Merge remote-tracking branch 'qatar/master'
[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 "vc1.h"
25 #include "vc1data.h"
26
27 struct dxva2_picture_context {
28     DXVA_PictureParameters pp;
29     DXVA_SliceInfo         si;
30
31     const uint8_t          *bitstream;
32     unsigned               bitstream_size;
33 };
34
35 static void fill_picture_parameters(AVCodecContext *avctx,
36                                     struct dxva_context *ctx, const VC1Context *v,
37                                     DXVA_PictureParameters *pp)
38 {
39     const MpegEncContext *s = &v->s;
40     const Picture *current_picture = s->current_picture_ptr;
41
42     memset(pp, 0, sizeof(*pp));
43     pp->wDecodedPictureIndex    =
44     pp->wDeblockedPictureIndex  = ff_dxva2_get_surface_index(ctx, current_picture);
45     if (s->pict_type != AV_PICTURE_TYPE_I)
46         pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(ctx, &s->last_picture);
47     else
48         pp->wForwardRefPictureIndex = 0xffff;
49     if (s->pict_type == AV_PICTURE_TYPE_B)
50         pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(ctx, &s->next_picture);
51     else
52         pp->wBackwardRefPictureIndex = 0xffff;
53     if (v->profile == PROFILE_ADVANCED) {
54         /* It is the cropped width/height -1 of the frame */
55         pp->wPicWidthInMBminus1 = avctx->width  - 1;
56         pp->wPicHeightInMBminus1= avctx->height - 1;
57     } else {
58         /* It is the coded width/height in macroblock -1 of the frame */
59         pp->wPicWidthInMBminus1 = s->mb_width  - 1;
60         pp->wPicHeightInMBminus1= s->mb_height - 1;
61     }
62     pp->bMacroblockWidthMinus1  = 15;
63     pp->bMacroblockHeightMinus1 = 15;
64     pp->bBlockWidthMinus1       = 7;
65     pp->bBlockHeightMinus1      = 7;
66     pp->bBPPminus1              = 7;
67     if (s->picture_structure & PICT_TOP_FIELD)
68         pp->bPicStructure      |= 0x01;
69     if (s->picture_structure & PICT_BOTTOM_FIELD)
70         pp->bPicStructure      |= 0x02;
71     pp->bSecondField            = v->interlace && v->fcm != ILACE_FIELD && !s->first_field;
72     pp->bPicIntra               = s->pict_type == AV_PICTURE_TYPE_I;
73     pp->bPicBackwardPrediction  = s->pict_type == AV_PICTURE_TYPE_B;
74     pp->bBidirectionalAveragingMode = (1                                           << 7) |
75                                       ((ctx->cfg->ConfigIntraResidUnsigned != 0)   << 6) |
76                                       ((ctx->cfg->ConfigResidDiffAccelerator != 0) << 5) |
77                                       ((v->lumscale != 32 || v->lumshift != 0)     << 4) |
78                                       ((v->profile == PROFILE_ADVANCED)            << 3);
79     pp->bMVprecisionAndChromaRelation = ((v->mv_mode == MV_PMODE_1MV_HPEL_BILIN) << 3) |
80                                         (1                                       << 2) |
81                                         (0                                       << 1) |
82                                         (!s->quarter_sample                          );
83     pp->bChromaFormat           = v->chromaformat;
84     ctx->report_id++;
85     if (ctx->report_id >= (1 << 16))
86         ctx->report_id = 1;
87     pp->bPicScanFixed           = ctx->report_id >> 8;
88     pp->bPicScanMethod          = ctx->report_id & 0xff;
89     pp->bPicReadbackRequests    = 0;
90     pp->bRcontrol               = v->rnd;
91     pp->bPicSpatialResid8       = (v->panscanflag  << 7) |
92                                   (v->refdist_flag << 6) |
93                                   (s->loop_filter  << 5) |
94                                   (v->fastuvmc     << 4) |
95                                   (v->extended_mv  << 3) |
96                                   (v->dquant       << 1) |
97                                   (v->vstransform      );
98     pp->bPicOverflowBlocks      = (v->quantizer_mode << 6) |
99                                   (v->multires       << 5) |
100                                   (s->resync_marker  << 4) |
101                                   (v->rangered       << 3) |
102                                   (s->max_b_frames       );
103     pp->bPicExtrapolation       = (!v->interlace || v->fcm == PROGRESSIVE) ? 1 : 2;
104     pp->bPicDeblocked           = ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) |
105                                   (s->loop_filter                                     << 1);
106     pp->bPicDeblockConfined     = (v->postprocflag             << 7) |
107                                   (v->broadcast                << 6) |
108                                   (v->interlace                << 5) |
109                                   (v->tfcntrflag               << 4) |
110                                   (v->finterpflag              << 3) |
111                                   ((s->pict_type != AV_PICTURE_TYPE_B) << 2) |
112                                   (v->psf                      << 1) |
113                                   (v->extended_dmv                 );
114     if (s->pict_type != AV_PICTURE_TYPE_I)
115         pp->bPic4MVallowed      = v->mv_mode == MV_PMODE_MIXED_MV ||
116                                   (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
117                                    v->mv_mode2 == MV_PMODE_MIXED_MV);
118     if (v->profile == PROFILE_ADVANCED)
119         pp->bPicOBMC            = (v->range_mapy_flag  << 7) |
120                                   (v->range_mapy       << 4) |
121                                   (v->range_mapuv_flag << 3) |
122                                   (v->range_mapuv          );
123     pp->bPicBinPB               = 0;
124     pp->bMV_RPS                 = 0;
125     pp->bReservedBits           = 0;
126     if (s->picture_structure == PICT_FRAME) {
127         pp->wBitstreamFcodes        = v->lumscale;
128         pp->wBitstreamPCEelements   = v->lumshift;
129     } else {
130         /* Syntax: (top_field_param << 8) | bottom_field_param */
131         pp->wBitstreamFcodes        = (v->lumscale << 8) | v->lumscale;
132         pp->wBitstreamPCEelements   = (v->lumshift << 8) | v->lumshift;
133     }
134     pp->bBitstreamConcealmentNeed   = 0;
135     pp->bBitstreamConcealmentMethod = 0;
136 }
137
138 static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice,
139                        unsigned position, unsigned size)
140 {
141     const VC1Context *v = avctx->priv_data;
142     const MpegEncContext *s = &v->s;
143
144     memset(slice, 0, sizeof(*slice));
145     slice->wHorizontalPosition = 0;
146     slice->wVerticalPosition   = s->mb_y;
147     slice->dwSliceBitsInBuffer = 8 * size;
148     slice->dwSliceDataLocation = position;
149     slice->bStartCodeBitOffset = 0;
150     slice->bReservedBits       = 0;
151     slice->wMBbitOffset        = get_bits_count(&s->gb);
152     slice->wNumberMBsInSlice   = s->mb_width * s->mb_height; /* XXX We assume 1 slice */
153     slice->wQuantizerScaleCode = v->pq;
154     slice->wBadSliceChopping   = 0;
155 }
156
157 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
158                                              DXVA2_DecodeBufferDesc *bs,
159                                              DXVA2_DecodeBufferDesc *sc)
160 {
161     const VC1Context *v = avctx->priv_data;
162     struct dxva_context *ctx = avctx->hwaccel_context;
163     const MpegEncContext *s = &v->s;
164     struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->f.hwaccel_picture_private;
165
166     DXVA_SliceInfo *slice = &ctx_pic->si;
167
168     static const uint8_t start_code[] = { 0, 0, 1, 0x0d };
169     const unsigned start_code_size = avctx->codec_id == CODEC_ID_VC1 ? sizeof(start_code) : 0;
170     const unsigned slice_size = slice->dwSliceBitsInBuffer / 8;
171     const unsigned padding = 128 - ((start_code_size + slice_size) & 127);
172     const unsigned data_size = start_code_size + slice_size + padding;
173
174     uint8_t  *dxva_data;
175     unsigned dxva_size;
176     int result;
177
178     if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder,
179                                               DXVA2_BitStreamDateBufferType,
180                                               &dxva_data, &dxva_size)))
181         return -1;
182
183     result = data_size <= dxva_size ? 0 : -1;
184     if (!result) {
185         if (start_code_size > 0)
186             memcpy(dxva_data, start_code, start_code_size);
187         memcpy(dxva_data + start_code_size,
188                ctx_pic->bitstream + slice->dwSliceDataLocation, slice_size);
189         if (padding > 0)
190             memset(dxva_data + start_code_size + slice_size, 0, padding);
191         slice->dwSliceBitsInBuffer = 8 * data_size;
192     }
193     if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder,
194                                                   DXVA2_BitStreamDateBufferType)))
195         return -1;
196     if (result)
197         return result;
198
199     memset(bs, 0, sizeof(*bs));
200     bs->CompressedBufferType = DXVA2_BitStreamDateBufferType;
201     bs->DataSize             = data_size;
202     bs->NumMBsInBuffer       = s->mb_width * s->mb_height;
203     assert((bs->DataSize & 127) == 0);
204
205     return ff_dxva2_commit_buffer(avctx, ctx, sc,
206                                   DXVA2_SliceControlBufferType,
207                                   slice, sizeof(*slice), bs->NumMBsInBuffer);
208 }
209
210 static int start_frame(AVCodecContext *avctx,
211                        av_unused const uint8_t *buffer,
212                        av_unused uint32_t size)
213 {
214     const VC1Context *v = avctx->priv_data;
215     struct dxva_context *ctx = avctx->hwaccel_context;
216     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->f.hwaccel_picture_private;
217
218     if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0)
219         return -1;
220     assert(ctx_pic);
221
222     fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp);
223
224     ctx_pic->bitstream_size = 0;
225     ctx_pic->bitstream      = NULL;
226     return 0;
227 }
228
229 static int decode_slice(AVCodecContext *avctx,
230                         const uint8_t *buffer, uint32_t size)
231 {
232     const VC1Context *v = avctx->priv_data;
233     const Picture *current_picture = v->s.current_picture_ptr;
234     struct dxva2_picture_context *ctx_pic = current_picture->f.hwaccel_picture_private;
235
236     if (ctx_pic->bitstream_size > 0)
237         return -1;
238
239     if (avctx->codec_id == CODEC_ID_VC1 &&
240         size >= 4 && IS_MARKER(AV_RB32(buffer))) {
241         buffer += 4;
242         size   -= 4;
243     }
244
245     ctx_pic->bitstream_size = size;
246     ctx_pic->bitstream      = buffer;
247
248     fill_slice(avctx, &ctx_pic->si, 0, size);
249     return 0;
250 }
251
252 static int end_frame(AVCodecContext *avctx)
253 {
254     VC1Context *v = avctx->priv_data;
255     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->f.hwaccel_picture_private;
256
257     if (ctx_pic->bitstream_size <= 0)
258         return -1;
259
260     return ff_dxva2_common_end_frame(avctx, &v->s,
261                                      &ctx_pic->pp, sizeof(ctx_pic->pp),
262                                      NULL, 0,
263                                      commit_bitstream_and_slice_buffer);
264 }
265
266 #if CONFIG_WMV3_DXVA2_HWACCEL
267 AVHWAccel ff_wmv3_dxva2_hwaccel = {
268     .name           = "wmv3_dxva2",
269     .type           = AVMEDIA_TYPE_VIDEO,
270     .id             = CODEC_ID_WMV3,
271     .pix_fmt        = PIX_FMT_DXVA2_VLD,
272     .start_frame    = start_frame,
273     .decode_slice   = decode_slice,
274     .end_frame      = end_frame,
275     .priv_data_size = sizeof(struct dxva2_picture_context),
276 };
277 #endif
278
279 AVHWAccel ff_vc1_dxva2_hwaccel = {
280     .name           = "vc1_dxva2",
281     .type           = AVMEDIA_TYPE_VIDEO,
282     .id             = CODEC_ID_VC1,
283     .pix_fmt        = PIX_FMT_DXVA2_VLD,
284     .start_frame    = start_frame,
285     .decode_slice   = decode_slice,
286     .end_frame      = end_frame,
287     .priv_data_size = sizeof(struct dxva2_picture_context),
288 };