Merge commit '9d218d573f8088c606d873e80df572582e6773ef'
[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 "mpegutils.h"
24 #include "vc1.h"
25 #include "vc1data.h"
26
27 // The headers above may include w32threads.h, which uses the original
28 // _WIN32_WINNT define, while dxva2_internal.h redefines it to target a
29 // potentially newer version.
30 #include "dxva2_internal.h"
31
32 struct dxva2_picture_context {
33     DXVA_PictureParameters pp;
34     DXVA_SliceInfo         si;
35
36     const uint8_t          *bitstream;
37     unsigned               bitstream_size;
38 };
39
40 static void fill_picture_parameters(AVCodecContext *avctx,
41                                     AVDXVAContext *ctx, const VC1Context *v,
42                                     DXVA_PictureParameters *pp)
43 {
44     const MpegEncContext *s = &v->s;
45     const Picture *current_picture = s->current_picture_ptr;
46     int intcomp = 0;
47
48     // determine if intensity compensation is needed
49     if (s->pict_type == AV_PICTURE_TYPE_P) {
50       if ((v->fcm == ILACE_FRAME && v->intcomp) || (v->fcm != ILACE_FRAME && v->mv_mode == MV_PMODE_INTENSITY_COMP)) {
51         if (v->lumscale != 32 || v->lumshift != 0 || (s->picture_structure != PICT_FRAME && (v->lumscale2 != 32 || v->lumshift2 != 0)))
52           intcomp = 1;
53       }
54     }
55
56     memset(pp, 0, sizeof(*pp));
57     pp->wDecodedPictureIndex    =
58     pp->wDeblockedPictureIndex  = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f);
59     if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type)
60         pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f);
61     else
62         pp->wForwardRefPictureIndex = 0xffff;
63     if (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type)
64         pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f);
65     else
66         pp->wBackwardRefPictureIndex = 0xffff;
67     if (v->profile == PROFILE_ADVANCED) {
68         /* It is the cropped width/height -1 of the frame */
69         pp->wPicWidthInMBminus1 = avctx->width  - 1;
70         pp->wPicHeightInMBminus1= avctx->height - 1;
71     } else {
72         /* It is the coded width/height in macroblock -1 of the frame */
73         pp->wPicWidthInMBminus1 = s->mb_width  - 1;
74         pp->wPicHeightInMBminus1= s->mb_height - 1;
75     }
76     pp->bMacroblockWidthMinus1  = 15;
77     pp->bMacroblockHeightMinus1 = 15;
78     pp->bBlockWidthMinus1       = 7;
79     pp->bBlockHeightMinus1      = 7;
80     pp->bBPPminus1              = 7;
81     if (s->picture_structure & PICT_TOP_FIELD)
82         pp->bPicStructure      |= 0x01;
83     if (s->picture_structure & PICT_BOTTOM_FIELD)
84         pp->bPicStructure      |= 0x02;
85     pp->bSecondField            = v->interlace && v->fcm == ILACE_FIELD && v->second_field;
86     pp->bPicIntra               = s->pict_type == AV_PICTURE_TYPE_I || v->bi_type;
87     pp->bPicBackwardPrediction  = s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
88     pp->bBidirectionalAveragingMode = (1                                           << 7) |
89                                       ((DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) != 0) << 6) |
90                                       ((DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) != 0) << 5) |
91                                       (intcomp                                     << 4) |
92                                       ((v->profile == PROFILE_ADVANCED)            << 3);
93     pp->bMVprecisionAndChromaRelation = ((v->mv_mode == MV_PMODE_1MV_HPEL_BILIN) << 3) |
94                                         (1                                       << 2) |
95                                         (0                                       << 1) |
96                                         (!s->quarter_sample                          );
97     pp->bChromaFormat           = v->chromaformat;
98     DXVA_CONTEXT_REPORT_ID(avctx, ctx)++;
99     if (DXVA_CONTEXT_REPORT_ID(avctx, ctx) >= (1 << 16))
100         DXVA_CONTEXT_REPORT_ID(avctx, ctx) = 1;
101     pp->bPicScanFixed           = DXVA_CONTEXT_REPORT_ID(avctx, ctx) >> 8;
102     pp->bPicScanMethod          = DXVA_CONTEXT_REPORT_ID(avctx, ctx) & 0xff;
103     pp->bPicReadbackRequests    = 0;
104     pp->bRcontrol               = v->rnd;
105     pp->bPicSpatialResid8       = (v->panscanflag  << 7) |
106                                   (v->refdist_flag << 6) |
107                                   (s->loop_filter  << 5) |
108                                   (v->fastuvmc     << 4) |
109                                   (v->extended_mv  << 3) |
110                                   (v->dquant       << 1) |
111                                   (v->vstransform      );
112     pp->bPicOverflowBlocks      = (v->quantizer_mode << 6) |
113                                   (v->multires       << 5) |
114                                   (v->resync_marker  << 4) |
115                                   (v->rangered       << 3) |
116                                   (s->max_b_frames       );
117     pp->bPicExtrapolation       = (!v->interlace || v->fcm == PROGRESSIVE) ? 1 : 2;
118     pp->bPicDeblocked           = ((!pp->bPicBackwardPrediction && v->overlap)        << 6) |
119                                   ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) |
120                                   (s->loop_filter                                     << 1);
121     pp->bPicDeblockConfined     = (v->postprocflag             << 7) |
122                                   (v->broadcast                << 6) |
123                                   (v->interlace                << 5) |
124                                   (v->tfcntrflag               << 4) |
125                                   (v->finterpflag              << 3) |
126                                   ((s->pict_type != AV_PICTURE_TYPE_B) << 2) |
127                                   (v->psf                      << 1) |
128                                   (v->extended_dmv                 );
129     if (s->pict_type != AV_PICTURE_TYPE_I)
130         pp->bPic4MVallowed      = v->mv_mode == MV_PMODE_MIXED_MV ||
131                                   (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
132                                    v->mv_mode2 == MV_PMODE_MIXED_MV);
133     if (v->profile == PROFILE_ADVANCED)
134         pp->bPicOBMC            = (v->range_mapy_flag  << 7) |
135                                   (v->range_mapy       << 4) |
136                                   (v->range_mapuv_flag << 3) |
137                                   (v->range_mapuv          );
138     pp->bPicBinPB               = 0;
139     pp->bMV_RPS                 = (v->fcm == ILACE_FIELD && pp->bPicBackwardPrediction) ? v->refdist + 9 : 0;
140     pp->bReservedBits           = v->pq;
141     if (s->picture_structure == PICT_FRAME) {
142         if (intcomp) {
143             pp->wBitstreamFcodes      = v->lumscale;
144             pp->wBitstreamPCEelements = v->lumshift;
145         } else {
146             pp->wBitstreamFcodes      = 32;
147             pp->wBitstreamPCEelements = 0;
148         }
149     } else {
150         /* Syntax: (top_field_param << 8) | bottom_field_param */
151         if (intcomp) {
152             pp->wBitstreamFcodes      = (v->lumscale << 8) | v->lumscale2;
153             pp->wBitstreamPCEelements = (v->lumshift << 8) | v->lumshift2;
154         } else {
155             pp->wBitstreamFcodes      = (32 << 8) | 32;
156             pp->wBitstreamPCEelements = 0;
157         }
158     }
159     pp->bBitstreamConcealmentNeed   = 0;
160     pp->bBitstreamConcealmentMethod = 0;
161 }
162
163 static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice,
164                        unsigned position, unsigned size)
165 {
166     const VC1Context *v = avctx->priv_data;
167     const MpegEncContext *s = &v->s;
168
169     memset(slice, 0, sizeof(*slice));
170     slice->wHorizontalPosition = 0;
171     slice->wVerticalPosition   = s->mb_y;
172     slice->dwSliceBitsInBuffer = 8 * size;
173     slice->dwSliceDataLocation = position;
174     slice->bStartCodeBitOffset = 0;
175     slice->bReservedBits       = (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type) ? v->bfraction_lut_index + 9 : 0;
176     slice->wMBbitOffset        = v->p_frame_skipped ? 0xffff : get_bits_count(&s->gb) + (avctx->codec_id == AV_CODEC_ID_VC1 ? 32 : 0);
177     slice->wNumberMBsInSlice   = s->mb_width * s->mb_height; /* XXX We assume 1 slice */
178     slice->wQuantizerScaleCode = v->pq;
179     slice->wBadSliceChopping   = 0;
180 }
181
182 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
183                                              DECODER_BUFFER_DESC *bs,
184                                              DECODER_BUFFER_DESC *sc)
185 {
186     const VC1Context *v = avctx->priv_data;
187     AVDXVAContext *ctx = avctx->hwaccel_context;
188     const MpegEncContext *s = &v->s;
189     struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->hwaccel_picture_private;
190
191     DXVA_SliceInfo *slice = &ctx_pic->si;
192
193     static const uint8_t start_code[] = { 0, 0, 1, 0x0d };
194     const unsigned start_code_size = avctx->codec_id == AV_CODEC_ID_VC1 ? sizeof(start_code) : 0;
195     const unsigned slice_size = slice->dwSliceBitsInBuffer / 8;
196     const unsigned padding = 128 - ((start_code_size + slice_size) & 127);
197     const unsigned data_size = start_code_size + slice_size + padding;
198
199     void     *dxva_data_ptr;
200     uint8_t  *dxva_data;
201     unsigned dxva_size;
202     int result;
203     unsigned type;
204
205 #if CONFIG_D3D11VA
206     if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
207         type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
208         if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
209                                                        D3D11VA_CONTEXT(ctx)->decoder,
210                                                        type,
211                                                        &dxva_size, &dxva_data_ptr)))
212             return -1;
213     }
214 #endif
215 #if CONFIG_DXVA2
216     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
217         type = DXVA2_BitStreamDateBufferType;
218         if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
219                                                   type,
220                                                   &dxva_data_ptr, &dxva_size)))
221             return -1;
222     }
223 #endif
224
225     dxva_data = dxva_data_ptr;
226     result = data_size <= dxva_size ? 0 : -1;
227     if (!result) {
228         if (start_code_size > 0) {
229             memcpy(dxva_data, start_code, start_code_size);
230             if (v->second_field)
231                 dxva_data[3] = 0x0c;
232         }
233         memcpy(dxva_data + start_code_size,
234                ctx_pic->bitstream + slice->dwSliceDataLocation, slice_size);
235         if (padding > 0)
236             memset(dxva_data + start_code_size + slice_size, 0, padding);
237         slice->dwSliceBitsInBuffer = 8 * data_size;
238     }
239 #if CONFIG_D3D11VA
240     if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD)
241         if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
242             return -1;
243 #endif
244 #if CONFIG_DXVA2
245     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
246         if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
247             return -1;
248 #endif
249     if (result)
250         return result;
251
252 #if CONFIG_D3D11VA
253     if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
254         D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
255         memset(dsc11, 0, sizeof(*dsc11));
256         dsc11->BufferType           = type;
257         dsc11->DataSize             = data_size;
258         dsc11->NumMBsInBuffer       = s->mb_width * s->mb_height;
259
260         type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
261     }
262 #endif
263 #if CONFIG_DXVA2
264     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
265         DXVA2_DecodeBufferDesc *dsc2 = bs;
266         memset(dsc2, 0, sizeof(*dsc2));
267         dsc2->CompressedBufferType = type;
268         dsc2->DataSize             = data_size;
269         dsc2->NumMBsInBuffer       = s->mb_width * s->mb_height;
270
271         type = DXVA2_SliceControlBufferType;
272     }
273 #endif
274     assert((data_size & 127) == 0);
275
276     return ff_dxva2_commit_buffer(avctx, ctx, sc,
277                                   type,
278                                   slice, sizeof(*slice), s->mb_width * s->mb_height);
279 }
280
281 static int dxva2_vc1_start_frame(AVCodecContext *avctx,
282                                  av_unused const uint8_t *buffer,
283                                  av_unused uint32_t size)
284 {
285     const VC1Context *v = avctx->priv_data;
286     AVDXVAContext *ctx = avctx->hwaccel_context;
287     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
288
289     if (DXVA_CONTEXT_DECODER(avctx, ctx) == NULL ||
290         DXVA_CONTEXT_CFG(avctx, ctx) == NULL ||
291         DXVA_CONTEXT_COUNT(avctx, ctx) <= 0)
292         return -1;
293     assert(ctx_pic);
294
295     fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp);
296
297     ctx_pic->bitstream_size = 0;
298     ctx_pic->bitstream      = NULL;
299     return 0;
300 }
301
302 static int dxva2_vc1_decode_slice(AVCodecContext *avctx,
303                                   const uint8_t *buffer,
304                                   uint32_t size)
305 {
306     const VC1Context *v = avctx->priv_data;
307     const Picture *current_picture = v->s.current_picture_ptr;
308     struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
309
310     if (ctx_pic->bitstream_size > 0)
311         return -1;
312
313     if (avctx->codec_id == AV_CODEC_ID_VC1 &&
314         size >= 4 && IS_MARKER(AV_RB32(buffer))) {
315         buffer += 4;
316         size   -= 4;
317     }
318
319     ctx_pic->bitstream_size = size;
320     ctx_pic->bitstream      = buffer;
321
322     fill_slice(avctx, &ctx_pic->si, 0, size);
323     return 0;
324 }
325
326 static int dxva2_vc1_end_frame(AVCodecContext *avctx)
327 {
328     VC1Context *v = avctx->priv_data;
329     struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
330     int ret;
331
332     if (ctx_pic->bitstream_size <= 0)
333         return -1;
334
335     ret = ff_dxva2_common_end_frame(avctx, v->s.current_picture_ptr->f,
336                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
337                                     NULL, 0,
338                                     commit_bitstream_and_slice_buffer);
339     if (!ret)
340         ff_mpeg_draw_horiz_band(&v->s, 0, avctx->height);
341     return ret;
342 }
343
344 #if CONFIG_WMV3_DXVA2_HWACCEL
345 AVHWAccel ff_wmv3_dxva2_hwaccel = {
346     .name           = "wmv3_dxva2",
347     .type           = AVMEDIA_TYPE_VIDEO,
348     .id             = AV_CODEC_ID_WMV3,
349     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
350     .start_frame    = dxva2_vc1_start_frame,
351     .decode_slice   = dxva2_vc1_decode_slice,
352     .end_frame      = dxva2_vc1_end_frame,
353     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
354 };
355 #endif
356
357 #if CONFIG_VC1_DXVA2_HWACCEL
358 AVHWAccel ff_vc1_dxva2_hwaccel = {
359     .name           = "vc1_dxva2",
360     .type           = AVMEDIA_TYPE_VIDEO,
361     .id             = AV_CODEC_ID_VC1,
362     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
363     .start_frame    = dxva2_vc1_start_frame,
364     .decode_slice   = dxva2_vc1_decode_slice,
365     .end_frame      = dxva2_vc1_end_frame,
366     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
367 };
368 #endif
369
370 #if CONFIG_WMV3_D3D11VA_HWACCEL
371 AVHWAccel ff_wmv3_d3d11va_hwaccel = {
372     .name           = "wmv3_d3d11va",
373     .type           = AVMEDIA_TYPE_VIDEO,
374     .id             = AV_CODEC_ID_WMV3,
375     .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
376     .start_frame    = dxva2_vc1_start_frame,
377     .decode_slice   = dxva2_vc1_decode_slice,
378     .end_frame      = dxva2_vc1_end_frame,
379     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
380 };
381 #endif
382
383 #if CONFIG_VC1_D3D11VA_HWACCEL
384 AVHWAccel ff_vc1_d3d11va_hwaccel = {
385     .name           = "vc1_d3d11va",
386     .type           = AVMEDIA_TYPE_VIDEO,
387     .id             = AV_CODEC_ID_VC1,
388     .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
389     .start_frame    = dxva2_vc1_start_frame,
390     .decode_slice   = dxva2_vc1_decode_slice,
391     .end_frame      = dxva2_vc1_end_frame,
392     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
393 };
394 #endif