]> git.sesse.net Git - ffmpeg/blob - libavcodec/dxva2_mpeg2.c
x86/jpeg2000dsp: add ff_ict_float_{fma3,fma4}
[ffmpeg] / libavcodec / dxva2_mpeg2.c
1 /*
2  * MPEG-2 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 "libavutil/log.h"
24 #include "mpegutils.h"
25 #include "mpegvideo.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     DXVA_QmatrixData       qm;
36     unsigned               slice_count;
37     DXVA_SliceInfo         slice[MAX_SLICES];
38
39     const uint8_t          *bitstream;
40     unsigned               bitstream_size;
41 };
42
43 static void fill_picture_parameters(AVCodecContext *avctx,
44                                     AVDXVAContext *ctx,
45                                     const struct MpegEncContext *s,
46                                     DXVA_PictureParameters *pp)
47 {
48     const Picture *current_picture = s->current_picture_ptr;
49     int is_field = s->picture_structure != PICT_FRAME;
50
51     memset(pp, 0, sizeof(*pp));
52     pp->wDecodedPictureIndex         = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f);
53     pp->wDeblockedPictureIndex       = 0;
54     if (s->pict_type != AV_PICTURE_TYPE_I)
55         pp->wForwardRefPictureIndex  = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f);
56     else
57         pp->wForwardRefPictureIndex  = 0xffff;
58     if (s->pict_type == AV_PICTURE_TYPE_B)
59         pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f);
60     else
61         pp->wBackwardRefPictureIndex = 0xffff;
62     pp->wPicWidthInMBminus1          = s->mb_width  - 1;
63     pp->wPicHeightInMBminus1         = (s->mb_height >> is_field) - 1;
64     pp->bMacroblockWidthMinus1       = 15;
65     pp->bMacroblockHeightMinus1      = 15;
66     pp->bBlockWidthMinus1            = 7;
67     pp->bBlockHeightMinus1           = 7;
68     pp->bBPPminus1                   = 7;
69     pp->bPicStructure                = s->picture_structure;
70     pp->bSecondField                 = is_field && !s->first_field;
71     pp->bPicIntra                    = s->pict_type == AV_PICTURE_TYPE_I;
72     pp->bPicBackwardPrediction       = s->pict_type == AV_PICTURE_TYPE_B;
73     pp->bBidirectionalAveragingMode  = 0;
74     pp->bMVprecisionAndChromaRelation= 0; /* FIXME */
75     pp->bChromaFormat                = s->chroma_format;
76     pp->bPicScanFixed                = 1;
77     pp->bPicScanMethod               = s->alternate_scan ? 1 : 0;
78     pp->bPicReadbackRequests         = 0;
79     pp->bRcontrol                    = 0;
80     pp->bPicSpatialResid8            = 0;
81     pp->bPicOverflowBlocks           = 0;
82     pp->bPicExtrapolation            = 0;
83     pp->bPicDeblocked                = 0;
84     pp->bPicDeblockConfined          = 0;
85     pp->bPic4MVallowed               = 0;
86     pp->bPicOBMC                     = 0;
87     pp->bPicBinPB                    = 0;
88     pp->bMV_RPS                      = 0;
89     pp->bReservedBits                = 0;
90     pp->wBitstreamFcodes             = (s->mpeg_f_code[0][0] << 12) |
91                                        (s->mpeg_f_code[0][1] <<  8) |
92                                        (s->mpeg_f_code[1][0] <<  4) |
93                                        (s->mpeg_f_code[1][1]      );
94     pp->wBitstreamPCEelements        = (s->intra_dc_precision         << 14) |
95                                        (s->picture_structure          << 12) |
96                                        (s->top_field_first            << 11) |
97                                        (s->frame_pred_frame_dct       << 10) |
98                                        (s->concealment_motion_vectors <<  9) |
99                                        (s->q_scale_type               <<  8) |
100                                        (s->intra_vlc_format           <<  7) |
101                                        (s->alternate_scan             <<  6) |
102                                        (s->repeat_first_field         <<  5) |
103                                        (s->chroma_420_type            <<  4) |
104                                        (s->progressive_frame          <<  3);
105     pp->bBitstreamConcealmentNeed    = 0;
106     pp->bBitstreamConcealmentMethod  = 0;
107 }
108
109 static void fill_quantization_matrices(AVCodecContext *avctx,
110                                        AVDXVAContext *ctx,
111                                        const struct MpegEncContext *s,
112                                        DXVA_QmatrixData *qm)
113 {
114     int i;
115     for (i = 0; i < 4; i++)
116         qm->bNewQmatrix[i] = 1;
117     for (i = 0; i < 64; i++) {
118         int n = s->idsp.idct_permutation[ff_zigzag_direct[i]];
119         qm->Qmatrix[0][i] = s->intra_matrix[n];
120         qm->Qmatrix[1][i] = s->inter_matrix[n];
121         qm->Qmatrix[2][i] = s->chroma_intra_matrix[n];
122         qm->Qmatrix[3][i] = s->chroma_inter_matrix[n];
123     }
124 }
125
126 static void fill_slice(AVCodecContext *avctx,
127                        const struct MpegEncContext *s,
128                        DXVA_SliceInfo *slice,
129                        unsigned position,
130                        const uint8_t *buffer, unsigned size)
131 {
132     int is_field = s->picture_structure != PICT_FRAME;
133     GetBitContext gb;
134
135     memset(slice, 0, sizeof(*slice));
136     slice->wHorizontalPosition = s->mb_x;
137     slice->wVerticalPosition   = s->mb_y >> is_field;
138     slice->dwSliceBitsInBuffer = 8 * size;
139     slice->dwSliceDataLocation = position;
140     slice->bStartCodeBitOffset = 0;
141     slice->bReservedBits       = 0;
142     /* XXX We store the index of the first MB and it will be fixed later */
143     slice->wNumberMBsInSlice   = (s->mb_y >> is_field) * s->mb_width + s->mb_x;
144     slice->wBadSliceChopping   = 0;
145
146     init_get_bits(&gb, &buffer[4], 8 * (size - 4));
147
148     slice->wQuantizerScaleCode = get_bits(&gb, 5);
149     skip_1stop_8data_bits(&gb);
150
151     slice->wMBbitOffset        = 4 * 8 + get_bits_count(&gb);
152 }
153 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
154                                              DECODER_BUFFER_DESC *bs,
155                                              DECODER_BUFFER_DESC *sc)
156 {
157     const struct MpegEncContext *s = avctx->priv_data;
158     AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
159     struct dxva2_picture_context *ctx_pic =
160         s->current_picture_ptr->hwaccel_picture_private;
161     const int is_field = s->picture_structure != PICT_FRAME;
162     const unsigned mb_count = s->mb_width * (s->mb_height >> is_field);
163     void     *dxva_data_ptr;
164     uint8_t  *dxva_data, *current, *end;
165     unsigned dxva_size;
166     unsigned i;
167     unsigned type;
168
169 #if CONFIG_D3D11VA
170     if (ff_dxva2_is_d3d11(avctx)) {
171         type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
172         if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
173                                                        D3D11VA_CONTEXT(ctx)->decoder,
174                                                        type,
175                                                        &dxva_size, &dxva_data_ptr)))
176             return -1;
177     }
178 #endif
179 #if CONFIG_DXVA2
180     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
181         type = DXVA2_BitStreamDateBufferType;
182         if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
183                                                   type,
184                                                   &dxva_data_ptr, &dxva_size)))
185             return -1;
186     }
187 #endif
188
189     dxva_data = dxva_data_ptr;
190     current = dxva_data;
191     end = dxva_data + dxva_size;
192
193     for (i = 0; i < ctx_pic->slice_count; i++) {
194         DXVA_SliceInfo *slice = &ctx_pic->slice[i];
195         unsigned position = slice->dwSliceDataLocation;
196         unsigned size     = slice->dwSliceBitsInBuffer / 8;
197         if (size > end - current) {
198             av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
199             break;
200         }
201         slice->dwSliceDataLocation = current - dxva_data;
202
203         if (i < ctx_pic->slice_count - 1)
204             slice->wNumberMBsInSlice =
205                 slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice;
206         else
207             slice->wNumberMBsInSlice =
208                 mb_count - slice[0].wNumberMBsInSlice;
209
210         memcpy(current, &ctx_pic->bitstream[position], size);
211         current += size;
212     }
213 #if CONFIG_D3D11VA
214     if (ff_dxva2_is_d3d11(avctx))
215         if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
216             return -1;
217 #endif
218 #if CONFIG_DXVA2
219     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
220         if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
221             return -1;
222 #endif
223     if (i < ctx_pic->slice_count)
224         return -1;
225
226 #if CONFIG_D3D11VA
227     if (ff_dxva2_is_d3d11(avctx)) {
228         D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
229         memset(dsc11, 0, sizeof(*dsc11));
230         dsc11->BufferType           = type;
231         dsc11->DataSize             = current - dxva_data;
232         dsc11->NumMBsInBuffer       = mb_count;
233
234         type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
235     }
236 #endif
237 #if CONFIG_DXVA2
238     if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
239         DXVA2_DecodeBufferDesc *dsc2 = bs;
240         memset(dsc2, 0, sizeof(*dsc2));
241         dsc2->CompressedBufferType = type;
242         dsc2->DataSize             = current - dxva_data;
243         dsc2->NumMBsInBuffer       = mb_count;
244
245         type = DXVA2_SliceControlBufferType;
246     }
247 #endif
248
249     return ff_dxva2_commit_buffer(avctx, ctx, sc,
250                                   type,
251                                   ctx_pic->slice,
252                                   ctx_pic->slice_count * sizeof(*ctx_pic->slice),
253                                   mb_count);
254 }
255
256 static int dxva2_mpeg2_start_frame(AVCodecContext *avctx,
257                                    av_unused const uint8_t *buffer,
258                                    av_unused uint32_t size)
259 {
260     const struct MpegEncContext *s = avctx->priv_data;
261     AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
262     struct dxva2_picture_context *ctx_pic =
263         s->current_picture_ptr->hwaccel_picture_private;
264
265     if (!DXVA_CONTEXT_VALID(avctx, ctx))
266         return -1;
267     assert(ctx_pic);
268
269     fill_picture_parameters(avctx, ctx, s, &ctx_pic->pp);
270     fill_quantization_matrices(avctx, ctx, s, &ctx_pic->qm);
271
272     ctx_pic->slice_count    = 0;
273     ctx_pic->bitstream_size = 0;
274     ctx_pic->bitstream      = NULL;
275     return 0;
276 }
277
278 static int dxva2_mpeg2_decode_slice(AVCodecContext *avctx,
279                                     const uint8_t *buffer, uint32_t size)
280 {
281     const struct MpegEncContext *s = avctx->priv_data;
282     struct dxva2_picture_context *ctx_pic =
283         s->current_picture_ptr->hwaccel_picture_private;
284     unsigned position;
285
286     if (ctx_pic->slice_count >= MAX_SLICES) {
287         avpriv_request_sample(avctx, "%d slices in dxva2",
288                               ctx_pic->slice_count);
289         return -1;
290     }
291     if (!ctx_pic->bitstream)
292         ctx_pic->bitstream = buffer;
293     ctx_pic->bitstream_size += size;
294
295     position = buffer - ctx_pic->bitstream;
296     fill_slice(avctx, s, &ctx_pic->slice[ctx_pic->slice_count++], position,
297                buffer, size);
298     return 0;
299 }
300
301 static int dxva2_mpeg2_end_frame(AVCodecContext *avctx)
302 {
303     struct MpegEncContext *s = avctx->priv_data;
304     struct dxva2_picture_context *ctx_pic =
305         s->current_picture_ptr->hwaccel_picture_private;
306     int ret;
307
308     if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
309         return -1;
310     ret = ff_dxva2_common_end_frame(avctx, s->current_picture_ptr->f,
311                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
312                                     &ctx_pic->qm, sizeof(ctx_pic->qm),
313                                     commit_bitstream_and_slice_buffer);
314     if (!ret)
315         ff_mpeg_draw_horiz_band(s, 0, avctx->height);
316     return ret;
317 }
318
319 #if CONFIG_MPEG2_DXVA2_HWACCEL
320 AVHWAccel ff_mpeg2_dxva2_hwaccel = {
321     .name           = "mpeg2_dxva2",
322     .type           = AVMEDIA_TYPE_VIDEO,
323     .id             = AV_CODEC_ID_MPEG2VIDEO,
324     .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
325     .init           = ff_dxva2_decode_init,
326     .uninit         = ff_dxva2_decode_uninit,
327     .start_frame    = dxva2_mpeg2_start_frame,
328     .decode_slice   = dxva2_mpeg2_decode_slice,
329     .end_frame      = dxva2_mpeg2_end_frame,
330     .frame_params   = ff_dxva2_common_frame_params,
331     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
332     .priv_data_size = sizeof(FFDXVASharedContext),
333 };
334 #endif
335
336 #if CONFIG_MPEG2_D3D11VA_HWACCEL
337 AVHWAccel ff_mpeg2_d3d11va_hwaccel = {
338     .name           = "mpeg2_d3d11va",
339     .type           = AVMEDIA_TYPE_VIDEO,
340     .id             = AV_CODEC_ID_MPEG2VIDEO,
341     .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
342     .init           = ff_dxva2_decode_init,
343     .uninit         = ff_dxva2_decode_uninit,
344     .start_frame    = dxva2_mpeg2_start_frame,
345     .decode_slice   = dxva2_mpeg2_decode_slice,
346     .end_frame      = dxva2_mpeg2_end_frame,
347     .frame_params   = ff_dxva2_common_frame_params,
348     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
349     .priv_data_size = sizeof(FFDXVASharedContext),
350 };
351 #endif
352
353 #if CONFIG_MPEG2_D3D11VA2_HWACCEL
354 AVHWAccel ff_mpeg2_d3d11va2_hwaccel = {
355     .name           = "mpeg2_d3d11va2",
356     .type           = AVMEDIA_TYPE_VIDEO,
357     .id             = AV_CODEC_ID_MPEG2VIDEO,
358     .pix_fmt        = AV_PIX_FMT_D3D11,
359     .init           = ff_dxva2_decode_init,
360     .uninit         = ff_dxva2_decode_uninit,
361     .start_frame    = dxva2_mpeg2_start_frame,
362     .decode_slice   = dxva2_mpeg2_decode_slice,
363     .end_frame      = dxva2_mpeg2_end_frame,
364     .frame_params   = ff_dxva2_common_frame_params,
365     .frame_priv_data_size = sizeof(struct dxva2_picture_context),
366     .priv_data_size = sizeof(FFDXVASharedContext),
367 };
368 #endif