D3D11va: add a Direct3D11 video decoder similar to DXVA2
[libav.git] / libavcodec / dxva2_mpeg2.c
CommitLineData
164aab02
LA
1/*
2 * MPEG-2 HW acceleration.
3 *
4 * copyright (c) 2010 Laurent Aimar
5 *
2912e87a 6 * This file is part of Libav.
164aab02 7 *
2912e87a 8 * Libav is free software; you can redistribute it and/or
164aab02
LA
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 *
2912e87a 13 * Libav is distributed in the hope that it will be useful,
164aab02
LA
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
2912e87a 19 * License along with Libav; if not, write to the Free Software
164aab02
LA
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
b6638210 23#include "libavutil/log.h"
164aab02 24#include "dxva2_internal.h"
e0c16e4e 25#include "mpegutils.h"
164aab02 26
b6638210 27#define MAX_SLICES 1024
164aab02
LA
28struct dxva2_picture_context {
29 DXVA_PictureParameters pp;
30 DXVA_QmatrixData qm;
31 unsigned slice_count;
32 DXVA_SliceInfo slice[MAX_SLICES];
33
34 const uint8_t *bitstream;
35 unsigned bitstream_size;
36};
37
38static void fill_picture_parameters(AVCodecContext *avctx,
d8039ef8 39 AVDXVAContext *ctx,
164aab02
LA
40 const struct MpegEncContext *s,
41 DXVA_PictureParameters *pp)
42{
43 const Picture *current_picture = s->current_picture_ptr;
44 int is_field = s->picture_structure != PICT_FRAME;
45
46 memset(pp, 0, sizeof(*pp));
d8039ef8 47 pp->wDecodedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f);
164aab02 48 pp->wDeblockedPictureIndex = 0;
975a1447 49 if (s->pict_type != AV_PICTURE_TYPE_I)
d8039ef8 50 pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f);
164aab02
LA
51 else
52 pp->wForwardRefPictureIndex = 0xffff;
975a1447 53 if (s->pict_type == AV_PICTURE_TYPE_B)
d8039ef8 54 pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f);
164aab02
LA
55 else
56 pp->wBackwardRefPictureIndex = 0xffff;
57 pp->wPicWidthInMBminus1 = s->mb_width - 1;
58 pp->wPicHeightInMBminus1 = (s->mb_height >> is_field) - 1;
59 pp->bMacroblockWidthMinus1 = 15;
60 pp->bMacroblockHeightMinus1 = 15;
61 pp->bBlockWidthMinus1 = 7;
62 pp->bBlockHeightMinus1 = 7;
63 pp->bBPPminus1 = 7;
64 pp->bPicStructure = s->picture_structure;
65 pp->bSecondField = is_field && !s->first_field;
975a1447
SS
66 pp->bPicIntra = s->pict_type == AV_PICTURE_TYPE_I;
67 pp->bPicBackwardPrediction = s->pict_type == AV_PICTURE_TYPE_B;
164aab02
LA
68 pp->bBidirectionalAveragingMode = 0;
69 pp->bMVprecisionAndChromaRelation= 0; /* FIXME */
70 pp->bChromaFormat = s->chroma_format;
71 pp->bPicScanFixed = 1;
72 pp->bPicScanMethod = s->alternate_scan ? 1 : 0;
73 pp->bPicReadbackRequests = 0;
74 pp->bRcontrol = 0;
75 pp->bPicSpatialResid8 = 0;
76 pp->bPicOverflowBlocks = 0;
77 pp->bPicExtrapolation = 0;
78 pp->bPicDeblocked = 0;
79 pp->bPicDeblockConfined = 0;
80 pp->bPic4MVallowed = 0;
81 pp->bPicOBMC = 0;
82 pp->bPicBinPB = 0;
83 pp->bMV_RPS = 0;
84 pp->bReservedBits = 0;
85 pp->wBitstreamFcodes = (s->mpeg_f_code[0][0] << 12) |
86 (s->mpeg_f_code[0][1] << 8) |
87 (s->mpeg_f_code[1][0] << 4) |
88 (s->mpeg_f_code[1][1] );
89 pp->wBitstreamPCEelements = (s->intra_dc_precision << 14) |
90 (s->picture_structure << 12) |
91 (s->top_field_first << 11) |
92 (s->frame_pred_frame_dct << 10) |
93 (s->concealment_motion_vectors << 9) |
94 (s->q_scale_type << 8) |
95 (s->intra_vlc_format << 7) |
96 (s->alternate_scan << 6) |
97 (s->repeat_first_field << 5) |
98 (s->chroma_420_type << 4) |
99 (s->progressive_frame << 3);
100 pp->bBitstreamConcealmentNeed = 0;
101 pp->bBitstreamConcealmentMethod = 0;
102}
103
104static void fill_quantization_matrices(AVCodecContext *avctx,
d8039ef8 105 AVDXVAContext *ctx,
164aab02
LA
106 const struct MpegEncContext *s,
107 DXVA_QmatrixData *qm)
108{
109 int i;
110 for (i = 0; i < 4; i++)
111 qm->bNewQmatrix[i] = 1;
112 for (i = 0; i < 64; i++) {
e3fcb143 113 int n = s->idsp.idct_permutation[ff_zigzag_direct[i]];
164aab02
LA
114 qm->Qmatrix[0][i] = s->intra_matrix[n];;
115 qm->Qmatrix[1][i] = s->inter_matrix[n];;
116 qm->Qmatrix[2][i] = s->chroma_intra_matrix[n];;
117 qm->Qmatrix[3][i] = s->chroma_inter_matrix[n];;
118 }
119}
120
121static void fill_slice(AVCodecContext *avctx,
122 const struct MpegEncContext *s,
123 DXVA_SliceInfo *slice,
124 unsigned position,
125 const uint8_t *buffer, unsigned size)
126{
127 int is_field = s->picture_structure != PICT_FRAME;
128 GetBitContext gb;
129
130 memset(slice, 0, sizeof(*slice));
131 slice->wHorizontalPosition = s->mb_x;
132 slice->wVerticalPosition = s->mb_y >> is_field;
133 slice->dwSliceBitsInBuffer = 8 * size;
134 slice->dwSliceDataLocation = position;
135 slice->bStartCodeBitOffset = 0;
136 slice->bReservedBits = 0;
137 /* XXX We store the index of the first MB and it will be fixed later */
138 slice->wNumberMBsInSlice = (s->mb_y >> is_field) * s->mb_width + s->mb_x;
139 slice->wBadSliceChopping = 0;
140
141 init_get_bits(&gb, &buffer[4], 8 * (size - 4));
142
143 slice->wQuantizerScaleCode = get_bits(&gb, 5);
144 while (get_bits1(&gb))
145 skip_bits(&gb, 8);
146
147 slice->wMBbitOffset = 4 * 8 + get_bits_count(&gb);
148}
149static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
d8039ef8
SL
150 DECODER_BUFFER_DESC *bs,
151 DECODER_BUFFER_DESC *sc)
164aab02
LA
152{
153 const struct MpegEncContext *s = avctx->priv_data;
d8039ef8 154 AVDXVAContext *ctx = avctx->hwaccel_context;
164aab02 155 struct dxva2_picture_context *ctx_pic =
ad0c9f2d 156 s->current_picture_ptr->hwaccel_picture_private;
164aab02
LA
157 const int is_field = s->picture_structure != PICT_FRAME;
158 const unsigned mb_count = s->mb_width * (s->mb_height >> is_field);
4600a85e 159 void *dxva_data_ptr;
164aab02
LA
160 uint8_t *dxva_data, *current, *end;
161 unsigned dxva_size;
162 unsigned i;
d8039ef8 163 unsigned type;
164aab02 164
d8039ef8
SL
165#if CONFIG_D3D11VA
166 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
167 type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
168 if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
169 D3D11VA_CONTEXT(ctx)->decoder,
170 type,
171 &dxva_size, &dxva_data_ptr)))
172 return -1;
173 }
174#endif
175#if CONFIG_DXVA2
176 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
177 type = DXVA2_BitStreamDateBufferType;
178 if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
179 type,
180 &dxva_data_ptr, &dxva_size)))
181 return -1;
182 }
183#endif
4600a85e
DB
184
185 dxva_data = dxva_data_ptr;
164aab02
LA
186 current = dxva_data;
187 end = dxva_data + dxva_size;
188
189 for (i = 0; i < ctx_pic->slice_count; i++) {
190 DXVA_SliceInfo *slice = &ctx_pic->slice[i];
191 unsigned position = slice->dwSliceDataLocation;
192 unsigned size = slice->dwSliceBitsInBuffer / 8;
193 if (size > end - current) {
194 av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
195 break;
196 }
197 slice->dwSliceDataLocation = current - dxva_data;
198
199 if (i < ctx_pic->slice_count - 1)
200 slice->wNumberMBsInSlice =
201 slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice;
202 else
203 slice->wNumberMBsInSlice =
204 mb_count - slice[0].wNumberMBsInSlice;
205
206 memcpy(current, &ctx_pic->bitstream[position], size);
207 current += size;
208 }
d8039ef8
SL
209#if CONFIG_D3D11VA
210 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD)
211 if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
212 return -1;
213#endif
214#if CONFIG_DXVA2
215 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
216 if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
217 return -1;
218#endif
164aab02
LA
219 if (i < ctx_pic->slice_count)
220 return -1;
221
d8039ef8
SL
222#if CONFIG_D3D11VA
223 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
224 D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
225 memset(dsc11, 0, sizeof(*dsc11));
226 dsc11->BufferType = type;
227 dsc11->DataSize = current - dxva_data;
228 dsc11->NumMBsInBuffer = mb_count;
229
230 type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
231 }
232#endif
233#if CONFIG_DXVA2
234 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
235 DXVA2_DecodeBufferDesc *dsc2 = bs;
236 memset(dsc2, 0, sizeof(*dsc2));
237 dsc2->CompressedBufferType = type;
238 dsc2->DataSize = current - dxva_data;
239 dsc2->NumMBsInBuffer = mb_count;
240
241 type = DXVA2_SliceControlBufferType;
242 }
243#endif
164aab02
LA
244
245 return ff_dxva2_commit_buffer(avctx, ctx, sc,
d8039ef8 246 type,
164aab02
LA
247 ctx_pic->slice,
248 ctx_pic->slice_count * sizeof(*ctx_pic->slice),
249 mb_count);
250}
251
c57a5936
DB
252static int dxva2_mpeg2_start_frame(AVCodecContext *avctx,
253 av_unused const uint8_t *buffer,
254 av_unused uint32_t size)
164aab02
LA
255{
256 const struct MpegEncContext *s = avctx->priv_data;
d8039ef8 257 AVDXVAContext *ctx = avctx->hwaccel_context;
164aab02 258 struct dxva2_picture_context *ctx_pic =
ad0c9f2d 259 s->current_picture_ptr->hwaccel_picture_private;
164aab02 260
d8039ef8
SL
261 if (DXVA_CONTEXT_DECODER(avctx, ctx) == NULL ||
262 DXVA_CONTEXT_CFG(avctx, ctx) == NULL ||
263 DXVA_CONTEXT_COUNT(avctx, ctx) <= 0)
164aab02
LA
264 return -1;
265 assert(ctx_pic);
266
267 fill_picture_parameters(avctx, ctx, s, &ctx_pic->pp);
268 fill_quantization_matrices(avctx, ctx, s, &ctx_pic->qm);
269
270 ctx_pic->slice_count = 0;
271 ctx_pic->bitstream_size = 0;
272 ctx_pic->bitstream = NULL;
273 return 0;
274}
275
c57a5936
DB
276static int dxva2_mpeg2_decode_slice(AVCodecContext *avctx,
277 const uint8_t *buffer, uint32_t size)
164aab02
LA
278{
279 const struct MpegEncContext *s = avctx->priv_data;
280 struct dxva2_picture_context *ctx_pic =
ad0c9f2d 281 s->current_picture_ptr->hwaccel_picture_private;
164aab02
LA
282 unsigned position;
283
b6638210
RH
284 if (ctx_pic->slice_count >= MAX_SLICES) {
285 avpriv_request_sample(avctx, "%d slices in dxva2",
286 ctx_pic->slice_count);
164aab02 287 return -1;
b6638210 288 }
164aab02
LA
289 if (!ctx_pic->bitstream)
290 ctx_pic->bitstream = buffer;
291 ctx_pic->bitstream_size += size;
292
293 position = buffer - ctx_pic->bitstream;
294 fill_slice(avctx, s, &ctx_pic->slice[ctx_pic->slice_count++], position,
295 buffer, size);
296 return 0;
297}
298
c57a5936 299static int dxva2_mpeg2_end_frame(AVCodecContext *avctx)
164aab02
LA
300{
301 struct MpegEncContext *s = avctx->priv_data;
302 struct dxva2_picture_context *ctx_pic =
ad0c9f2d 303 s->current_picture_ptr->hwaccel_picture_private;
2c541554 304 int ret;
164aab02
LA
305
306 if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
307 return -1;
f6774f90 308 ret = ff_dxva2_common_end_frame(avctx, s->current_picture_ptr->f,
2c541554
AK
309 &ctx_pic->pp, sizeof(ctx_pic->pp),
310 &ctx_pic->qm, sizeof(ctx_pic->qm),
311 commit_bitstream_and_slice_buffer);
312 if (!ret)
313 ff_mpeg_draw_horiz_band(s, 0, avctx->height);
314 return ret;
164aab02
LA
315}
316
d8039ef8 317#if CONFIG_MPEG2_DXVA2_HWACCEL
d1b6f33b 318AVHWAccel ff_mpeg2_dxva2_hwaccel = {
164aab02
LA
319 .name = "mpeg2_dxva2",
320 .type = AVMEDIA_TYPE_VIDEO,
36ef5369 321 .id = AV_CODEC_ID_MPEG2VIDEO,
716d413c 322 .pix_fmt = AV_PIX_FMT_DXVA2_VLD,
c57a5936
DB
323 .start_frame = dxva2_mpeg2_start_frame,
324 .decode_slice = dxva2_mpeg2_decode_slice,
325 .end_frame = dxva2_mpeg2_end_frame,
a871ef0c 326 .frame_priv_data_size = sizeof(struct dxva2_picture_context),
164aab02 327};
d8039ef8
SL
328#endif
329
330#if CONFIG_MPEG2_D3D11VA_HWACCEL
331AVHWAccel ff_mpeg2_d3d11va_hwaccel = {
332 .name = "mpeg2_d3d11va",
333 .type = AVMEDIA_TYPE_VIDEO,
334 .id = AV_CODEC_ID_MPEG2VIDEO,
335 .pix_fmt = AV_PIX_FMT_D3D11VA_VLD,
336 .start_frame = dxva2_mpeg2_start_frame,
337 .decode_slice = dxva2_mpeg2_decode_slice,
338 .end_frame = dxva2_mpeg2_end_frame,
339 .frame_priv_data_size = sizeof(struct dxva2_picture_context),
340};
341#endif