_used = 0;
190 }
191
192 void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
193                                     int buf_size, int slice_count)
194 {
195     struct vdpau_render_state *render, *last, *next;
196     int i;
197
198     if (!s->current_picture_ptr) return;
199
200     render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
201     assert(render);
202
203     /* fill VdpPictureInfoMPEG1Or2 struct */
204     render->info.mpeg.picture_structure          = s->picture_structure;
205     render->info.mpeg.picture_coding_type        = s->pict_type;
206     render->info.mpeg.intra_dc_precision         = s->intra_dc_precision;
207     render->info.mpeg.frame_pred_frame_dct       = s->frame_pred_frame_dct;
208     render->info.mpeg.concealment_motion_vectors = s->concealment_motion_vectors;
209     render->info.mpeg.intra_vlc_format           = s->intra_vlc_format;
210     render->info.mpeg.alternate_scan             = s->alternate_scan;
211     render->info.mpeg.q_scale_type               = s->q_scale_type;
212     render->info.mpeg.top_field_first            = s->top_field_first;
213     render->info.mpeg.full_pel_forward_vector    = s->full_pel[0]; // MPEG-1 only.  Set 0 for MPEG-2
214     render->info.mpeg.full_pel_backward_vector   = s->full_pel[1]; // MPEG-1 only.  Set 0 for MPEG-2
215     render->info.mpeg.f_code[0][0]               = s->mpeg_f_code[0][0]; // For MPEG-1 fill both horiz. & vert.
216     render->info.mpeg.f_code[0][1]               = s->mpeg_f_code[0][1];
217     render->info.mpeg.f_code[1][0]               = s->mpeg_f_code[1][0];
218     render->info.mpeg.f_code[1][1]               = s->mpeg_f_code[1][1];
219     for (i = 0; i < 64; ++i) {
220         render->info.mpeg.intra_quantizer_matrix[i]     = s->intra_matrix[i];
221         render->info.mpeg.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
222     }
223
224     render->info.mpeg.forward_reference          = VDP_INVALID_HANDLE;
225     render->info.mpeg.backward_reference         = VDP_INVALID_HANDLE;
226
227     switch(s->pict_type){
228     case  FF_B_TYPE:
229         next = (struct vdpau_render_state *)s->next_picture.data[0];
230         assert(next);
231         render->info.mpeg.backward_reference     = next->surface;
232         // no return here, going to set forward prediction
233     case  FF_P_TYPE:
234         last = (struct vdpau_render_state *)s->last_picture.data[0];
235         if (!last) // FIXME: Does this test make sense?
236             last = render; // predict second field from the first
237         render->info.mpeg.forward_reference      = last->surface;
238     }
239
240     ff_vdpau_add_data_chunk(s, buf, buf_size);
241
242     render->info.mpeg.slice_count                = slice_count;
243
244     if (slice_count)
245         ff_draw_horiz_band(s, 0, s->avctx->height);
246     render->bitstream_buffers_used               = 0;
247 }
248
249 void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
250                                  int buf_size)
251 {
252     VC1Context *v = s->avctx->priv_data;
253     struct vdpau_render_state *render, *last, *next;
254
255     render = (struct vdpau_render_state *)s->current_picture.data[0];
256     assert(render);
257
258     /*  fill LvPictureInfoVC1 struct */
259     render->info.vc1.frame_coding_mode  = v->fcm;
260     render->info.vc1.postprocflag       = v->postprocflag;
261     render->info.vc1.pulldown           = v->broadcast;
262     render->info.vc1.interlace          = v->interlace;
263     render->info.vc1.tfcntrflag         = v->tfcntrflag;
264     render->info.vc1.finterpflag        = v->finterpflag;
265     render->info.vc1.psf                = v->psf;
266     render->info.vc1.dquant             = v->dquant;
267     render->info.vc1.panscan_flag       = v->panscanflag;
268     render->info.vc1.refdist_flag       = v->refdist_flag;
269     render->info.vc1.quantizer          = v->quantizer_mode;
270     render->info.vc1.extended_mv        = v->extended_mv;
271     render->info.vc1.extended_dmv       = v->extended_dmv;
272     render->info.vc1.overlap            = v->overlap;
273     render->info.vc1.vstransform        = v->vstransform;
274     render->info.vc1.loopfilter         = v->s.loop_filter;
275     render->info.vc1.fastuvmc           = v->fastuvmc;
276     render->info.vc1.range_mapy_flag    = v->range_mapy_flag;
277     render->info.vc1.range_mapy         = v->range_mapy;
278     render->info.vc1.range_mapuv_flag   = v->range_mapuv_flag;
279     render->info.vc1.range_mapuv        = v->range_mapuv;
280     /* Specific to simple/main profile only */
281     render->info.vc1.multires           = v->multires;
282     render->info.vc1.syncmarker         = v->s.resync_marker;
283     render->info.vc1.rangered           = v->rangered | (v->rangeredfrm << 1);
284     render->info.vc1.maxbframes         = v->s.max_b_frames;
285
286     render->info.vc1.deblockEnable      = v->postprocflag & 1;
287     render->info.vc1.pquant             = v->pq;
288
289     render->info.vc1.forward_reference  = VDP_INVALID_HANDLE;
290     render->info.vc1.backward_reference = VDP_INVALID_HANDLE;
291
292     if (v->bi_type)
293         render->info.vc1.picture_type = 4;
294     else
295         render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
296
297     switch(s->pict_type){
298     case  FF_B_TYPE:
299         next = (struct vdpau_render_state *)s->next_picture.data[0];
300         assert(next);
301         render->info.vc1.backward_reference = next->surface;
302         // no break here, going to set forward prediction
303     case  FF_P_TYPE:
304         last = (struct vdpau_render_state *)s->last_picture.data[0];
305         if (!last) // FIXME: Does this test make sense?
306             last = render; // predict second field from the first
307         render->info.vc1.forward_reference = last->surface;
308     }
309
310     ff_vdpau_add_data_chunk(s, buf, buf_size);
311
312     render->info.vc1.slice_count          = 1;
313
314     ff_draw_horiz_band(s, 0, s->avctx->height);
315     render->bitstream_buffers_used        = 0;
316 }
317
318 void ff_vdpau_mpeg4_decode_picture(MpegEncContext *s, const uint8_t *buf,
319                                    int buf_size)
320 {
321     struct vdpau_render_state *render, *last, *next;
322     int i;
323
324     if (!s->current_picture_ptr) return;
325
326     render = (struct vdpau_render_state *)s->current_picture_ptr->data[0];
327     assert(render);
328
329     /* fill VdpPictureInfoMPEG4Part2 struct */
330     render->info.mpeg4.trd[0]                            = s->pp_time;
331     render->info.mpeg4.trb[0]                            = s->pb_time;
332     render->info.mpeg4.trd[1]                            = s->pp_field_time >> 1;
333     render->info.mpeg4.trb[1]                            = s->pb_field_time >> 1;
334     render->info.mpeg4.vop_time_increment_resolution     = s->avctx->time_base.den;
335     render->info.mpeg4.vop_coding_type                   = 0;
336     render->info.mpeg4.vop_fcode_forward                 = s->f_code;
337     render->info.mpeg4.vop_fcode_backward                = s->b_code;
338     render->info.mpeg4.resync_marker_disable             = !s->resync_marker;
339     render->info.mpeg4.interlaced                        = !s->progressive_sequence;
340     render->info.mpeg4.quant_type                        = s->mpeg_quant;
341     render->info.mpeg4.quarter_sample                    = s->quarter_sample;
342     render->info.mpeg4.short_video_header                = s->avctx->codec->id == CODEC_ID_H263;
343     render->info.mpeg4.rounding_control                  = s->no_rounding;
344     render->info.mpeg4.alternate_vertical_scan_flag      = s->alternate_scan;
345     render->info.mpeg4.top_field_first                   = s->top_field_first;
346     for (i = 0; i < 64; ++i) {
347         render->info.mpeg4.intra_quantizer_matrix[i]     = s->intra_matrix[i];
348         render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
349     }
350     render->info.mpeg4.forward_reference                 = VDP_INVALID_HANDLE;
351     render->info.mpeg4.backward_reference                = VDP_INVALID_HANDLE;
352
353     switch (s->pict_type) {
354     case FF_B_TYPE:
355         next = (struct vdpau_render_state *)s->next_picture.data[0];
356         assert(next);
357         render->info.mpeg4.backward_reference     = next->surface;
358         render->info.mpeg4.vop_coding_type        = 2;
359         // no break here, going to set forward prediction
360     case FF_P_TYPE:
361         last = (struct vdpau_render_state *)s->last_picture.data[0];
362         assert(last);
363         render->info.mpeg4.forward_reference      = last->surface;
364     }
365
366     ff_vdpau_add_data_chunk(s, buf, buf_size);
367
368     ff_draw_horiz_band(s, 0, s->avctx->height);
369     render->bitstream_buffers_used = 0;
370 }
371
372 /* @}*/