a456149e6e4ce96f6b10012d20c9a023cec86f78
[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 "internal.h"
24 #include "vaapi_decode.h"
25 #include "vc1.h"
26 #include "vc1data.h"
27
28 /** Translate FFmpeg MV modes to VA API */
29 static int get_VAMvModeVC1(enum MVModes mv_mode)
30 {
31     switch (mv_mode) {
32     case MV_PMODE_1MV_HPEL_BILIN: return VAMvMode1MvHalfPelBilinear;
33     case MV_PMODE_1MV:            return VAMvMode1Mv;
34     case MV_PMODE_1MV_HPEL:       return VAMvMode1MvHalfPel;
35     case MV_PMODE_MIXED_MV:       return VAMvModeMixedMv;
36     case MV_PMODE_INTENSITY_COMP: return VAMvModeIntensityCompensation;
37     }
38     return 0;
39 }
40
41 /** Check whether the MVTYPEMB bitplane is present */
42 static inline int vc1_has_MVTYPEMB_bitplane(const VC1Context *v)
43 {
44     if (v->mv_type_is_raw)
45         return 0;
46     return v->s.pict_type == AV_PICTURE_TYPE_P &&
47            (v->mv_mode == MV_PMODE_MIXED_MV ||
48             (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
49              v->mv_mode2 == MV_PMODE_MIXED_MV));
50 }
51
52 /** Check whether the SKIPMB bitplane is present */
53 static inline int vc1_has_SKIPMB_bitplane(const VC1Context *v)
54 {
55     if (v->skip_is_raw)
56         return 0;
57     return v->s.pict_type == AV_PICTURE_TYPE_P ||
58            (v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type);
59 }
60
61 /** Check whether the DIRECTMB bitplane is present */
62 static inline int vc1_has_DIRECTMB_bitplane(const VC1Context *v)
63 {
64     if (v->dmb_is_raw)
65         return 0;
66     return v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
67 }
68
69 /** Check whether the ACPRED bitplane is present */
70 static inline int vc1_has_ACPRED_bitplane(const VC1Context *v)
71 {
72     if (v->acpred_is_raw)
73         return 0;
74     return v->profile == PROFILE_ADVANCED &&
75            (v->s.pict_type == AV_PICTURE_TYPE_I ||
76             (v->s.pict_type == AV_PICTURE_TYPE_B && v->bi_type));
77 }
78
79 /** Check whether the OVERFLAGS bitplane is present */
80 static inline int vc1_has_OVERFLAGS_bitplane(const VC1Context *v)
81 {
82     if (v->overflg_is_raw)
83         return 0;
84     return v->profile == PROFILE_ADVANCED &&
85            (v->s.pict_type == AV_PICTURE_TYPE_I ||
86             (v->s.pict_type == AV_PICTURE_TYPE_B && v->bi_type)) &&
87            (v->overlap && v->pq <= 8) &&
88            v->condover == CONDOVER_SELECT;
89 }
90
91 /** Reconstruct bitstream PTYPE (7.1.1.4, index into Table-35) */
92 static int vc1_get_PTYPE(const VC1Context *v)
93 {
94     const MpegEncContext *s = &v->s;
95     switch (s->pict_type) {
96     case AV_PICTURE_TYPE_I: return 0;
97     case AV_PICTURE_TYPE_P: return v->p_frame_skipped ? 4 : 1;
98     case AV_PICTURE_TYPE_B: return v->bi_type         ? 3 : 2;
99     }
100     return 0;
101 }
102
103 /** Reconstruct bitstream MVMODE (7.1.1.32) */
104 static inline VAMvModeVC1 vc1_get_MVMODE(const VC1Context *v)
105 {
106     if (v->s.pict_type == AV_PICTURE_TYPE_P ||
107         (v->s.pict_type == AV_PICTURE_TYPE_B && !v->bi_type))
108         return get_VAMvModeVC1(v->mv_mode);
109     return 0;
110 }
111
112 /** Reconstruct bitstream MVMODE2 (7.1.1.33) */
113 static inline VAMvModeVC1 vc1_get_MVMODE2(const VC1Context *v)
114 {
115     if (v->s.pict_type == AV_PICTURE_TYPE_P && v->mv_mode == MV_PMODE_INTENSITY_COMP)
116         return get_VAMvModeVC1(v->mv_mode2);
117     return 0;
118 }
119
120 /** Reconstruct bitstream TTFRM (7.1.1.41, Table-53) */
121 static inline int vc1_get_TTFRM(const VC1Context *v)
122 {
123     switch (v->ttfrm) {
124     case TT_8X8: return 0;
125     case TT_8X4: return 1;
126     case TT_4X8: return 2;
127     case TT_4X4: return 3;
128     }
129     return 0;
130 }
131
132 /** Pack FFmpeg bitplanes into a VABitPlaneBuffer element */
133 static inline void vc1_pack_bitplanes(uint8_t *bitplane, int n, const uint8_t *ff_bp[3], int x, int y, int stride)
134 {
135     const int bitplane_index = n / 2;
136     const int ff_bp_index = y * stride + x;
137     uint8_t v = 0;
138     if (ff_bp[0])
139         v = ff_bp[0][ff_bp_index];
140     if (ff_bp[1])
141         v |= ff_bp[1][ff_bp_index] << 1;
142     if (ff_bp[2])
143         v |= ff_bp[2][ff_bp_index] << 2;
144     bitplane[bitplane_index] = (bitplane[bitplane_index] << 4) | v;
145 }
146
147 static int vaapi_vc1_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size)
148 {
149     const VC1Context *v = avctx->priv_data;
150     const MpegEncContext *s = &v->s;
151     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
152     VAPictureParameterBufferVC1 pic_param;
153     int err;
154
155     pic->output_surface = ff_vaapi_get_surface_id(s->current_picture_ptr->f);
156
157     pic_param = (VAPictureParameterBufferVC1) {
158         .forward_reference_picture         = VA_INVALID_ID,
159         .backward_reference_picture        = VA_INVALID_ID,
160         .inloop_decoded_picture            = VA_INVALID_ID,
161         .sequence_fields.bits = {
162             .pulldown                      = v->broadcast,
163             .interlace                     = v->interlace,
164             .tfcntrflag                    = v->tfcntrflag,
165             .finterpflag                   = v->finterpflag,
166             .psf                           = v->psf,
167             .multires                      = v->multires,
168             .overlap                       = v->overlap,
169             .syncmarker                    = v->resync_marker,
170             .rangered                      = v->rangered,
171             .max_b_frames                  = s->avctx->max_b_frames,
172             .profile                       = v->profile,
173         },
174         .coded_width                       = s->avctx->coded_width,
175         .coded_height                      = s->avctx->coded_height,
176         .entrypoint_fields.bits = {
177             .broken_link                   = v->broken_link,
178             .closed_entry                  = v->closed_entry,
179             .panscan_flag                  = v->panscanflag,
180             .loopfilter                    = s->loop_filter,
181         },
182         .conditional_overlap_flag          = v->condover,
183         .fast_uvmc_flag                    = v->fastuvmc,
184         .range_mapping_fields.bits = {
185             .luma_flag                     = v->range_mapy_flag,
186             .luma                          = v->range_mapy,
187             .chroma_flag                   = v->range_mapuv_flag,
188             .chroma                        = v->range_mapuv,
189         },
190         .b_picture_fraction                = v->bfraction_lut_index,
191         .cbp_table                         = v->cbpcy_vlc ? v->cbpcy_vlc - ff_vc1_cbpcy_p_vlc : 0,
192         .mb_mode_table                     = 0, /* XXX: interlaced frame */
193         .range_reduction_frame             = v->rangeredfrm,
194         .rounding_control                  = v->rnd,
195         .post_processing                   = v->postproc,
196         .picture_resolution_index          = v->respic,
197         .luma_scale                        = v->lumscale,
198         .luma_shift                        = v->lumshift,
199         .picture_fields.bits = {
200             .picture_type                  = vc1_get_PTYPE(v),
201             .frame_coding_mode             = v->fcm,
202             .top_field_first               = v->tff,
203             .is_first_field                = v->fcm == 0, /* XXX: interlaced frame */
204             .intensity_compensation        = v->mv_mode == MV_PMODE_INTENSITY_COMP,
205         },
206         .raw_coding.flags = {
207             .mv_type_mb                    = v->mv_type_is_raw,
208             .direct_mb                     = v->dmb_is_raw,
209             .skip_mb                       = v->skip_is_raw,
210             .field_tx                      = 0, /* XXX: interlaced frame */
211             .forward_mb                    = 0, /* XXX: interlaced frame */
212             .ac_pred                       = v->acpred_is_raw,
213             .overflags                     = v->overflg_is_raw,
214         },
215         .bitplane_present.flags = {
216             .bp_mv_type_mb                 = vc1_has_MVTYPEMB_bitplane(v),
217             .bp_direct_mb                  = vc1_has_DIRECTMB_bitplane(v),
218             .bp_skip_mb                    = vc1_has_SKIPMB_bitplane(v),
219             .bp_field_tx                   = 0, /* XXX: interlaced frame */
220             .bp_forward_mb                 = 0, /* XXX: interlaced frame */
221             .bp_ac_pred                    = vc1_has_ACPRED_bitplane(v),
222             .bp_overflags                  = vc1_has_OVERFLAGS_bitplane(v),
223         },
224         .reference_fields.bits = {
225             .reference_distance_flag       = v->refdist_flag,
226             .reference_distance            = 0, /* XXX: interlaced frame */
227             .num_reference_pictures        = 0, /* XXX: interlaced frame */
228             .reference_field_pic_indicator = 0, /* XXX: interlaced frame */
229         },
230         .mv_fields.bits = {
231             .mv_mode                       = vc1_get_MVMODE(v),
232             .mv_mode2                      = vc1_get_MVMODE2(v),
233             .mv_table                      = s->mv_table_index,
234             .two_mv_block_pattern_table    = 0, /* XXX: interlaced frame */
235             .four_mv_switch                = 0, /* XXX: interlaced frame */
236             .four_mv_block_pattern_table   = 0, /* XXX: interlaced frame */
237             .extended_mv_flag              = v->extended_mv,
238             .extended_mv_range             = v->mvrange,
239             .extended_dmv_flag             = v->extended_dmv,
240             .extended_dmv_range            = 0, /* XXX: interlaced frame */
241         },
242         .pic_quantizer_fields.bits = {
243             .dquant                        = v->dquant,
244             .quantizer                     = v->quantizer_mode,
245             .half_qp                       = v->halfpq,
246             .pic_quantizer_scale           = v->pq,
247             .pic_quantizer_type            = v->pquantizer,
248             .dq_frame                      = v->dquantfrm,
249             .dq_profile                    = v->dqprofile,
250             .dq_sb_edge                    = v->dqprofile == DQPROFILE_SINGLE_EDGE  ? v->dqsbedge : 0,
251             .dq_db_edge                    = v->dqprofile == DQPROFILE_DOUBLE_EDGES ? v->dqsbedge : 0,
252             .dq_binary_level               = v->dqbilevel,
253             .alt_pic_quantizer             = v->altpq,
254         },
255         .transform_fields.bits = {
256             .variable_sized_transform_flag = v->vstransform,
257             .mb_level_transform_type_flag  = v->ttmbf,
258             .frame_level_transform_type    = vc1_get_TTFRM(v),
259             .transform_ac_codingset_idx1   = v->c_ac_table_index,
260             .transform_ac_codingset_idx2   = v->y_ac_table_index,
261             .intra_transform_dc_table      = v->s.dc_table_index,
262         },
263     };
264
265     switch (s->pict_type) {
266     case AV_PICTURE_TYPE_B:
267         pic_param.backward_reference_picture = ff_vaapi_get_surface_id(s->next_picture.f);
268         // fall-through
269     case AV_PICTURE_TYPE_P:
270         pic_param.forward_reference_picture = ff_vaapi_get_surface_id(s->last_picture.f);
271         break;
272     }
273
274     err = ff_vaapi_decode_make_param_buffer(avctx, pic,
275                                             VAPictureParameterBufferType,
276                                             &pic_param, sizeof(pic_param));
277     if (err)
278         goto fail;
279
280     if (pic_param.bitplane_present.value) {
281         uint8_t *bitplane;
282         const uint8_t *ff_bp[3];
283         int x, y, n;
284         size_t size = (s->mb_width * s->mb_height + 1) / 2;
285
286         bitplane = av_mallocz(size);
287         if (!bitplane) {
288             err = AVERROR(ENOMEM);
289             goto fail;
290         }
291
292         switch (s->pict_type) {
293         case AV_PICTURE_TYPE_P:
294             ff_bp[0] = pic_param.bitplane_present.flags.bp_direct_mb  ? v->direct_mb_plane    : NULL;
295             ff_bp[1] = pic_param.bitplane_present.flags.bp_skip_mb    ? s->mbskip_table       : NULL;
296             ff_bp[2] = pic_param.bitplane_present.flags.bp_mv_type_mb ? v->mv_type_mb_plane   : NULL;
297             break;
298         case AV_PICTURE_TYPE_B:
299             if (!v->bi_type) {
300                 ff_bp[0] = pic_param.bitplane_present.flags.bp_direct_mb ? v->direct_mb_plane : NULL;
301                 ff_bp[1] = pic_param.bitplane_present.flags.bp_skip_mb   ? s->mbskip_table    : NULL;
302                 ff_bp[2] = NULL; /* XXX: interlaced frame (FORWARD plane) */
303                 break;
304             }
305             /* fall-through (BI-type) */
306         case AV_PICTURE_TYPE_I:
307             ff_bp[0] = NULL; /* XXX: interlaced frame (FIELDTX plane) */
308             ff_bp[1] = pic_param.bitplane_present.flags.bp_ac_pred    ? v->acpred_plane       : NULL;
309             ff_bp[2] = pic_param.bitplane_present.flags.bp_overflags  ? v->over_flags_plane   : NULL;
310             break;
311         default:
312             ff_bp[0] = NULL;
313             ff_bp[1] = NULL;
314             ff_bp[2] = NULL;
315             break;
316         }
317
318         n = 0;
319         for (y = 0; y < s->mb_height; y++)
320             for (x = 0; x < s->mb_width; x++, n++)
321                 vc1_pack_bitplanes(bitplane, n, ff_bp, x, y, s->mb_stride);
322         if (n & 1) /* move last nibble to the high order */
323             bitplane[n/2] <<= 4;
324
325         err = ff_vaapi_decode_make_param_buffer(avctx, pic,
326                                                 VABitPlaneBufferType,
327                                                 bitplane, size);
328         av_free(bitplane);
329         if (err)
330             goto fail;
331     }
332     return 0;
333
334 fail:
335     ff_vaapi_decode_cancel(avctx, pic);
336     return err;
337 }
338
339 static int vaapi_vc1_end_frame(AVCodecContext *avctx)
340 {
341     VC1Context *v = avctx->priv_data;
342     MpegEncContext *s = &v->s;
343     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
344     int ret;
345
346     ret = ff_vaapi_decode_issue(avctx, pic);
347     if (ret < 0)
348         goto fail;
349
350     ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
351
352 fail:
353     return ret;
354 }
355
356 static int vaapi_vc1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
357 {
358     const VC1Context *v = avctx->priv_data;
359     const MpegEncContext *s = &v->s;
360     VAAPIDecodePicture *pic = s->current_picture_ptr->hwaccel_picture_private;
361     VASliceParameterBufferVC1 slice_param;
362     int err;
363
364     /* Current bit buffer is beyond any marker for VC-1, so skip it */
365     if (avctx->codec_id == AV_CODEC_ID_VC1 && IS_MARKER(AV_RB32(buffer))) {
366         buffer += 4;
367         size -= 4;
368     }
369
370     slice_param = (VASliceParameterBufferVC1) {
371         .slice_data_size         = size,
372         .slice_data_offset       = 0,
373         .slice_data_flag         = VA_SLICE_DATA_FLAG_ALL,
374         .macroblock_offset       = get_bits_count(&s->gb),
375         .slice_vertical_position = s->mb_y,
376     };
377
378     err = ff_vaapi_decode_make_slice_buffer(avctx, pic,
379                                             &slice_param, sizeof(slice_param),
380                                             buffer, size);
381     if (err < 0) {
382         ff_vaapi_decode_cancel(avctx, pic);
383         return err;
384     }
385
386     return 0;
387 }
388
389 #if CONFIG_WMV3_VAAPI_HWACCEL
390 AVHWAccel ff_wmv3_vaapi_hwaccel = {
391     .name                 = "wmv3_vaapi",
392     .type                 = AVMEDIA_TYPE_VIDEO,
393     .id                   = AV_CODEC_ID_WMV3,
394     .pix_fmt              = AV_PIX_FMT_VAAPI,
395     .start_frame          = &vaapi_vc1_start_frame,
396     .end_frame            = &vaapi_vc1_end_frame,
397     .decode_slice         = &vaapi_vc1_decode_slice,
398     .frame_priv_data_size = sizeof(VAAPIDecodePicture),
399     .init                 = &ff_vaapi_decode_init,
400     .uninit               = &ff_vaapi_decode_uninit,
401     .priv_data_size       = sizeof(VAAPIDecodeContext),
402 };
403 #endif
404
405 AVHWAccel ff_vc1_vaapi_hwaccel = {
406     .name                 = "vc1_vaapi",
407     .type                 = AVMEDIA_TYPE_VIDEO,
408     .id                   = AV_CODEC_ID_VC1,
409     .pix_fmt              = AV_PIX_FMT_VAAPI,
410     .start_frame          = &vaapi_vc1_start_frame,
411     .end_frame            = &vaapi_vc1_end_frame,
412     .decode_slice         = &vaapi_vc1_decode_slice,
413     .frame_priv_data_size = sizeof(VAAPIDecodePicture),
414     .init                 = &ff_vaapi_decode_init,
415     .uninit               = &ff_vaapi_decode_uninit,
416     .priv_data_size       = sizeof(VAAPIDecodeContext),
417 };