2 * MPEG-2 HW acceleration.
4 * copyright (c) 2010 Laurent Aimar
6 * This file is part of Libav.
8 * Libav 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.
13 * Libav 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.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/log.h"
24 #include "mpegutils.h"
25 #include "mpegvideo.h"
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"
32 #define MAX_SLICES 1024
33 struct dxva2_picture_context {
34 DXVA_PictureParameters pp;
37 DXVA_SliceInfo slice[MAX_SLICES];
39 const uint8_t *bitstream;
40 unsigned bitstream_size;
43 static void fill_picture_parameters(AVCodecContext *avctx,
45 const struct MpegEncContext *s,
46 DXVA_PictureParameters *pp)
48 const Picture *current_picture = s->current_picture_ptr;
49 int is_field = s->picture_structure != PICT_FRAME;
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);
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);
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;
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;
80 pp->bPicSpatialResid8 = 0;
81 pp->bPicOverflowBlocks = 0;
82 pp->bPicExtrapolation = 0;
83 pp->bPicDeblocked = 0;
84 pp->bPicDeblockConfined = 0;
85 pp->bPic4MVallowed = 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;
109 static void fill_quantization_matrices(AVCodecContext *avctx,
111 const struct MpegEncContext *s,
112 DXVA_QmatrixData *qm)
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];;
126 static void fill_slice(AVCodecContext *avctx,
127 const struct MpegEncContext *s,
128 DXVA_SliceInfo *slice,
130 const uint8_t *buffer, unsigned size)
132 int is_field = s->picture_structure != PICT_FRAME;
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;
146 init_get_bits(&gb, &buffer[4], 8 * (size - 4));
148 slice->wQuantizerScaleCode = get_bits(&gb, 5);
149 while (get_bits1(&gb))
152 slice->wMBbitOffset = 4 * 8 + get_bits_count(&gb);
154 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
155 DECODER_BUFFER_DESC *bs,
156 DECODER_BUFFER_DESC *sc)
158 const struct MpegEncContext *s = avctx->priv_data;
159 AVDXVAContext *ctx = avctx->hwaccel_context;
160 struct dxva2_picture_context *ctx_pic =
161 s->current_picture_ptr->hwaccel_picture_private;
162 const int is_field = s->picture_structure != PICT_FRAME;
163 const unsigned mb_count = s->mb_width * (s->mb_height >> is_field);
165 uint8_t *dxva_data, *current, *end;
171 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
172 type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
173 if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
174 D3D11VA_CONTEXT(ctx)->decoder,
176 &dxva_size, &dxva_data_ptr)))
181 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
182 type = DXVA2_BitStreamDateBufferType;
183 if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
185 &dxva_data_ptr, &dxva_size)))
190 dxva_data = dxva_data_ptr;
192 end = dxva_data + dxva_size;
194 for (i = 0; i < ctx_pic->slice_count; i++) {
195 DXVA_SliceInfo *slice = &ctx_pic->slice[i];
196 unsigned position = slice->dwSliceDataLocation;
197 unsigned size = slice->dwSliceBitsInBuffer / 8;
198 if (size > end - current) {
199 av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
202 slice->dwSliceDataLocation = current - dxva_data;
204 if (i < ctx_pic->slice_count - 1)
205 slice->wNumberMBsInSlice =
206 slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice;
208 slice->wNumberMBsInSlice =
209 mb_count - slice[0].wNumberMBsInSlice;
211 memcpy(current, &ctx_pic->bitstream[position], size);
215 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD)
216 if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
220 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
221 if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
224 if (i < ctx_pic->slice_count)
228 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
229 D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
230 memset(dsc11, 0, sizeof(*dsc11));
231 dsc11->BufferType = type;
232 dsc11->DataSize = current - dxva_data;
233 dsc11->NumMBsInBuffer = mb_count;
235 type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
239 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
240 DXVA2_DecodeBufferDesc *dsc2 = bs;
241 memset(dsc2, 0, sizeof(*dsc2));
242 dsc2->CompressedBufferType = type;
243 dsc2->DataSize = current - dxva_data;
244 dsc2->NumMBsInBuffer = mb_count;
246 type = DXVA2_SliceControlBufferType;
250 return ff_dxva2_commit_buffer(avctx, ctx, sc,
253 ctx_pic->slice_count * sizeof(*ctx_pic->slice),
257 static int dxva2_mpeg2_start_frame(AVCodecContext *avctx,
258 av_unused const uint8_t *buffer,
259 av_unused uint32_t size)
261 const struct MpegEncContext *s = avctx->priv_data;
262 AVDXVAContext *ctx = avctx->hwaccel_context;
263 struct dxva2_picture_context *ctx_pic =
264 s->current_picture_ptr->hwaccel_picture_private;
266 if (DXVA_CONTEXT_DECODER(avctx, ctx) == NULL ||
267 DXVA_CONTEXT_CFG(avctx, ctx) == NULL ||
268 DXVA_CONTEXT_COUNT(avctx, ctx) <= 0)
272 fill_picture_parameters(avctx, ctx, s, &ctx_pic->pp);
273 fill_quantization_matrices(avctx, ctx, s, &ctx_pic->qm);
275 ctx_pic->slice_count = 0;
276 ctx_pic->bitstream_size = 0;
277 ctx_pic->bitstream = NULL;
281 static int dxva2_mpeg2_decode_slice(AVCodecContext *avctx,
282 const uint8_t *buffer, uint32_t size)
284 const struct MpegEncContext *s = avctx->priv_data;
285 struct dxva2_picture_context *ctx_pic =
286 s->current_picture_ptr->hwaccel_picture_private;
289 if (ctx_pic->slice_count >= MAX_SLICES) {
290 avpriv_request_sample(avctx, "%d slices in dxva2",
291 ctx_pic->slice_count);
294 if (!ctx_pic->bitstream)
295 ctx_pic->bitstream = buffer;
296 ctx_pic->bitstream_size += size;
298 position = buffer - ctx_pic->bitstream;
299 fill_slice(avctx, s, &ctx_pic->slice[ctx_pic->slice_count++], position,
304 static int dxva2_mpeg2_end_frame(AVCodecContext *avctx)
306 struct MpegEncContext *s = avctx->priv_data;
307 struct dxva2_picture_context *ctx_pic =
308 s->current_picture_ptr->hwaccel_picture_private;
311 if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
313 ret = ff_dxva2_common_end_frame(avctx, s->current_picture_ptr->f,
314 &ctx_pic->pp, sizeof(ctx_pic->pp),
315 &ctx_pic->qm, sizeof(ctx_pic->qm),
316 commit_bitstream_and_slice_buffer);
318 ff_mpeg_draw_horiz_band(s, 0, avctx->height);
322 #if CONFIG_MPEG2_DXVA2_HWACCEL
323 AVHWAccel ff_mpeg2_dxva2_hwaccel = {
324 .name = "mpeg2_dxva2",
325 .type = AVMEDIA_TYPE_VIDEO,
326 .id = AV_CODEC_ID_MPEG2VIDEO,
327 .pix_fmt = AV_PIX_FMT_DXVA2_VLD,
328 .start_frame = dxva2_mpeg2_start_frame,
329 .decode_slice = dxva2_mpeg2_decode_slice,
330 .end_frame = dxva2_mpeg2_end_frame,
331 .frame_priv_data_size = sizeof(struct dxva2_picture_context),
335 #if CONFIG_MPEG2_D3D11VA_HWACCEL
336 AVHWAccel ff_mpeg2_d3d11va_hwaccel = {
337 .name = "mpeg2_d3d11va",
338 .type = AVMEDIA_TYPE_VIDEO,
339 .id = AV_CODEC_ID_MPEG2VIDEO,
340 .pix_fmt = AV_PIX_FMT_D3D11VA_VLD,
341 .start_frame = dxva2_mpeg2_start_frame,
342 .decode_slice = dxva2_mpeg2_decode_slice,
343 .end_frame = dxva2_mpeg2_end_frame,
344 .frame_priv_data_size = sizeof(struct dxva2_picture_context),