pthread_frame: do not run hwaccel decoding asynchronously unless it's safe
[ffmpeg.git] / libavcodec / vaapi_vc1.c
1 /*
2  * VC-1 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 "hwaccel.h"
24 #include "internal.h"
25 #include "vaapi_decode.h"
26 #include "vc1.h"
27 #include "vc1data.h"
28
29 /** Translate FFmpeg MV modes to VA API */
30 static int get_VAMvModeVC1(enum MVModes mv_mode)
31 {
32     switch (mv_mode) {
33     case MV_PMODE_1MV_HPEL_BILIN: return VAMvMode1MvHalfPelBilinear;
34     case MV_PMODE_1MV:            return VAMvMode1Mv;
35     case MV_PMODE_1MV_HPEL:       return VAMvMode1MvHalfPel;
36     case MV_PMODE_MIXED_MV:       return VAMvModeMixedMv;
37     case MV_PMODE_INTENSITY_COMP: return VAMvModeIntensityCompensation;
38     }
39     return 0;
40 }
41
42 /** Check whether the MVTYPEMB bitplane is present */
43 static inline int vc1_has_MVTYPEMB_bitplane(const VC1Context *v)
44 {
45     if (v->mv_type_is_raw)
46         return 0;
47     return v->s.pict_type == AV_PICTURE_TYPE_P &&
48            (v->mv_mode == MV_PMODE_MIXED_MV ||
49             (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
50              v->mv_mode2 == MV_PMODE_MIXED_MV));
51 }
52
53 /** Check whether the SKIPMB bitplane is present */
54 static inline int vc1_has_SKIPMB_bitplane(const VC1Context *v)
55 {
56     if (v->skip_is_raw)
57         return 0;
58     return v->s.pict_type == AV_PICTURE_TYPE_P ||
59            (v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type);
60 }
61
62 /** Check whether the DIRECTMB bitplane is present */
63 static inline int vc1_has_DIRECTMB_bitplane(const VC1Context *v)
64 {
65     if (v->dmb_is_raw)
66         return 0;
67     return v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
68 }
69
70 /** Check whether the ACPRED bitplane is present */
71 static inline int vc1_has_ACPRED_bitplane(const VC1Context *v)
72 {
73     if (v->acpred_is_raw)
74         return 0;
75     return v->profile == PROFILE_ADVANCED &&
76            (v->s.pict_type == AV_PICTURE_TYPE_I ||
77             (v->s.pict_type == AV_PICTURE_TYPE_B && v->bi_type));
78 }
79
80 /** Check whether the OVERFLAGS bitplane is present */
81 static inline int vc1_has_OVERFLAGS_bitplane(const VC1Context *v)
82 {
83     if (v->overflg_is_raw)
84         return 0;
85     return v->profile == PROFILE_ADVANCED &&
86            (v->s.pict_type == AV_PICTURE_TYPE_I ||
87             (v->s.pict_type == AV_PICTURE_TYPE_B && v->bi_type)) &&
88            (v->overlap && v->pq <= 8) &&
89            v->condover == CONDOVER_SELECT;
90 }
91
92 /** Reconstruct bitstream PTYPE (7.1.1.4, index into Table-35) */
93 static int vc1_get_PTYPE(const VC1Context *v)
94 {
95     const MpegEncContext *s = &v->s;
96     switch (s->pict_type) {
97     case AV_PICTURE_TYPE_I: return 0;
98     case AV_PICTURE_TYPE_P: return v->p_frame_skipped ? 4 : 1;
99     case AV_PICTURE_TYPE_B: return v->bi_type         ? 3 : 2;
100     }
101     return 0;
102 }
103
104 /** Reconstruct bitstream MVMODE (7.1.1.32) */
105 static inline VAMvModeVC1 vc1_get_MVMODE(const VC1Context *v)
106 {
107     if (v->s.pict_type == AV_PICTURE_TYPE_P ||
108         (v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type))
109         return get_VAMvModeVC1(v->mv_mode);
110     return 0;
111 }
112
113 /** Reconstruct bitstream MVMODE2 (7.1.1.33) */
114 static inline VAMvModeVC1 vc1_get_MVMODE2(const VC1Context *v)
115 {
116     if (v->s.pict_type == AV_PICTURE_TYPE_P && v->mv_mode == MV_PMODE_INTENSITY_COMP)
117         return get_VAMvModeVC1(v->mv_mode2);
118     return 0;
119 }
120
121 /** Reconstruct bitstream TTFRM (7.1.1.41, Table-53) */
122 static inline int vc1_get_TTFRM(const VC1Context *v)
123 {
124     switch (v->ttfrm) {
125     case TT_8X8: return 0;
126     case TT_8X4: return 1;
127     case TT_4X8: return 2;
128     case TT_4X4: return 3;
129     }
130     return 0;
131 }
132
133 /** Pack FFmpeg bitplanes into a VABitPlaneBuffer element */
134 static inline void vc1_pack_bitplanes(uint8_t *bitplane, int n, const uint8_t *ff_bp[3], int x, int y, int stride)
135 {
136     const int bitplane_index = n / 2;
137     const int ff_bp_index = y * stride + x;
138     uint8_t v = 0;
139     if (ff_bp[0])
140         v = ff_bp[0][ff_bp_index];
141     if (ff_bp[1])
142         v |= ff_bp[1][ff_bp_index] << 1;
143     if (ff_bp[2])
144         v |= ff_bp[2][ff_bp_index] << 2;
145     bitplane[bitplane_index] = (bitplane[bitplane_index] << 4) | v;
146 }
147
148 static int vaapi_vc1_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size)
149 {
150     const VC1Context *v = avctx->priv_data;
151     const MpegEncContext *s = &v->s;
152     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
153     VAPictureParameterBufferVC1 pic_param;
154     int err;
155
156     pic->output_surface = ff_vaapi_get_surface_id(s->current_picture_ptr->f);
157
158     pic_param = (VAPictureParameterBufferVC1) {
159         .forward_reference_picture         = VA_INVALID_ID,
160         .backward_reference_picture        = VA_INVALID_ID,
161         .inloop_decoded_picture            = VA_INVALID_ID,
162         .sequence_fields.bits = {
163             .pulldown                      = v->broadcast,
164             .interlace                     = v->interlace,
165             .tfcntrflag                    = v->tfcntrflag,
166             .finterpflag                   = v->finterpflag,
167             .psf                           = v->psf,
168             .multires                      = v->multires,
169             .overlap                       = v->overlap,
170             .syncmarker                    = v->resync_marker,
171             .rangered                      = v->rangered,
172             .max_b_frames                  = s->avctx->max_b_frames,
173             .profile                       = v->profile,
174         },
175         .coded_width                       = s->avctx->coded_width,
176         .coded_height                      = s->avctx->coded_height,
177         .entrypoint_fields.bits = {
178             .broken_link                   = v->broken_link,
179             .closed_entry                  = v->closed_entry,
180             .panscan_flag                  = v->panscanflag,
181             .loopfilter                    = s->loop_filter,
182         },
183         .conditional_overlap_flag          = v->condover,
184         .fast_uvmc_flag                    = v->fastuvmc,
185         .range_mapping_fields.bits = {
186             .luma_flag                     = v->range_mapy_flag,
187             .luma                          = v->range_mapy,
188             .chroma_flag                   = v->range_mapuv_flag,
189             .chroma                        = v->range_mapuv,
190         },
191         .b_picture_fraction                = v->bfraction_lut_index,
192         .cbp_table                         = v->cbpcy_vlc ? v->cbpcy_vlc - ff_vc1_cbpcy_p_vlc : 0,
193         .mb_mode_table                     = 0, /* XXX: interlaced frame */
194         .range_reduction_frame             = v->rangeredfrm,
195         .rounding_control                  = v->rnd,
196         .post_processing                   = v->postproc,
197         .picture_resolution_index          = v->respic,
198         .luma_scale                        = v->lumscale,
199         .luma_shift                        = v->lumshift,
200         .picture_fields.bits = {
201             .picture_type                  = vc1_get_PTYPE(v),
202             .frame_coding_mode             = v->fcm,
203             .top_field_first               = v->tff,
204             .is_first_field                = v->fcm == 0, /* XXX: interlaced frame */
205             .intensity_compensation        = v->mv_mode == MV_PMODE_INTENSITY_COMP,
206         },
207         .raw_coding.flags = {
208             .mv_type_mb                    = v->mv_type_is_raw,
209             .direct_mb                     = v->dmb_is_raw,
210             .skip_mb                       = v->skip_is_raw,
211             .field_tx                      = 0, /* XXX: interlaced frame */
212             .forward_mb                    = 0, /* XXX: interlaced frame */
213             .ac_pred                       = v->acpred_is_raw,
214             .overflags                     = v->overflg_is_raw,
215         },
216         .bitplane_present.flags = {
217             .bp_mv_type_mb                 = vc1_has_MVTYPEMB_bitplane(v),
218             .bp_direct_mb                  = vc1_has_DIRECTMB_bitplane(v),
219             .bp_skip_mb                    = vc1_has_SKIPMB_bitplane(v),
220             .bp_field_tx                   = 0, /* XXX: interlaced frame */
221             .bp_forward_mb                 = 0, /* XXX: interlaced frame */
222             .bp_ac_pred                    = vc1_has_ACPRED_bitplane(v),
223             .bp_overflags                  = vc1_has_OVERFLAGS_bitplane(v),
224         },
225         .reference_fields.bits = {
226             .reference_distance_flag       = v->refdist_flag,
227             .reference_distance            = 0, /* XXX: interlaced frame */
228             .num_reference_pictures        = 0, /* XXX: interlaced frame */
229             .reference_field_pic_indicator = 0, /* XXX: interlaced frame */
230         },
231         .mv_fields.bits = {
232             .mv_mode                       = vc1_get_MVMODE(v),
233             .mv_mode2                      = vc1_get_MVMODE2(v),
234             .mv_table                      = s->mv_table_index,
235             .two_mv_block_pattern_table    = 0, /* XXX: interlaced frame */
236             .four_mv_switch                = 0, /* XXX: interlaced frame */
237             .four_mv_block_pattern_table   = 0, /* XXX: interlaced frame */
238             .extended_mv_flag              = v->extended_mv,
239             .extended_mv_range             = v->mvrange,
240             .extended_dmv_flag             = v->extended_dmv,
241             .extended_dmv_range            = 0, /* XXX: interlaced frame */
242         },
243         .pic_quantizer_fields.bits = {
244             .dquant                        = v->dquant,
245             .quantizer                     = v->quantizer_mode,
246             .half_qp                       = v->halfpq,
247             .pic_quantizer_scale           = v->pq,
248             .pic_quantizer_type            = v->pquantizer,
249             .dq_frame                      = v->dquantfrm,
250             .dq_profile                    = v->dqprofile,
251             .dq_sb_edge                    = v->dqprofile == DQPROFILE_SINGLE_EDGE  ? v->dqsbedge : 0,
252             .dq_db_edge                    = v->dqprofile == DQPROFILE_DOUBLE_EDGES ? v->dqsbedge : 0,
253             .dq_binary_level               = v->dqbilevel,
254             .alt_pic_quantizer             = v->altpq,
255         },
256         .transform_fields.bits = {
257             .variable_sized_transform_flag = v->vstransform,
258             .mb_level_transform_type_flag  = v->ttmbf,
259             .frame_level_transform_type    = vc1_get_TTFRM(v),
260             .transform_ac_codingset_idx1   = v->c_ac_table_index,
261             .transform_ac_codingset_idx2   = v->y_ac_table_index,
262             .intra_transform_dc_table      = v->s.dc_table_index,
263         },
264     };
265
266     switch (s->pict_type) {
267     case AV_PICTURE_TYPE_B:
268         pic_param.backward_reference_picture = ff_vaapi_get_surface_id(s->next_picture.f);
269         // fall-through
270     case AV_PICTURE_TYPE_P:
271         pic_param.forward_reference_picture = ff_vaapi_get_surface_id(s->last_picture.f);
272         break;
273     }
274
275     err = ff_vaapi_decode_make_param_buffer(avctx, pic,
276                                             VAPictureParameterBufferType,
277                                             &pic_param, sizeof(pic_param));
278     if (err)
279         goto fail;
280
281     if (pic_param.bitplane_present.value) {
282         uint8_t *bitplane;
283         const uint8_t *ff_bp[3];
284         int x, y, n;
285         size_t size = (s->mb_width * s->mb_height + 1) / 2;
286
287         bitplane = av_mallocz(size);
288         if (!bitplane) {
289             err = AVERROR(ENOMEM);
290             goto fail;
291         }
292
293         switch (s->pict_type) {
294         case AV_PICTURE_TYPE_P:
295             ff_bp[0] = pic_param.bitplane_present.flags.bp_direct_mb  ? v->direct_mb_plane    : NULL;
296             ff_bp[1] = pic_param.bitplane_present.flags.bp_skip_mb    ? s->mbskip_table       : NULL;
297             ff_bp[2] = pic_param.bitplane_present.flags.bp_mv_type_mb ? v->mv_type_mb_plane   : NULL;
298             break;
299         case AV_PICTURE_TYPE_B:
300             if (!v->bi_type) {
301                 ff_bp[0] = pic_param.bitplane_present.flags.bp_direct_mb ? v->direct_mb_plane : NULL;
302                 ff_bp[1] = pic_param.bitplane_present.flags.bp_skip_mb   ? s->mbskip_table    : NULL;
303                 ff_bp[2] = NULL; /* XXX: interlaced frame (FORWARD plane) */
304                 break;
305             }
306             /* fall-through (BI-type) */
307         case AV_PICTURE_TYPE_I:
308             ff_bp[0] = NULL; /* XXX: interlaced frame (FIELDTX plane) */
309             ff_bp[1] = pic_param.bitplane_present.flags.bp_ac_pred    ? v->acpred_plane       : NULL;
310             ff_bp[2] = pic_param.bitplane_present.flags.bp_overflags  ? v->over_flags_plane   : NULL;
311             break;
312         default:
313             ff_bp[0] = NULL;
314             ff_bp[1] = NULL;
315             ff_bp[2] = NULL;
316             break;
317         }
318
319         n = 0;
320         for (y = 0; y < s->mb_height; y++)
321             for (x = 0; x < s->mb_width; x++, n++)
322                 vc1_pack_bitplanes(bitplane, n, ff_bp, x, y, s->mb_stride);
323         if (n & 1) /* move last nibble to the high order */
324             bitplane[n/2] <<= 4;
325
326         err = ff_vaapi_decode_make_param_buffer(avctx, pic,
327                                                 VABitPlaneBufferType,
328                                                 bitplane, size);
329         av_free(bitplane);
330         if (err)
331             goto fail;
332     }
333     return 0;
334
335 fail:
336     ff_vaapi_decode_cancel(avctx, pic);
337     return err;
338 }
339
340 static int vaapi_vc1_end_frame(AVCodecContext *avctx)
341 {
342     VC1Context *v = avctx->priv_data;
343     MpegEncContext *s = &v->s;
344     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
345     int ret;
346
347     ret = ff_vaapi_decode_issue(avctx, pic);
348     if (ret < 0)
349         goto fail;
350
351     ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
352
353 fail:
354     return ret;
355 }
356
357 static int vaapi_vc1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
358 {
359     const VC1Context *v = avctx->priv_data;
360     const MpegEncContext *s = &v->s;
361     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
362     VASliceParameterBufferVC1 slice_param;
363     int err;
364
365     /* Current bit buffer is beyond any marker for VC-1, so skip it */
366     if (avctx->codec_id == AV_CODEC_ID_VC1 && IS_MARKER(AV_RB32(buffer))) {
367         buffer += 4;
368         size -= 4;
369     }
370
371     slice_param = (VASliceParameterBufferVC1) {
372         .slice_data_size         = size,
373         .slice_data_offset       = 0,
374         .slice_data_flag         = VA_SLICE_DATA_FLAG_ALL,
375         .macroblock_offset       = get_bits_count(&s->gb),
376         .slice_vertical_position = s->mb_y,
377     };
378
379     err = ff_vaapi_decode_make_slice_buffer(avctx, pic,
380                                             &slice_param, sizeof(slice_param),
381                                             buffer, size);
382     if (err < 0) {
383         ff_vaapi_decode_cancel(avctx, pic);
384         return err;
385     }
386
387     return 0;
388 }
389
390 #if CONFIG_WMV3_VAAPI_HWACCEL
391 AVHWAccel ff_wmv3_vaapi_hwaccel = {
392     .name                 = "wmv3_vaapi",
393     .type                 = AVMEDIA_TYPE_VIDEO,
394     .id                   = AV_CODEC_ID_WMV3,
395     .pix_fmt              = AV_PIX_FMT_VAAPI,
396     .start_frame          = &vaapi_vc1_start_frame,
397     .end_frame            = &vaapi_vc1_end_frame,
398     .decode_slice         = &vaapi_vc1_decode_slice,
399     .frame_priv_data_size = sizeof(VAAPIDecodePicture),
400     .init                 = &ff_vaapi_decode_init,
401     .uninit               = &ff_vaapi_decode_uninit,
402     .priv_data_size       = sizeof(VAAPIDecodeContext),
403     .caps_internal        = HWACCEL_CAP_ASYNC_SAFE,
404 };
405 #endif
406
407 AVHWAccel ff_vc1_vaapi_hwaccel = {
408     .name                 = "vc1_vaapi",
409     .type                 = AVMEDIA_TYPE_VIDEO,
410     .id                   = AV_CODEC_ID_VC1,
411     .pix_fmt              = AV_PIX_FMT_VAAPI,
412     .start_frame          = &vaapi_vc1_start_frame,
413     .end_frame            = &vaapi_vc1_end_frame,
414     .decode_slice         = &vaapi_vc1_decode_slice,
415     .frame_priv_data_size = sizeof(VAAPIDecodePicture),
416     .init                 = &ff_vaapi_decode_init,
417     .uninit               = &ff_vaapi_decode_uninit,
418     .priv_data_size       = sizeof(VAAPIDecodeContext),
419     .caps_internal        = HWACCEL_CAP_ASYNC_SAFE,
420 };