blob: cc97d7b6c646c06a58959dd539159e19c702b8e9
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 | #define MAX_SLICES 1024 |
33 | struct dxva2_picture_context { |
34 | DXVA_PictureParameters pp; |
35 | unsigned slice_count; |
36 | DXVA_SliceInfo slice[MAX_SLICES]; |
37 | |
38 | const uint8_t *bitstream; |
39 | unsigned bitstream_size; |
40 | }; |
41 | |
42 | static void fill_picture_parameters(AVCodecContext *avctx, |
43 | AVDXVAContext *ctx, const VC1Context *v, |
44 | DXVA_PictureParameters *pp) |
45 | { |
46 | const MpegEncContext *s = &v->s; |
47 | const Picture *current_picture = s->current_picture_ptr; |
48 | int intcomp = 0; |
49 | |
50 | // determine if intensity compensation is needed |
51 | if (s->pict_type == AV_PICTURE_TYPE_P) { |
52 | if ((v->fcm == ILACE_FRAME && v->intcomp) || (v->fcm != ILACE_FRAME && v->mv_mode == MV_PMODE_INTENSITY_COMP)) { |
53 | if (v->lumscale != 32 || v->lumshift != 0 || (s->picture_structure != PICT_FRAME && (v->lumscale2 != 32 || v->lumshift2 != 0))) |
54 | intcomp = 1; |
55 | } |
56 | } |
57 | |
58 | memset(pp, 0, sizeof(*pp)); |
59 | pp->wDecodedPictureIndex = |
60 | pp->wDeblockedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f); |
61 | if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type) |
62 | pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f); |
63 | else |
64 | pp->wForwardRefPictureIndex = 0xffff; |
65 | if (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type) |
66 | pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f); |
67 | else |
68 | pp->wBackwardRefPictureIndex = 0xffff; |
69 | if (v->profile == PROFILE_ADVANCED) { |
70 | /* It is the cropped width/height -1 of the frame */ |
71 | pp->wPicWidthInMBminus1 = avctx->width - 1; |
72 | pp->wPicHeightInMBminus1= avctx->height - 1; |
73 | } else { |
74 | /* It is the coded width/height in macroblock -1 of the frame */ |
75 | pp->wPicWidthInMBminus1 = s->mb_width - 1; |
76 | pp->wPicHeightInMBminus1= s->mb_height - 1; |
77 | } |
78 | pp->bMacroblockWidthMinus1 = 15; |
79 | pp->bMacroblockHeightMinus1 = 15; |
80 | pp->bBlockWidthMinus1 = 7; |
81 | pp->bBlockHeightMinus1 = 7; |
82 | pp->bBPPminus1 = 7; |
83 | if (s->picture_structure & PICT_TOP_FIELD) |
84 | pp->bPicStructure |= 0x01; |
85 | if (s->picture_structure & PICT_BOTTOM_FIELD) |
86 | pp->bPicStructure |= 0x02; |
87 | pp->bSecondField = v->interlace && v->fcm == ILACE_FIELD && v->second_field; |
88 | pp->bPicIntra = s->pict_type == AV_PICTURE_TYPE_I || v->bi_type; |
89 | pp->bPicBackwardPrediction = s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type; |
90 | pp->bBidirectionalAveragingMode = (1 << 7) | |
91 | ((DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) != 0) << 6) | |
92 | ((DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) != 0) << 5) | |
93 | (intcomp << 4) | |
94 | ((v->profile == PROFILE_ADVANCED) << 3); |
95 | pp->bMVprecisionAndChromaRelation = ((v->mv_mode == MV_PMODE_1MV_HPEL_BILIN) << 3) | |
96 | (1 << 2) | |
97 | (0 << 1) | |
98 | (!s->quarter_sample ); |
99 | pp->bChromaFormat = v->chromaformat; |
100 | DXVA_CONTEXT_REPORT_ID(avctx, ctx)++; |
101 | if (DXVA_CONTEXT_REPORT_ID(avctx, ctx) >= (1 << 16)) |
102 | DXVA_CONTEXT_REPORT_ID(avctx, ctx) = 1; |
103 | pp->bPicScanFixed = DXVA_CONTEXT_REPORT_ID(avctx, ctx) >> 8; |
104 | pp->bPicScanMethod = DXVA_CONTEXT_REPORT_ID(avctx, ctx) & 0xff; |
105 | pp->bPicReadbackRequests = 0; |
106 | pp->bRcontrol = v->rnd; |
107 | pp->bPicSpatialResid8 = (v->panscanflag << 7) | |
108 | (v->refdist_flag << 6) | |
109 | (s->loop_filter << 5) | |
110 | (v->fastuvmc << 4) | |
111 | (v->extended_mv << 3) | |
112 | (v->dquant << 1) | |
113 | (v->vstransform ); |
114 | pp->bPicOverflowBlocks = (v->quantizer_mode << 6) | |
115 | (v->multires << 5) | |
116 | (v->resync_marker << 4) | |
117 | (v->rangered << 3) | |
118 | (s->max_b_frames ); |
119 | pp->bPicExtrapolation = (!v->interlace || v->fcm == PROGRESSIVE) ? 1 : 2; |
120 | pp->bPicDeblocked = ((!pp->bPicBackwardPrediction && v->overlap) << 6) | |
121 | ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) | |
122 | (s->loop_filter << 1); |
123 | pp->bPicDeblockConfined = (v->postprocflag << 7) | |
124 | (v->broadcast << 6) | |
125 | (v->interlace << 5) | |
126 | (v->tfcntrflag << 4) | |
127 | (v->finterpflag << 3) | |
128 | ((s->pict_type != AV_PICTURE_TYPE_B) << 2) | |
129 | (v->psf << 1) | |
130 | (v->extended_dmv ); |
131 | if (s->pict_type != AV_PICTURE_TYPE_I) |
132 | pp->bPic4MVallowed = v->mv_mode == MV_PMODE_MIXED_MV || |
133 | (v->mv_mode == MV_PMODE_INTENSITY_COMP && |
134 | v->mv_mode2 == MV_PMODE_MIXED_MV); |
135 | if (v->profile == PROFILE_ADVANCED) |
136 | pp->bPicOBMC = (v->range_mapy_flag << 7) | |
137 | (v->range_mapy << 4) | |
138 | (v->range_mapuv_flag << 3) | |
139 | (v->range_mapuv ); |
140 | pp->bPicBinPB = 0; |
141 | pp->bMV_RPS = (v->fcm == ILACE_FIELD && pp->bPicBackwardPrediction) ? v->refdist + 9 : 0; |
142 | pp->bReservedBits = v->pq; |
143 | if (s->picture_structure == PICT_FRAME) { |
144 | if (intcomp) { |
145 | pp->wBitstreamFcodes = v->lumscale; |
146 | pp->wBitstreamPCEelements = v->lumshift; |
147 | } else { |
148 | pp->wBitstreamFcodes = 32; |
149 | pp->wBitstreamPCEelements = 0; |
150 | } |
151 | } else { |
152 | /* Syntax: (top_field_param << 8) | bottom_field_param */ |
153 | if (intcomp) { |
154 | pp->wBitstreamFcodes = (v->lumscale << 8) | v->lumscale2; |
155 | pp->wBitstreamPCEelements = (v->lumshift << 8) | v->lumshift2; |
156 | } else { |
157 | pp->wBitstreamFcodes = (32 << 8) | 32; |
158 | pp->wBitstreamPCEelements = 0; |
159 | } |
160 | } |
161 | pp->bBitstreamConcealmentNeed = 0; |
162 | pp->bBitstreamConcealmentMethod = 0; |
163 | } |
164 | |
165 | static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, |
166 | unsigned position, unsigned size) |
167 | { |
168 | const VC1Context *v = avctx->priv_data; |
169 | const MpegEncContext *s = &v->s; |
170 | |
171 | memset(slice, 0, sizeof(*slice)); |
172 | slice->wHorizontalPosition = 0; |
173 | slice->wVerticalPosition = s->mb_y; |
174 | slice->dwSliceBitsInBuffer = 8 * size; |
175 | slice->dwSliceDataLocation = position; |
176 | slice->bStartCodeBitOffset = 0; |
177 | slice->bReservedBits = (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type) ? v->bfraction_lut_index + 9 : 0; |
178 | slice->wMBbitOffset = v->p_frame_skipped ? 0xffff : get_bits_count(&s->gb) + (avctx->codec_id == AV_CODEC_ID_VC1 ? 32 : 0); |
179 | /* XXX We store the index of the first MB and it will be fixed later */ |
180 | slice->wNumberMBsInSlice = (s->mb_y >> v->field_mode) * s->mb_width + s->mb_x; |
181 | slice->wQuantizerScaleCode = v->pq; |
182 | slice->wBadSliceChopping = 0; |
183 | } |
184 | |
185 | static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx, |
186 | DECODER_BUFFER_DESC *bs, |
187 | DECODER_BUFFER_DESC *sc) |
188 | { |
189 | const VC1Context *v = avctx->priv_data; |
190 | AVDXVAContext *ctx = avctx->hwaccel_context; |
191 | const MpegEncContext *s = &v->s; |
192 | struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; |
193 | |
194 | static const uint8_t start_code[] = { 0, 0, 1, 0x0d }; |
195 | const unsigned start_code_size = avctx->codec_id == AV_CODEC_ID_VC1 ? sizeof(start_code) : 0; |
196 | const unsigned mb_count = s->mb_width * (s->mb_height >> v->field_mode); |
197 | DXVA_SliceInfo *slice = NULL; |
198 | void *dxva_data_ptr; |
199 | uint8_t *dxva_data, *current, *end; |
200 | unsigned dxva_size; |
201 | unsigned padding; |
202 | unsigned i; |
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 | current = dxva_data; |
227 | end = dxva_data + dxva_size; |
228 | |
229 | for (i = 0; i < ctx_pic->slice_count; i++) { |
230 | unsigned position, size; |
231 | slice = &ctx_pic->slice[i]; |
232 | position = slice->dwSliceDataLocation; |
233 | size = slice->dwSliceBitsInBuffer / 8; |
234 | if (start_code_size + size > end - current) { |
235 | av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream"); |
236 | break; |
237 | } |
238 | slice->dwSliceDataLocation = current - dxva_data; |
239 | |
240 | if (i < ctx_pic->slice_count - 1) |
241 | slice->wNumberMBsInSlice = |
242 | slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice; |
243 | else |
244 | slice->wNumberMBsInSlice = |
245 | mb_count - slice[0].wNumberMBsInSlice; |
246 | |
247 | /* write the appropriate frame, field or slice start code */ |
248 | if (start_code_size) { |
249 | memcpy(current, start_code, start_code_size); |
250 | if (i == 0 && v->second_field) |
251 | current[3] = 0x0c; |
252 | else if (i > 0) |
253 | current[3] = 0x0b; |
254 | |
255 | current += start_code_size; |
256 | slice->dwSliceBitsInBuffer += start_code_size * 8; |
257 | } |
258 | |
259 | memcpy(current, &ctx_pic->bitstream[position], size); |
260 | current += size; |
261 | } |
262 | padding = FFMIN(128 - ((current - dxva_data) & 127), end - current); |
263 | if (slice && padding > 0) { |
264 | memset(current, 0, padding); |
265 | current += padding; |
266 | slice->dwSliceBitsInBuffer += padding * 8; |
267 | } |
268 | |
269 | #if CONFIG_D3D11VA |
270 | if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) |
271 | if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type))) |
272 | return -1; |
273 | #endif |
274 | #if CONFIG_DXVA2 |
275 | if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) |
276 | if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type))) |
277 | return -1; |
278 | #endif |
279 | if (i < ctx_pic->slice_count) |
280 | return -1; |
281 | |
282 | #if CONFIG_D3D11VA |
283 | if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) { |
284 | D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs; |
285 | memset(dsc11, 0, sizeof(*dsc11)); |
286 | dsc11->BufferType = type; |
287 | dsc11->DataSize = current - dxva_data; |
288 | dsc11->NumMBsInBuffer = mb_count; |
289 | |
290 | type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL; |
291 | } |
292 | #endif |
293 | #if CONFIG_DXVA2 |
294 | if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) { |
295 | DXVA2_DecodeBufferDesc *dsc2 = bs; |
296 | memset(dsc2, 0, sizeof(*dsc2)); |
297 | dsc2->CompressedBufferType = type; |
298 | dsc2->DataSize = current - dxva_data; |
299 | dsc2->NumMBsInBuffer = mb_count; |
300 | |
301 | type = DXVA2_SliceControlBufferType; |
302 | } |
303 | #endif |
304 | |
305 | return ff_dxva2_commit_buffer(avctx, ctx, sc, |
306 | type, |
307 | ctx_pic->slice, |
308 | ctx_pic->slice_count * sizeof(*ctx_pic->slice), |
309 | mb_count); |
310 | } |
311 | |
312 | static int dxva2_vc1_start_frame(AVCodecContext *avctx, |
313 | av_unused const uint8_t *buffer, |
314 | av_unused uint32_t size) |
315 | { |
316 | const VC1Context *v = avctx->priv_data; |
317 | AVDXVAContext *ctx = avctx->hwaccel_context; |
318 | struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private; |
319 | |
320 | if (!DXVA_CONTEXT_VALID(avctx, ctx)) |
321 | return -1; |
322 | assert(ctx_pic); |
323 | |
324 | fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp); |
325 | |
326 | ctx_pic->slice_count = 0; |
327 | ctx_pic->bitstream_size = 0; |
328 | ctx_pic->bitstream = NULL; |
329 | return 0; |
330 | } |
331 | |
332 | static int dxva2_vc1_decode_slice(AVCodecContext *avctx, |
333 | const uint8_t *buffer, |
334 | uint32_t size) |
335 | { |
336 | const VC1Context *v = avctx->priv_data; |
337 | const Picture *current_picture = v->s.current_picture_ptr; |
338 | struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private; |
339 | unsigned position; |
340 | |
341 | if (ctx_pic->slice_count >= MAX_SLICES) { |
342 | avpriv_request_sample(avctx, "%d slices in dxva2", |
343 | ctx_pic->slice_count); |
344 | return -1; |
345 | } |
346 | |
347 | if (avctx->codec_id == AV_CODEC_ID_VC1 && |
348 | size >= 4 && IS_MARKER(AV_RB32(buffer))) { |
349 | buffer += 4; |
350 | size -= 4; |
351 | } |
352 | |
353 | if (!ctx_pic->bitstream) |
354 | ctx_pic->bitstream = buffer; |
355 | ctx_pic->bitstream_size += size; |
356 | |
357 | position = buffer - ctx_pic->bitstream; |
358 | fill_slice(avctx, &ctx_pic->slice[ctx_pic->slice_count++], position, size); |
359 | return 0; |
360 | } |
361 | |
362 | static int dxva2_vc1_end_frame(AVCodecContext *avctx) |
363 | { |
364 | VC1Context *v = avctx->priv_data; |
365 | struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private; |
366 | int ret; |
367 | |
368 | if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) |
369 | return -1; |
370 | |
371 | ret = ff_dxva2_common_end_frame(avctx, v->s.current_picture_ptr->f, |
372 | &ctx_pic->pp, sizeof(ctx_pic->pp), |
373 | NULL, 0, |
374 | commit_bitstream_and_slice_buffer); |
375 | if (!ret) |
376 | ff_mpeg_draw_horiz_band(&v->s, 0, avctx->height); |
377 | return ret; |
378 | } |
379 | |
380 | #if CONFIG_WMV3_DXVA2_HWACCEL |
381 | AVHWAccel ff_wmv3_dxva2_hwaccel = { |
382 | .name = "wmv3_dxva2", |
383 | .type = AVMEDIA_TYPE_VIDEO, |
384 | .id = AV_CODEC_ID_WMV3, |
385 | .pix_fmt = AV_PIX_FMT_DXVA2_VLD, |
386 | .start_frame = dxva2_vc1_start_frame, |
387 | .decode_slice = dxva2_vc1_decode_slice, |
388 | .end_frame = dxva2_vc1_end_frame, |
389 | .frame_priv_data_size = sizeof(struct dxva2_picture_context), |
390 | }; |
391 | #endif |
392 | |
393 | #if CONFIG_VC1_DXVA2_HWACCEL |
394 | AVHWAccel ff_vc1_dxva2_hwaccel = { |
395 | .name = "vc1_dxva2", |
396 | .type = AVMEDIA_TYPE_VIDEO, |
397 | .id = AV_CODEC_ID_VC1, |
398 | .pix_fmt = AV_PIX_FMT_DXVA2_VLD, |
399 | .start_frame = dxva2_vc1_start_frame, |
400 | .decode_slice = dxva2_vc1_decode_slice, |
401 | .end_frame = dxva2_vc1_end_frame, |
402 | .frame_priv_data_size = sizeof(struct dxva2_picture_context), |
403 | }; |
404 | #endif |
405 | |
406 | #if CONFIG_WMV3_D3D11VA_HWACCEL |
407 | AVHWAccel ff_wmv3_d3d11va_hwaccel = { |
408 | .name = "wmv3_d3d11va", |
409 | .type = AVMEDIA_TYPE_VIDEO, |
410 | .id = AV_CODEC_ID_WMV3, |
411 | .pix_fmt = AV_PIX_FMT_D3D11VA_VLD, |
412 | .start_frame = dxva2_vc1_start_frame, |
413 | .decode_slice = dxva2_vc1_decode_slice, |
414 | .end_frame = dxva2_vc1_end_frame, |
415 | .frame_priv_data_size = sizeof(struct dxva2_picture_context), |
416 | }; |
417 | #endif |
418 | |
419 | #if CONFIG_VC1_D3D11VA_HWACCEL |
420 | AVHWAccel ff_vc1_d3d11va_hwaccel = { |
421 | .name = "vc1_d3d11va", |
422 | .type = AVMEDIA_TYPE_VIDEO, |
423 | .id = AV_CODEC_ID_VC1, |
424 | .pix_fmt = AV_PIX_FMT_D3D11VA_VLD, |
425 | .start_frame = dxva2_vc1_start_frame, |
426 | .decode_slice = dxva2_vc1_decode_slice, |
427 | .end_frame = dxva2_vc1_end_frame, |
428 | .frame_priv_data_size = sizeof(struct dxva2_picture_context), |
429 | }; |
430 | #endif |
431 |