Update to MPlayer SVN rev 30134 and FFmpeg SVN rev 20947.
[vaapi:mplayer.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 "vaapi_internal.h"
24 #include "vc1.h"
25 #include "vc1data.h"
26
27 /** Translates FFmpeg MV modes to VA API */
28 static int get_VAMvModeVC1(enum MVModes mv_mode)
29 {
30     switch (mv_mode) {
31     case MV_PMODE_1MV_HPEL_BILIN: return VAMvMode1MvHalfPelBilinear;
32     case MV_PMODE_1MV:            return VAMvMode1Mv;
33     case MV_PMODE_1MV_HPEL:       return VAMvMode1MvHalfPel;
34     case MV_PMODE_MIXED_MV:       return VAMvModeMixedMv;
35     case MV_PMODE_INTENSITY_COMP: return VAMvModeIntensityCompensation;
36     }
37     return 0;
38 }
39
40 /** Checks whether the MVTYPEMB bitplane is present */
41 static inline int vc1_has_MVTYPEMB_bitplane(VC1Context *v)
42 {
43     if (v->mv_type_is_raw)
44         return 0;
45     return (v->s.pict_type == FF_P_TYPE &&
46             (v->mv_mode == MV_PMODE_MIXED_MV ||
47              (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
48               v->mv_mode2 == MV_PMODE_MIXED_MV)));
49 }
50
51 /** Checks whether the SKIPMB bitplane is present */
52 static inline int vc1_has_SKIPMB_bitplane(VC1Context *v)
53 {
54     if (v->skip_is_raw)
55         return 0;
56     return (v->s.pict_type == FF_P_TYPE ||
57             (v->s.pict_type == FF_B_TYPE && !v->bi_type));
58 }
59
60 /** Checks whether the DIRECTMB bitplane is present */
61 static inline int vc1_has_DIRECTMB_bitplane(VC1Context *v)
62 {
63     if (v->dmb_is_raw)
64         return 0;
65     return v->s.pict_type == FF_B_TYPE && !v->bi_type;
66 }
67
68 /** Checks whether the ACPRED bitplane is present */
69 static inline int vc1_has_ACPRED_bitplane(VC1Context *v)
70 {
71     if (v->acpred_is_raw)
72         return 0;
73     return (v->profile == PROFILE_ADVANCED &&
74             (v->s.pict_type == FF_I_TYPE ||
75              (v->s.pict_type == FF_B_TYPE && v->bi_type)));
76 }
77
78 /** Check whether the OVERFLAGS bitplane is present */
79 static inline int vc1_has_OVERFLAGS_bitplane(VC1Context *v)
80 {
81     if (v->overflg_is_raw)
82         return 0;
83     return (v->profile == PROFILE_ADVANCED &&
84             (v->s.pict_type == FF_I_TYPE ||
85              (v->s.pict_type == FF_B_TYPE && v->bi_type)) &&
86             (v->overlap && v->pq <= 8) &&
87             v->condover == CONDOVER_SELECT);
88 }
89
90 /** Reconstruct bitstream PTYPE (7.1.1.4, index into Table-35) */
91 static int vc1_get_PTYPE(VC1Context *v)
92 {
93     MpegEncContext * const s = &v->s;
94     switch (s->pict_type) {
95     case FF_I_TYPE: return 0;
96     case FF_P_TYPE: return v->p_frame_skipped ? 4 : 1;
97     case FF_B_TYPE: return v->bi_type         ? 3 : 2;
98     }
99     return 0;
100 }
101
102 /** Reconstruct bitstream MVMODE (7.1.1.32) */
103 static inline VAMvModeVC1 vc1_get_MVMODE(VC1Context *v)
104 {
105     if (v->s.pict_type == FF_P_TYPE ||
106         (v->s.pict_type == FF_B_TYPE && !v->bi_type))
107         return get_VAMvModeVC1(v->mv_mode);
108     return 0;
109 }
110
111 /** Reconstruct bitstream MVMODE2 (7.1.1.33) */
112 static inline VAMvModeVC1 vc1_get_MVMODE2(VC1Context *v)
113 {
114     if (v->s.pict_type == FF_P_TYPE && v->mv_mode == MV_PMODE_INTENSITY_COMP)
115         return get_VAMvModeVC1(v->mv_mode2);
116     return 0;
117 }
118
119 /** Pack FFmpeg bitplanes into a VABitPlaneBuffer element */
120 static inline uint8_t vc1_pack_bitplanes(const uint8_t *ff_bp[3], int x, int y, int stride)
121 {
122     const int n = y * stride + x;
123     uint8_t v = 0;
124     if (ff_bp[0])
125         v = ff_bp[0][n];
126     if (ff_bp[1])
127         v |= ff_bp[1][n] << 1;
128     if (ff_bp[2])
129         v |= ff_bp[2][n] << 2;
130     return v;
131 }
132
133 static int vaapi_vc1_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size)
134 {
135     VC1Context * const v = avctx->priv_data;
136     MpegEncContext * const s = &v->s;
137     struct vaapi_context * const vactx = avctx->hwaccel_context;
138     VAPictureParameterBufferVC1 *pic_param;
139
140     dprintf(avctx, "vaapi_vc1_start_frame()\n");
141
142     vactx->slice_param_size = sizeof(VASliceParameterBufferVC1);
143
144     /* Fill in VAPictureParameterBufferVC1 */
145     pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferVC1));
146     if (!pic_param)
147         return -1;
148     pic_param->forward_reference_picture                            = 0xffffffff;
149     pic_param->backward_reference_picture                           = 0xffffffff;
150     pic_param->inloop_decoded_picture                               = 0xffffffff;
151     pic_param->sequence_fields.value                                = 0; /* reset all bits */
152     pic_param->sequence_fields.bits.pulldown                        = v->broadcast;
153     pic_param->sequence_fields.bits.interlace                       = v->interlace;
154     pic_param->sequence_fields.bits.tfcntrflag                      = v->tfcntrflag;
155     pic_param->sequence_fields.bits.finterpflag                     = v->finterpflag;
156     pic_param->sequence_fields.bits.psf                             = v->psf;
157     pic_param->sequence_fields.bits.multires                        = v->multires;
158     pic_param->sequence_fields.bits.overlap                         = v->overlap;
159     pic_param->sequence_fields.bits.syncmarker                      = s->resync_marker;
160     pic_param->sequence_fields.bits.rangered                        = v->rangered;
161     pic_param->sequence_fields.bits.max_b_frames                    = s->avctx->max_b_frames;
162     pic_param->coded_width                                          = s->avctx->coded_width;
163     pic_param->coded_height                                         = s->avctx->coded_height;
164     pic_param->entrypoint_fields.value                              = 0; /* reset all bits */
165     pic_param->entrypoint_fields.bits.broken_link                   = v->broken_link;
166     pic_param->entrypoint_fields.bits.closed_entry                  = v->closed_entry;
167     pic_param->entrypoint_fields.bits.panscan_flag                  = v->panscanflag;
168     pic_param->entrypoint_fields.bits.loopfilter                    = s->loop_filter;
169     pic_param->conditional_overlap_flag                             = v->condover;
170     pic_param->fast_uvmc_flag                                       = v->fastuvmc;
171     pic_param->range_mapping_fields.value                           = 0; /* reset all bits */
172     pic_param->range_mapping_fields.bits.luma_flag                  = v->range_mapy_flag;
173     pic_param->range_mapping_fields.bits.luma                       = v->range_mapy;
174     pic_param->range_mapping_fields.bits.chroma_flag                = v->range_mapuv_flag;
175     pic_param->range_mapping_fields.bits.chroma                     = v->range_mapuv;
176     pic_param->b_picture_fraction                                   = v->bfraction_lut_index;
177     pic_param->cbp_table                                            = v->cbpcy_vlc ? v->cbpcy_vlc - ff_vc1_cbpcy_p_vlc : 0;
178     pic_param->mb_mode_table                                        = 0; /* XXX: interlaced frame */
179     pic_param->range_reduction_frame                                = v->rangeredfrm;
180     pic_param->rounding_control                                     = v->rnd;
181     pic_param->post_processing                                      = v->postproc;
182     pic_param->picture_resolution_index                             = v->respic;
183     pic_param->luma_scale                                           = v->lumscale;
184     pic_param->luma_shift                                           = v->lumshift;
185     pic_param->picture_fields.value                                 = 0; /* reset all bits */
186     pic_param->picture_fields.bits.picture_type                     = vc1_get_PTYPE(v);
187     pic_param->picture_fields.bits.frame_coding_mode                = v->fcm;
188     pic_param->picture_fields.bits.top_field_first                  = v->tff;
189     pic_param->picture_fields.bits.is_first_field                   = v->fcm == 0; /* XXX: interlaced frame */
190     pic_param->picture_fields.bits.intensity_compensation           = v->mv_mode == MV_PMODE_INTENSITY_COMP;
191     pic_param->raw_coding.value                                     = 0; /* reset all bits */
192     pic_param->raw_coding.flags.mv_type_mb                          = v->mv_type_is_raw;
193     pic_param->raw_coding.flags.direct_mb                           = v->dmb_is_raw;
194     pic_param->raw_coding.flags.skip_mb                             = v->skip_is_raw;
195     pic_param->raw_coding.flags.field_tx                            = 0; /* XXX: interlaced frame */
196     pic_param->raw_coding.flags.forward_mb                          = 0; /* XXX: interlaced frame */
197     pic_param->raw_coding.flags.ac_pred                             = v->acpred_is_raw;
198     pic_param->raw_coding.flags.overflags                           = v->overflg_is_raw;
199     pic_param->bitplane_present.value                               = 0; /* reset all bits */
200     pic_param->bitplane_present.flags.bp_mv_type_mb                 = vc1_has_MVTYPEMB_bitplane(v);
201     pic_param->bitplane_present.flags.bp_direct_mb                  = vc1_has_DIRECTMB_bitplane(v);
202     pic_param->bitplane_present.flags.bp_skip_mb                    = vc1_has_SKIPMB_bitplane(v);
203     pic_param->bitplane_present.flags.bp_field_tx                   = 0; /* XXX: interlaced frame */
204     pic_param->bitplane_present.flags.bp_forward_mb                 = 0; /* XXX: interlaced frame */
205     pic_param->bitplane_present.flags.bp_ac_pred                    = vc1_has_ACPRED_bitplane(v);
206     pic_param->bitplane_present.flags.bp_overflags                  = vc1_has_OVERFLAGS_bitplane(v);
207     pic_param->reference_fields.value                               = 0; /* reset all bits */
208     pic_param->reference_fields.bits.reference_distance_flag        = v->refdist_flag;
209     pic_param->reference_fields.bits.reference_distance             = 0; /* XXX: interlaced frame */
210     pic_param->reference_fields.bits.num_reference_pictures         = 0; /* XXX: interlaced frame */
211     pic_param->reference_fields.bits.reference_field_pic_indicator  = 0; /* XXX: interlaced frame */
212     pic_param->mv_fields.value                                      = 0; /* reset all bits */
213     pic_param->mv_fields.bits.mv_mode                               = vc1_get_MVMODE(v);
214     pic_param->mv_fields.bits.mv_mode2                              = vc1_get_MVMODE2(v);
215     pic_param->mv_fields.bits.mv_table                              = s->mv_table_index;
216     pic_param->mv_fields.bits.two_mv_block_pattern_table            = 0; /* XXX: interlaced frame */
217     pic_param->mv_fields.bits.four_mv_switch                        = 0; /* XXX: interlaced frame */
218     pic_param->mv_fields.bits.four_mv_block_pattern_table           = 0; /* XXX: interlaced frame */
219     pic_param->mv_fields.bits.extended_mv_flag                      = v->extended_mv;
220     pic_param->mv_fields.bits.extended_mv_range                     = v->mvrange;
221     pic_param->mv_fields.bits.extended_dmv_flag                     = v->extended_dmv;
222     pic_param->mv_fields.bits.extended_dmv_range                    = 0; /* XXX: interlaced frame */
223     pic_param->pic_quantizer_fields.value                           = 0; /* reset all bits */
224     pic_param->pic_quantizer_fields.bits.dquant                     = v->dquant;
225     pic_param->pic_quantizer_fields.bits.quantizer                  = v->quantizer_mode;
226     pic_param->pic_quantizer_fields.bits.half_qp                    = v->halfpq;
227     pic_param->pic_quantizer_fields.bits.pic_quantizer_scale        = v->pq;
228     pic_param->pic_quantizer_fields.bits.pic_quantizer_type         = v->pquantizer;
229     pic_param->pic_quantizer_fields.bits.dq_frame                   = v->dquantfrm;
230     pic_param->pic_quantizer_fields.bits.dq_profile                 = v->dqprofile;
231     pic_param->pic_quantizer_fields.bits.dq_sb_edge                 = v->dqprofile == DQPROFILE_SINGLE_EDGE  ? v->dqsbedge : 0;
232     pic_param->pic_quantizer_fields.bits.dq_db_edge                 = v->dqprofile == DQPROFILE_DOUBLE_EDGES ? v->dqsbedge : 0;
233     pic_param->pic_quantizer_fields.bits.dq_binary_level            = v->dqbilevel;
234     pic_param->pic_quantizer_fields.bits.alt_pic_quantizer          = v->altpq;
235     pic_param->transform_fields.value                               = 0; /* reset all bits */
236     pic_param->transform_fields.bits.variable_sized_transform_flag  = v->vstransform;
237     pic_param->transform_fields.bits.mb_level_transform_type_flag   = v->ttmbf;
238     pic_param->transform_fields.bits.frame_level_transform_type     = v->ttfrm;
239     pic_param->transform_fields.bits.transform_ac_codingset_idx1    = v->c_ac_table_index;
240     pic_param->transform_fields.bits.transform_ac_codingset_idx2    = v->y_ac_table_index;
241     pic_param->transform_fields.bits.intra_transform_dc_table       = v->s.dc_table_index;
242
243     switch (s->pict_type) {
244     case FF_B_TYPE:
245         pic_param->backward_reference_picture = ff_vaapi_get_surface_id(&s->next_picture);
246         // fall-through
247     case FF_P_TYPE:
248         pic_param->forward_reference_picture = ff_vaapi_get_surface_id(&s->last_picture);
249         break;
250     }
251
252     if (pic_param->bitplane_present.value) {
253         uint8_t *bitplane;
254         const uint8_t *ff_bp[3];
255         int x, y, n;
256
257         switch (s->pict_type) {
258         case FF_P_TYPE:
259             ff_bp[0] = pic_param->bitplane_present.flags.bp_direct_mb  ? v->direct_mb_plane    : NULL;
260             ff_bp[1] = pic_param->bitplane_present.flags.bp_skip_mb    ? s->mbskip_table       : NULL;
261             ff_bp[2] = pic_param->bitplane_present.flags.bp_mv_type_mb ? v->mv_type_mb_plane   : NULL;
262             break;
263         case FF_B_TYPE:
264             if (!v->bi_type) {
265                 ff_bp[0] = pic_param->bitplane_present.flags.bp_direct_mb ? v->direct_mb_plane : NULL;
266                 ff_bp[1] = pic_param->bitplane_present.flags.bp_skip_mb   ? s->mbskip_table    : NULL;
267                 ff_bp[2] = NULL; /* XXX: interlaced frame (FORWARD plane) */
268                 break;
269             }
270             /* fall-through (BI-type) */
271         case FF_I_TYPE:
272             ff_bp[0] = NULL; /* XXX: interlaced frame (FIELDTX plane) */
273             ff_bp[1] = pic_param->bitplane_present.flags.bp_ac_pred    ? v->acpred_plane       : NULL;
274             ff_bp[2] = pic_param->bitplane_present.flags.bp_overflags  ? v->over_flags_plane   : NULL;
275             break;
276         default:
277             ff_bp[0] = NULL;
278             ff_bp[1] = NULL;
279             ff_bp[2] = NULL;
280             break;
281         }
282
283         bitplane = ff_vaapi_alloc_bitplane(vactx, s->mb_height * ((s->mb_width + 1) / 2));
284         if (!bitplane)
285             return -1;
286
287         n = 0;
288         for (y = 0; y < s->mb_height; y++) {
289             for (x = 0; x < s->mb_width; x += 2) {
290                 bitplane[n] = vc1_pack_bitplanes(ff_bp, x+1, y, s->mb_stride);
291                 bitplane[n] |= (vc1_pack_bitplanes(ff_bp, x, y, s->mb_stride) << 4);
292                 ++n;
293             }
294         }
295     }
296     return 0;
297 }
298
299 static int vaapi_vc1_end_frame(AVCodecContext *avctx)
300 {
301     VC1Context * const v = avctx->priv_data;
302
303     return ff_vaapi_common_end_frame(&v->s);
304 }
305
306 static int vaapi_vc1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
307 {
308     VC1Context * const v = avctx->priv_data;
309     MpegEncContext * const s = &v->s;
310     VASliceParameterBufferVC1 *slice_param;
311
312     dprintf(avctx, "vaapi_vc1_decode_slice(): buffer %p, size %d\n", buffer, size);
313
314     /* Current bit buffer is beyond any marker for VC-1, so skip it */
315     if (avctx->codec_id == CODEC_ID_VC1 && IS_MARKER(AV_RB32(buffer))) {
316         buffer += 4;
317         size -= 4;
318     }
319
320     /* Fill in VASliceParameterBufferVC1 */
321     slice_param = (VASliceParameterBufferVC1 *)ff_vaapi_alloc_slice(avctx->hwaccel_context, buffer, size);
322     if (!slice_param)
323         return -1;
324     slice_param->macroblock_offset       = get_bits_count(&s->gb);
325     slice_param->slice_vertical_position = s->mb_y;
326     return 0;
327 }
328
329 #if CONFIG_WMV3_VAAPI_HWACCEL
330 AVHWAccel wmv3_vaapi_hwaccel = {
331     .name           = "wmv3_vaapi",
332     .type           = CODEC_TYPE_VIDEO,
333     .id             = CODEC_ID_WMV3,
334     .pix_fmt        = PIX_FMT_VAAPI_VLD,
335     .capabilities   = 0,
336     .start_frame    = vaapi_vc1_start_frame,
337     .end_frame      = vaapi_vc1_end_frame,
338     .decode_slice   = vaapi_vc1_decode_slice,
339     .priv_data_size = 0,
340 };
341 #endif
342
343 AVHWAccel vc1_vaapi_hwaccel = {
344     .name           = "vc1_vaapi",
345     .type           = CODEC_TYPE_VIDEO,
346     .id             = CODEC_ID_VC1,
347     .pix_fmt        = PIX_FMT_VAAPI_VLD,
348     .capabilities   = 0,
349     .start_frame    = vaapi_vc1_start_frame,
350     .end_frame      = vaapi_vc1_end_frame,
351     .decode_slice   = vaapi_vc1_decode_slice,
352     .priv_data_size = 0,
353 };