Merge commit 'e0c16e4e3259cf50b5bac4c23bb6e517f397c74b'
[ffmpeg.git] / libavcodec / dxva2_vc1.c
1 /*
2  * DXVA2 WMV3/VC-1 HW acceleration.
3  *
4  * copyright (c) 2010 Laurent Aimar
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 "dxva2_internal.h"
24 #include "mpegutils.h"
25 #include "vc1.h"
26 #include "vc1data.h"
27
28 struct dxva2_picture_context {
29     DXVA_PictureParameters pp;
30     DXVA_SliceInfo         si;
31
32     const uint8_t          *bitstream;
33     unsigned               bitstream_size;
34 };
35
36 static void fill_picture_parameters(AVCodecContext *avctx,
37                                     struct dxva_context *ctx, const VC1Context *v,
38                                     DXVA_PictureParameters *pp)
39 {
40     const MpegEncContext *s = &v->s;
41     const Picture *current_picture = s->current_picture_ptr;
42     int intcomp = 0;
43
44     // determine if intensity compensation is needed
45     if (s->pict_type == AV_PICTURE_TYPE_P) {
46       if ((v->fcm == ILACE_FRAME && v->intcomp) || (v->fcm != ILACE_FRAME && v->mv_mode == MV_PMODE_INTENSITY_COMP)) {
47         if (v->lumscale != 32 || v->lumshift != 0 || (s->picture_structure != PICT_FRAME && (v->lumscale2 != 32 || v->lumshift2 != 0)))
48           intcomp = 1;
49       }
50     }
51
52     memset(pp, 0, sizeof(*pp));
53     pp->wDecodedPictureIndex    =
54     pp->wDeblockedPictureIndex  = ff_dxva2_get_surface_index(ctx, current_picture);
55     if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type)
56         pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(ctx, &s->last_picture);
57     else
58         pp->wForwardRefPictureIndex = 0xffff;
59     if (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type)
60         pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(ctx, &s->next_picture);
61     else
62         pp->wBackwardRefPictureIndex = 0xffff;
63     if (v->profile == PROFILE_ADVANCED) {
64         /* It is the cropped width/height -1 of the frame */
65         pp->wPicWidthInMBminus1 = avctx->width  - 1;
66         pp->wPicHeightInMBminus1= avctx->height - 1;
67     } else {
68         /* It is the coded width/height in macroblock -1 of the frame */
69         pp->wPicWidthInMBminus1 = s->mb_width  - 1;
70         pp->wPicHeightInMBminus1= s->mb_height - 1;
71     }
72     pp->bMacroblockWidthMinus1  = 15;
73     pp->bMacroblockHeightMinus1 = 15;
74     pp->bBlockWidthMinus1       = 7;
75     pp->bBlockHeightMinus1      = 7;
76     pp->bBPPminus1              = 7;
77     if (s->picture_structure & PICT_TOP_FIELD)
78         pp->bPicStructure      |= 0x01;
79     if (s->picture_structure & PICT_BOTTOM_FIELD)
80         pp->bPicStructure      |= 0x02;
81     pp->bSecondField            = v->interlace && v->fcm == ILACE_FIELD && v->second_field;
82     pp->bPicIntra               = s->pict_type == AV_PICTURE_TYPE_I || v->bi_type;
83     pp->bPicBackwardPrediction  = s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
84     pp->bBidirectionalAveragingMode = (1                                           << 7) |
85                                       ((ctx->cfg->ConfigIntraResidUnsigned != 0)   << 6) |
86                                       ((ctx->cfg->ConfigResidDiffAccelerator != 0) << 5) |
87                                       (intcomp                                     << 4) |
88                                       ((v->profile == PROFILE_ADVANCED)            << 3);
89     pp->bMVprecisionAndChromaRelation = ((v->mv_mode == MV_PMODE_1MV_HPEL_BILIN) << 3) |
90                                         (1                                       << 2) |
91                                         (0                                       << 1) |
92                                         (!s->quarter_sample                          );
93     pp->bChromaFormat           = v->chromaformat;
94     ctx->report_id++;
95     if (ctx->report_id >= (1 << 16))
96         ctx->report_id = 1;
97     pp->bPicScanFixed           = ctx->report_id >> 8;
98     pp->bPicScanMethod          = ctx->report_id & 0xff;
99     pp->bPicReadbackRequests    = 0;
100     pp->bRcontrol               = v->rnd;
101     pp->bPicSpatialResid8       = (v->panscanflag  << 7) |
102                                   (v->refdist_flag << 6) |
103                                   (s->loop_filter  << 5) |
104                                   (v->fastuvmc     << 4) |
105                                   (v->extended_mv  << 3) |
106                                   (v->dquant       << 1) |
107                                   (v->vstransform      );
108     pp->bPicOverflowBlocks      = (v->quantizer_mode << 6) |
109                                   (v->multires       << 5) |
110                                   (v->resync_marker  << 4) |
111                                   (v->rangered       << 3) |
112                                   (s->max_b_frames       );
113     pp->bPicExtrapolation       = (!v->interlace || v->fcm == PROGRESSIVE) ? 1 : 2;
114     pp->bPicDeblocked           = ((!pp->bPicBackwardPrediction && v->overlap)        << 6) |
115                                   ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) |
116                                   (s->loop_filter                                     << 1);
117     pp->bPicDeblockConfined     = (v->postprocflag             << 7) |
118                                   (v->broadcast                << 6) |
119                                   (v->interlace                << 5) |
120                                   (v->tfcntrflag               << 4) |
121                                   (v->finterpflag              << 3) |
122                                   ((s->pict_type != AV_PICTURE_TYPE_B) << 2) |
123                                   (v->psf                      << 1) |
124                                   (v->extended_dmv                 );
125     if (s->pict_type != AV_PICTURE_TYPE_I)
126         pp->bPic4MVallowed      = v->mv_mode == MV_PMODE_MIXED_MV ||
127                                   (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
128                                    v->mv_mode2 == MV_PMODE_MIXED_MV);
129     if (v->profile == PROFILE_ADVANCED)
130         pp->bPicOBMC            = (v->range_mapy_flag  << 7) |
131                                   (v->range_mapy       << 4) |
132                                   (v->range_mapuv_flag << 3) |
133                                   (v->range_mapuv          );
134     pp->bPicBinPB               = 0;
135     pp->bMV_RPS                 = (v->fcm == ILACE_FIELD && pp->bPicBackwardPrediction) ? v->refdist + 9 : 0;
136     pp->bReservedBits           = v->pq;
137     if (s->picture_structure == PICT_FRAME) {
138         if (intcomp) {
139             pp->wBitstreamFcodes      = v->lumscale;
140             pp->wBitstreamPCEelements = v->lumshift;
141         } else {
142             pp->wBitstreamFcodes      = 32;
143             pp->wBitstreamPCEelements = 0;
144         }
145     } else {
146         /* Syntax: (top_field_param << 8) | bottom_field_param */
147         if (intcomp) {
148             pp->wBitstreamFcodes      = (v->lumscale << 8) | v->lumscale2;
149             pp->wBitstreamPCEelements = (v->lumshift << 8) | v->lumshift2;
150         } else {
151             pp->wBitstreamFcodes      = (32 << 8) | 32;
152             pp->wBitstreamPCEelements = 0;
153         }
154     }
155     pp->bBitstreamConcealmentNeed   = 0;
156     pp->bBitstreamConcealmentMethod = 0;
157 }
158
159 static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice,
160                        unsigned position, unsigned size)
161 {
162     const VC1Context *v = avctx->priv_data;
163     const MpegEncContext *s = &v->s;
164
165     memset(slice, 0, sizeof(*slice));
166     slice->wHorizontalPosition = 0;
167     slice->wVerticalPosition   = s->mb_y;
168     slice->dwSliceBitsInBuffer = 8 * size;
169     slice->dwSliceDataLocation = position;
170     slice->bStartCodeBitOffset = 0;
171     slice->bReservedBits       = (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type) ? v->bfraction_lut_index + 9 : 0;
172     slice->wMBbitOffset        = v->p_frame_skipped ? 0xffff : get_bits_count(&s->gb) + (avctx->codec_id == AV_CODEC_ID_VC1 ? 32 : 0);
173     slice->wNumberMBsInSlice   = s->mb_width * s->mb_height; /* XXX We assume 1 slice */
174     slice->wQuantizerScaleCode = v->pq;
175     slice->wBadSliceChopping   = 0;
176 }
177
178 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
179                                              DXVA2_DecodeBufferDesc *bs,
180                                              DXVA2_DecodeBufferDesc *sc)
181 {
182     const VC1Context *v = avctx->priv_data;
183     struct dxva_context *ctx = avctx->hwaccel_context;
184     const MpegEncContext *s = &v->s;
185     struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->hwaccel_picture_private;
186
187     DXVA_SliceInfo *slice = &ctx_pic->si;
188
189     static const uint8_t start_code[] = { 0, 0, 1, 0x0d };
190     const unsigned start_code_size = avctx->codec_id == AV_CODEC_ID_VC1 ? sizeof(start_code) : 0;
191     const unsigned slice_size = slice->dwSliceBitsInBuffer / 8;
192     const unsigned padding = 128 - ((start_code_size + slice_size) & 127);
193     const unsigned data_size = start_code_size + slice_size + padding;
194
195     uint8_t  *dxva_data;
196     unsigned dxva_size;
197     int result;
198
199     if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder,
200                                               DXVA2_BitStreamDateBufferType,
201                                               (void **)&dxva_data, &dxva_size)))
202         return -1;
203
204     result = data_size <= dxva_size ? 0 : -1;
205     if (!result) {
206         if (start_code_size > 0) {
207             memcpy(dxva_data, start_code, start_code_size);
208             if (v->second_field)
209                 dxva_data[3] = 0x0c;
210         }
211         memcpy(dxva_data + start_code_size,
212                ctx_pic->bitstream + slice->dwSliceDataLocation, slice_size);
213         if (padding > 0)
214             memset(dxva_data + start_code_size + slice_size, 0, padding);
215         slice->dwSliceBitsInBuffer = 8 * data_size;
216     }
217     if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder,
218                                                   DXVA2_BitStreamDateBufferType)))
219         return -1;
220     if (result)
221         return result;
222
223     memset(bs, 0, sizeof(*bs));
224     bs->CompressedBufferType = DXVA2_BitStreamDateBufferType;
225     bs->DataSize             = data_size;
226     bs->NumMBsInBuffer       = s->mb_width * s->mb_height;
227     assert((bs->DataSize & 127) == 0);
228
229     return ff_dxva2_commit_buffer(avctx, ctx, sc,
230                                   DXVA2_SliceControlBufferType,
231                                   slice, sizeof(*slice), bs->NumMBsInBuffer);
232 }
233
234 static int dxva2_vc1_start_frame(AVCodecContext *avctx,
235                                  av_unused const uint8_t *buffer,
236                                  av_unused uint32_t size)
237 {
238     const VC1Context *v = avctx->priv_data;
239     struct dxva_context *ctx = avctx->hwaccel_context;
240     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
241
242     if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0)
243         return -1;
244     assert(ctx_pic);
245
246     fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp);
247
248     ctx_pic->bitstream_size = 0;
249     ctx_pic->bitstream      = NULL;
250     return 0;
251 }
252
253 static int dxva2_vc1_decode_slice(AVCodecContext *avctx,
254                                   const uint8_t *buffer,
255                                   uint32_t size)
256 {
257     const VC1Context *v = avctx->priv_data;
258     const Picture *current_picture = v->s.current_picture_ptr;
259     struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
260
261     if (ctx_pic->bitstream_size > 0)
262         return -1;
263
264     if (avctx->codec_id == AV_CODEC_ID_VC1 &&
265         size >= 4 && IS_MARKER(AV_RB32(buffer))) {
266         buffer += 4;
267         size   -= 4;
268     }
269
270     ctx_pic->bitstream_size = size;
271     ctx_pic->bitstream      = buffer;
272
273     fill_slice(avctx, &ctx_pic->si, 0, size);
274     return 0;
275 }
276
277 static int dxva2_vc1_end_frame(AVCodecContext *avctx)
278 {
279     VC1Context *v = avctx->priv_data;
280     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
281     int ret;
282
283     if (ctx_pic->bitstream_size <= 0)
284         return -1;
285
286     ret = ff_dxva2_common_end_frame(avctx, v->s.current_picture_ptr,
287                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
288                                     NULL, 0,
289                                     commit_bitstream_and_slice_buffer);
290     if (!ret)
291         ff_mpeg_draw_horiz_band(&v->s, 0, avctx->height);
292     return ret;
293 }
294
295 #if CONFIG_WMV3_DXVA2_HWACCEL
296 AVHWAccel ff_wmv3_dxva2_hwaccel = {
297     .name           = "wmv3_dxva2",
298     .type           = AVMEDIA_TYPE_VIDEO,
299     .id             = AV_CODEC_ID_WMV3,
300     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
301     .start_frame    = dxva2_vc1_start_frame,
302     .decode_slice   = dxva2_vc1_decode_slice,
303     .end_frame      = dxva2_vc1_end_frame,
304     .priv_data_size = sizeof(struct dxva2_picture_context),
305 };
306 #endif
307
308 AVHWAccel ff_vc1_dxva2_hwaccel = {
309     .name           = "vc1_dxva2",
310     .type           = AVMEDIA_TYPE_VIDEO,
311     .id             = AV_CODEC_ID_VC1,
312     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
313     .start_frame    = dxva2_vc1_start_frame,
314     .decode_slice   = dxva2_vc1_decode_slice,
315     .end_frame      = dxva2_vc1_end_frame,
316     .priv_data_size = sizeof(struct dxva2_picture_context),
317 };