svq3: use hpeldsp instead of dsputil for half-pel functions.
[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 && !v->bi_type)
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 && !v->bi_type)
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 && v->second_field;
72     pp->bPicIntra               = s->pict_type == AV_PICTURE_TYPE_I || v->bi_type;
73     pp->bPicBackwardPrediction  = s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
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           = ((!pp->bPicBackwardPrediction && v->overlap)        << 6) |
105                                   ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) |
106                                   (s->loop_filter                                     << 1);
107     pp->bPicDeblockConfined     = (v->postprocflag             << 7) |
108                                   (v->broadcast                << 6) |
109                                   (v->interlace                << 5) |
110                                   (v->tfcntrflag               << 4) |
111                                   (v->finterpflag              << 3) |
112                                   ((s->pict_type != AV_PICTURE_TYPE_B) << 2) |
113                                   (v->psf                      << 1) |
114                                   (v->extended_dmv                 );
115     if (s->pict_type != AV_PICTURE_TYPE_I)
116         pp->bPic4MVallowed      = v->mv_mode == MV_PMODE_MIXED_MV ||
117                                   (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
118                                    v->mv_mode2 == MV_PMODE_MIXED_MV);
119     if (v->profile == PROFILE_ADVANCED)
120         pp->bPicOBMC            = (v->range_mapy_flag  << 7) |
121                                   (v->range_mapy       << 4) |
122                                   (v->range_mapuv_flag << 3) |
123                                   (v->range_mapuv          );
124     pp->bPicBinPB               = 0;
125     pp->bMV_RPS                 = 0;
126     pp->bReservedBits           = 0;
127     if (s->picture_structure == PICT_FRAME) {
128         pp->wBitstreamFcodes        = v->lumscale;
129         pp->wBitstreamPCEelements   = v->lumshift;
130     } else {
131         /* Syntax: (top_field_param << 8) | bottom_field_param */
132         pp->wBitstreamFcodes        = (v->lumscale << 8) | v->lumscale;
133         pp->wBitstreamPCEelements   = (v->lumshift << 8) | v->lumshift;
134     }
135     pp->bBitstreamConcealmentNeed   = 0;
136     pp->bBitstreamConcealmentMethod = 0;
137 }
138
139 static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice,
140                        unsigned position, unsigned size)
141 {
142     const VC1Context *v = avctx->priv_data;
143     const MpegEncContext *s = &v->s;
144
145     memset(slice, 0, sizeof(*slice));
146     slice->wHorizontalPosition = 0;
147     slice->wVerticalPosition   = s->mb_y;
148     slice->dwSliceBitsInBuffer = 8 * size;
149     slice->dwSliceDataLocation = position;
150     slice->bStartCodeBitOffset = 0;
151     slice->bReservedBits       = 0;
152     slice->wMBbitOffset        = get_bits_count(&s->gb);
153     slice->wNumberMBsInSlice   = s->mb_width * s->mb_height; /* XXX We assume 1 slice */
154     slice->wQuantizerScaleCode = v->pq;
155     slice->wBadSliceChopping   = 0;
156 }
157
158 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
159                                              DXVA2_DecodeBufferDesc *bs,
160                                              DXVA2_DecodeBufferDesc *sc)
161 {
162     const VC1Context *v = avctx->priv_data;
163     struct dxva_context *ctx = avctx->hwaccel_context;
164     const MpegEncContext *s = &v->s;
165     struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->hwaccel_picture_private;
166
167     DXVA_SliceInfo *slice = &ctx_pic->si;
168
169     static const uint8_t start_code[] = { 0, 0, 1, 0x0d };
170     const unsigned start_code_size = avctx->codec_id == AV_CODEC_ID_VC1 ? sizeof(start_code) : 0;
171     const unsigned slice_size = slice->dwSliceBitsInBuffer / 8;
172     const unsigned padding = 128 - ((start_code_size + slice_size) & 127);
173     const unsigned data_size = start_code_size + slice_size + padding;
174
175     uint8_t  *dxva_data;
176     unsigned dxva_size;
177     int result;
178
179     if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder,
180                                               DXVA2_BitStreamDateBufferType,
181                                               (void **)&dxva_data, &dxva_size)))
182         return -1;
183
184     result = data_size <= dxva_size ? 0 : -1;
185     if (!result) {
186         if (start_code_size > 0) {
187             memcpy(dxva_data, start_code, start_code_size);
188             if (v->second_field)
189                 dxva_data[3] = 0x0c;
190         }
191         memcpy(dxva_data + start_code_size,
192                ctx_pic->bitstream + slice->dwSliceDataLocation, slice_size);
193         if (padding > 0)
194             memset(dxva_data + start_code_size + slice_size, 0, padding);
195         slice->dwSliceBitsInBuffer = 8 * data_size;
196     }
197     if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder,
198                                                   DXVA2_BitStreamDateBufferType)))
199         return -1;
200     if (result)
201         return result;
202
203     memset(bs, 0, sizeof(*bs));
204     bs->CompressedBufferType = DXVA2_BitStreamDateBufferType;
205     bs->DataSize             = data_size;
206     bs->NumMBsInBuffer       = s->mb_width * s->mb_height;
207     assert((bs->DataSize & 127) == 0);
208
209     return ff_dxva2_commit_buffer(avctx, ctx, sc,
210                                   DXVA2_SliceControlBufferType,
211                                   slice, sizeof(*slice), bs->NumMBsInBuffer);
212 }
213
214 static int dxva2_vc1_start_frame(AVCodecContext *avctx,
215                                  av_unused const uint8_t *buffer,
216                                  av_unused uint32_t size)
217 {
218     const VC1Context *v = avctx->priv_data;
219     struct dxva_context *ctx = avctx->hwaccel_context;
220     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
221
222     if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0)
223         return -1;
224     assert(ctx_pic);
225
226     fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp);
227
228     ctx_pic->bitstream_size = 0;
229     ctx_pic->bitstream      = NULL;
230     return 0;
231 }
232
233 static int dxva2_vc1_decode_slice(AVCodecContext *avctx,
234                                   const uint8_t *buffer,
235                                   uint32_t size)
236 {
237     const VC1Context *v = avctx->priv_data;
238     const Picture *current_picture = v->s.current_picture_ptr;
239     struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
240
241     if (ctx_pic->bitstream_size > 0)
242         return -1;
243
244     if (avctx->codec_id == AV_CODEC_ID_VC1 &&
245         size >= 4 && IS_MARKER(AV_RB32(buffer))) {
246         buffer += 4;
247         size   -= 4;
248     }
249
250     ctx_pic->bitstream_size = size;
251     ctx_pic->bitstream      = buffer;
252
253     fill_slice(avctx, &ctx_pic->si, 0, size);
254     return 0;
255 }
256
257 static int dxva2_vc1_end_frame(AVCodecContext *avctx)
258 {
259     VC1Context *v = avctx->priv_data;
260     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
261     int ret;
262
263     if (ctx_pic->bitstream_size <= 0)
264         return -1;
265
266     ret = ff_dxva2_common_end_frame(avctx, v->s.current_picture_ptr,
267                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
268                                     NULL, 0,
269                                     commit_bitstream_and_slice_buffer);
270     if (!ret)
271         ff_mpeg_draw_horiz_band(&v->s, 0, avctx->height);
272     return ret;
273 }
274
275 #if CONFIG_WMV3_DXVA2_HWACCEL
276 AVHWAccel ff_wmv3_dxva2_hwaccel = {
277     .name           = "wmv3_dxva2",
278     .type           = AVMEDIA_TYPE_VIDEO,
279     .id             = AV_CODEC_ID_WMV3,
280     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
281     .start_frame    = dxva2_vc1_start_frame,
282     .decode_slice   = dxva2_vc1_decode_slice,
283     .end_frame      = dxva2_vc1_end_frame,
284     .priv_data_size = sizeof(struct dxva2_picture_context),
285 };
286 #endif
287
288 AVHWAccel ff_vc1_dxva2_hwaccel = {
289     .name           = "vc1_dxva2",
290     .type           = AVMEDIA_TYPE_VIDEO,
291     .id             = AV_CODEC_ID_VC1,
292     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
293     .start_frame    = dxva2_vc1_start_frame,
294     .decode_slice   = dxva2_vc1_decode_slice,
295     .end_frame      = dxva2_vc1_end_frame,
296     .priv_data_size = sizeof(struct dxva2_picture_context),
297 };