2 * DXVA2 H264 HW acceleration.
4 * copyright (c) 2009 Laurent Aimar
6 * This file is part of FFmpeg.
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.
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.
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
25 #include "mpegutils.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 struct dxva2_picture_context {
33 DXVA_PicParams_H264 pp;
36 DXVA_Slice_H264_Short slice_short[MAX_SLICES];
37 DXVA_Slice_H264_Long slice_long[MAX_SLICES];
38 const uint8_t *bitstream;
39 unsigned bitstream_size;
42 static void fill_picture_entry(DXVA_PicEntry_H264 *pic,
43 unsigned index, unsigned flag)
45 assert((index&0x7f) == index && (flag&0x01) == flag);
46 pic->bPicEntry = index | (flag << 7);
49 static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const H264Context *h,
50 DXVA_PicParams_H264 *pp)
52 const H264Picture *current_picture = h->cur_pic_ptr;
55 memset(pp, 0, sizeof(*pp));
56 /* Configure current picture */
57 fill_picture_entry(&pp->CurrPic,
58 ff_dxva2_get_surface_index(avctx, ctx, current_picture->f),
59 h->picture_structure == PICT_BOTTOM_FIELD);
60 /* Configure the set of references */
61 pp->UsedForReferenceFlags = 0;
62 pp->NonExistingFrameFlags = 0;
63 for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) {
65 if (j < h->short_ref_count) {
66 r = h->short_ref[j++];
69 while (!r && j < h->short_ref_count + 16)
70 r = h->long_ref[j++ - h->short_ref_count];
73 fill_picture_entry(&pp->RefFrameList[i],
74 ff_dxva2_get_surface_index(avctx, ctx, r->f),
77 if ((r->reference & PICT_TOP_FIELD) && r->field_poc[0] != INT_MAX)
78 pp->FieldOrderCntList[i][0] = r->field_poc[0];
79 if ((r->reference & PICT_BOTTOM_FIELD) && r->field_poc[1] != INT_MAX)
80 pp->FieldOrderCntList[i][1] = r->field_poc[1];
82 pp->FrameNumList[i] = r->long_ref ? r->pic_id : r->frame_num;
83 if (r->reference & PICT_TOP_FIELD)
84 pp->UsedForReferenceFlags |= 1 << (2*i + 0);
85 if (r->reference & PICT_BOTTOM_FIELD)
86 pp->UsedForReferenceFlags |= 1 << (2*i + 1);
88 pp->RefFrameList[i].bPicEntry = 0xff;
89 pp->FieldOrderCntList[i][0] = 0;
90 pp->FieldOrderCntList[i][1] = 0;
91 pp->FrameNumList[i] = 0;
95 pp->wFrameWidthInMbsMinus1 = h->mb_width - 1;
96 pp->wFrameHeightInMbsMinus1 = h->mb_height - 1;
97 pp->num_ref_frames = h->sps.ref_frame_count;
99 pp->wBitFields = ((h->picture_structure != PICT_FRAME) << 0) |
101 (h->picture_structure == PICT_FRAME)) << 1) |
102 (h->sps.residual_color_transform_flag << 2) |
103 /* sp_for_switch_flag (not implemented by FFmpeg) */
105 (h->sps.chroma_format_idc << 4) |
106 ((h->nal_ref_idc != 0) << 6) |
107 (h->pps.constrained_intra_pred << 7) |
108 (h->pps.weighted_pred << 8) |
109 (h->pps.weighted_bipred_idc << 9) |
110 /* MbsConsecutiveFlag */
112 (h->sps.frame_mbs_only_flag << 12) |
113 (h->pps.transform_8x8_mode << 13) |
114 ((h->sps.level_idc >= 31) << 14) |
115 /* IntraPicFlag (Modified if we detect a non
116 * intra slice in dxva2_h264_decode_slice) */
119 pp->bit_depth_luma_minus8 = h->sps.bit_depth_luma - 8;
120 pp->bit_depth_chroma_minus8 = h->sps.bit_depth_chroma - 8;
121 if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG)
122 pp->Reserved16Bits = 0;
123 else if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO)
124 pp->Reserved16Bits = 0x34c;
126 pp->Reserved16Bits = 3; /* FIXME is there a way to detect the right mode ? */
127 pp->StatusReportFeedbackNumber = 1 + DXVA_CONTEXT_REPORT_ID(avctx, ctx)++;
128 pp->CurrFieldOrderCnt[0] = 0;
129 if ((h->picture_structure & PICT_TOP_FIELD) &&
130 current_picture->field_poc[0] != INT_MAX)
131 pp->CurrFieldOrderCnt[0] = current_picture->field_poc[0];
132 pp->CurrFieldOrderCnt[1] = 0;
133 if ((h->picture_structure & PICT_BOTTOM_FIELD) &&
134 current_picture->field_poc[1] != INT_MAX)
135 pp->CurrFieldOrderCnt[1] = current_picture->field_poc[1];
136 pp->pic_init_qs_minus26 = h->pps.init_qs - 26;
137 pp->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0];
138 pp->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1];
139 pp->ContinuationFlag = 1;
140 pp->pic_init_qp_minus26 = h->pps.init_qp - 26;
141 pp->num_ref_idx_l0_active_minus1 = h->pps.ref_count[0] - 1;
142 pp->num_ref_idx_l1_active_minus1 = h->pps.ref_count[1] - 1;
143 pp->Reserved8BitsA = 0;
144 pp->frame_num = h->frame_num;
145 pp->log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4;
146 pp->pic_order_cnt_type = h->sps.poc_type;
147 if (h->sps.poc_type == 0)
148 pp->log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4;
149 else if (h->sps.poc_type == 1)
150 pp->delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
151 pp->direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag;
152 pp->entropy_coding_mode_flag = h->pps.cabac;
153 pp->pic_order_present_flag = h->pps.pic_order_present;
154 pp->num_slice_groups_minus1 = h->pps.slice_group_count - 1;
155 pp->slice_group_map_type = h->pps.mb_slice_group_map_type;
156 pp->deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
157 pp->redundant_pic_cnt_present_flag= h->pps.redundant_pic_cnt_present;
158 pp->Reserved8BitsB = 0;
159 pp->slice_group_change_rate_minus1= 0; /* XXX not implemented by FFmpeg */
160 //pp->SliceGroupMap[810]; /* XXX not implemented by FFmpeg */
163 static void fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, const H264Context *h, DXVA_Qmatrix_H264 *qm)
166 memset(qm, 0, sizeof(*qm));
167 if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG) {
168 for (i = 0; i < 6; i++)
169 for (j = 0; j < 16; j++)
170 qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][j];
172 for (i = 0; i < 64; i++) {
173 qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][i];
174 qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][i];
177 for (i = 0; i < 6; i++)
178 for (j = 0; j < 16; j++)
179 qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][zigzag_scan[j]];
181 for (i = 0; i < 64; i++) {
182 qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][ff_zigzag_direct[i]];
183 qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][ff_zigzag_direct[i]];
188 static int is_slice_short(const AVCodecContext *avctx, AVDXVAContext *ctx)
190 assert(DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) == 1 ||
191 DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) == 2);
192 return DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) == 2;
195 static void fill_slice_short(DXVA_Slice_H264_Short *slice,
196 unsigned position, unsigned size)
198 memset(slice, 0, sizeof(*slice));
199 slice->BSNALunitDataLocation = position;
200 slice->SliceBytesInBuffer = size;
201 slice->wBadSliceChopping = 0;
204 static int get_refpic_index(const DXVA_PicParams_H264 *pp, int surface_index)
207 for (i = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) {
208 if ((pp->RefFrameList[i].bPicEntry & 0x7f) == surface_index)
214 static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
215 const DXVA_PicParams_H264 *pp, unsigned position, unsigned size)
217 const H264Context *h = avctx->priv_data;
218 H264SliceContext *sl = &h->slice_ctx[0];
219 AVDXVAContext *ctx = avctx->hwaccel_context;
222 memset(slice, 0, sizeof(*slice));
223 slice->BSNALunitDataLocation = position;
224 slice->SliceBytesInBuffer = size;
225 slice->wBadSliceChopping = 0;
227 slice->first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x;
228 slice->NumMbsForSlice = 0; /* XXX it is set once we have all slices */
229 slice->BitOffsetToSliceData = get_bits_count(&sl->gb);
230 slice->slice_type = ff_h264_get_slice_type(sl);
231 if (sl->slice_type_fixed)
232 slice->slice_type += 5;
233 slice->luma_log2_weight_denom = sl->luma_log2_weight_denom;
234 slice->chroma_log2_weight_denom = sl->chroma_log2_weight_denom;
235 if (sl->list_count > 0)
236 slice->num_ref_idx_l0_active_minus1 = sl->ref_count[0] - 1;
237 if (sl->list_count > 1)
238 slice->num_ref_idx_l1_active_minus1 = sl->ref_count[1] - 1;
239 slice->slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2;
240 slice->slice_beta_offset_div2 = sl->slice_beta_offset / 2;
241 slice->Reserved8Bits = 0;
243 for (list = 0; list < 2; list++) {
245 for (i = 0; i < FF_ARRAY_ELEMS(slice->RefPicList[list]); i++) {
246 if (list < sl->list_count && i < sl->ref_count[list]) {
247 const H264Picture *r = sl->ref_list[list][i].parent;
250 if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO)
251 index = ff_dxva2_get_surface_index(avctx, ctx, r->f);
253 index = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, r->f));
254 fill_picture_entry(&slice->RefPicList[list][i], index,
255 r->reference == PICT_BOTTOM_FIELD);
256 for (plane = 0; plane < 3; plane++) {
258 if (plane == 0 && sl->luma_weight_flag[list]) {
259 w = sl->luma_weight[i][list][0];
260 o = sl->luma_weight[i][list][1];
261 } else if (plane >= 1 && sl->chroma_weight_flag[list]) {
262 w = sl->chroma_weight[i][list][plane-1][0];
263 o = sl->chroma_weight[i][list][plane-1][1];
265 w = 1 << (plane == 0 ? sl->luma_log2_weight_denom :
266 sl->chroma_log2_weight_denom);
269 slice->Weights[list][i][plane][0] = w;
270 slice->Weights[list][i][plane][1] = o;
274 slice->RefPicList[list][i].bPicEntry = 0xff;
275 for (plane = 0; plane < 3; plane++) {
276 slice->Weights[list][i][plane][0] = 0;
277 slice->Weights[list][i][plane][1] = 0;
282 slice->slice_qs_delta = 0; /* XXX not implemented by FFmpeg */
283 slice->slice_qp_delta = sl->qscale - h->pps.init_qp;
284 slice->redundant_pic_cnt = sl->redundant_pic_count;
285 if (sl->slice_type == AV_PICTURE_TYPE_B)
286 slice->direct_spatial_mv_pred_flag = sl->direct_spatial_mv_pred;
287 slice->cabac_init_idc = h->pps.cabac ? sl->cabac_init_idc : 0;
288 if (sl->deblocking_filter < 2)
289 slice->disable_deblocking_filter_idc = 1 - sl->deblocking_filter;
291 slice->disable_deblocking_filter_idc = sl->deblocking_filter;
292 slice->slice_id = h->current_slice - 1;
295 static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
296 DECODER_BUFFER_DESC *bs,
297 DECODER_BUFFER_DESC *sc)
299 const H264Context *h = avctx->priv_data;
300 const unsigned mb_count = h->mb_width * h->mb_height;
301 AVDXVAContext *ctx = avctx->hwaccel_context;
302 const H264Picture *current_picture = h->cur_pic_ptr;
303 struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
304 DXVA_Slice_H264_Short *slice = NULL;
305 void *dxva_data_ptr = NULL;
306 uint8_t *dxva_data, *current, *end;
307 unsigned dxva_size = 0;
314 /* Create an annex B bitstream buffer with only slice NAL and finalize slice */
316 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
317 type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
318 if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
319 D3D11VA_CONTEXT(ctx)->decoder,
321 &dxva_size, &dxva_data_ptr)))
326 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
327 type = DXVA2_BitStreamDateBufferType;
328 if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
330 &dxva_data_ptr, &dxva_size)))
335 dxva_data = dxva_data_ptr;
337 end = dxva_data + dxva_size;
339 for (i = 0; i < ctx_pic->slice_count; i++) {
340 static const uint8_t start_code[] = { 0, 0, 1 };
341 static const unsigned start_code_size = sizeof(start_code);
342 unsigned position, size;
344 assert(offsetof(DXVA_Slice_H264_Short, BSNALunitDataLocation) ==
345 offsetof(DXVA_Slice_H264_Long, BSNALunitDataLocation));
346 assert(offsetof(DXVA_Slice_H264_Short, SliceBytesInBuffer) ==
347 offsetof(DXVA_Slice_H264_Long, SliceBytesInBuffer));
349 if (is_slice_short(avctx, ctx))
350 slice = &ctx_pic->slice_short[i];
352 slice = (DXVA_Slice_H264_Short*)&ctx_pic->slice_long[i];
354 position = slice->BSNALunitDataLocation;
355 size = slice->SliceBytesInBuffer;
356 if (start_code_size + size > end - current) {
357 av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream");
361 slice->BSNALunitDataLocation = current - dxva_data;
362 slice->SliceBytesInBuffer = start_code_size + size;
364 if (!is_slice_short(avctx, ctx)) {
365 DXVA_Slice_H264_Long *slice_long = (DXVA_Slice_H264_Long*)slice;
366 if (i < ctx_pic->slice_count - 1)
367 slice_long->NumMbsForSlice =
368 slice_long[1].first_mb_in_slice - slice_long[0].first_mb_in_slice;
370 slice_long->NumMbsForSlice = mb_count - slice_long->first_mb_in_slice;
373 memcpy(current, start_code, start_code_size);
374 current += start_code_size;
376 memcpy(current, &ctx_pic->bitstream[position], size);
379 padding = FFMIN(128 - ((current - dxva_data) & 127), end - current);
380 if (slice && padding > 0) {
381 memset(current, 0, padding);
384 slice->SliceBytesInBuffer += padding;
387 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD)
388 if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
392 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
393 if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
396 if (i < ctx_pic->slice_count)
400 if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
401 D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
402 memset(dsc11, 0, sizeof(*dsc11));
403 dsc11->BufferType = type;
404 dsc11->DataSize = current - dxva_data;
405 dsc11->NumMBsInBuffer = mb_count;
407 type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
411 if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
412 DXVA2_DecodeBufferDesc *dsc2 = bs;
413 memset(dsc2, 0, sizeof(*dsc2));
414 dsc2->CompressedBufferType = type;
415 dsc2->DataSize = current - dxva_data;
416 dsc2->NumMBsInBuffer = mb_count;
418 type = DXVA2_SliceControlBufferType;
422 if (is_slice_short(avctx, ctx)) {
423 slice_data = ctx_pic->slice_short;
424 slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_short);
426 slice_data = ctx_pic->slice_long;
427 slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_long);
429 assert((bs->DataSize & 127) == 0);
430 return ff_dxva2_commit_buffer(avctx, ctx, sc,
432 slice_data, slice_size, mb_count);
436 static int dxva2_h264_start_frame(AVCodecContext *avctx,
437 av_unused const uint8_t *buffer,
438 av_unused uint32_t size)
440 const H264Context *h = avctx->priv_data;
441 AVDXVAContext *ctx = avctx->hwaccel_context;
442 struct dxva2_picture_context *ctx_pic = h->cur_pic_ptr->hwaccel_picture_private;
444 if (DXVA_CONTEXT_DECODER(avctx, ctx) == NULL ||
445 DXVA_CONTEXT_CFG(avctx, ctx) == NULL ||
446 DXVA_CONTEXT_COUNT(avctx, ctx) <= 0)
450 /* Fill up DXVA_PicParams_H264 */
451 fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp);
453 /* Fill up DXVA_Qmatrix_H264 */
454 fill_scaling_lists(avctx, ctx, h, &ctx_pic->qm);
456 ctx_pic->slice_count = 0;
457 ctx_pic->bitstream_size = 0;
458 ctx_pic->bitstream = NULL;
462 static int dxva2_h264_decode_slice(AVCodecContext *avctx,
463 const uint8_t *buffer,
466 const H264Context *h = avctx->priv_data;
467 const H264SliceContext *sl = &h->slice_ctx[0];
468 AVDXVAContext *ctx = avctx->hwaccel_context;
469 const H264Picture *current_picture = h->cur_pic_ptr;
470 struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
473 if (ctx_pic->slice_count >= MAX_SLICES)
476 if (!ctx_pic->bitstream)
477 ctx_pic->bitstream = buffer;
478 ctx_pic->bitstream_size += size;
480 position = buffer - ctx_pic->bitstream;
481 if (is_slice_short(avctx, ctx))
482 fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count],
485 fill_slice_long(avctx, &ctx_pic->slice_long[ctx_pic->slice_count],
486 &ctx_pic->pp, position, size);
487 ctx_pic->slice_count++;
489 if (sl->slice_type != AV_PICTURE_TYPE_I && sl->slice_type != AV_PICTURE_TYPE_SI)
490 ctx_pic->pp.wBitFields &= ~(1 << 15); /* Set IntraPicFlag to 0 */
494 static int dxva2_h264_end_frame(AVCodecContext *avctx)
496 H264Context *h = avctx->priv_data;
497 H264SliceContext *sl = &h->slice_ctx[0];
498 struct dxva2_picture_context *ctx_pic =
499 h->cur_pic_ptr->hwaccel_picture_private;
502 if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
504 ret = ff_dxva2_common_end_frame(avctx, h->cur_pic_ptr->f,
505 &ctx_pic->pp, sizeof(ctx_pic->pp),
506 &ctx_pic->qm, sizeof(ctx_pic->qm),
507 commit_bitstream_and_slice_buffer);
509 ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height);
513 #if CONFIG_H264_DXVA2_HWACCEL
514 AVHWAccel ff_h264_dxva2_hwaccel = {
515 .name = "h264_dxva2",
516 .type = AVMEDIA_TYPE_VIDEO,
517 .id = AV_CODEC_ID_H264,
518 .pix_fmt = AV_PIX_FMT_DXVA2_VLD,
519 .start_frame = dxva2_h264_start_frame,
520 .decode_slice = dxva2_h264_decode_slice,
521 .end_frame = dxva2_h264_end_frame,
522 .frame_priv_data_size = sizeof(struct dxva2_picture_context),
526 #if CONFIG_H264_D3D11VA_HWACCEL
527 AVHWAccel ff_h264_d3d11va_hwaccel = {
528 .name = "h264_d3d11va",
529 .type = AVMEDIA_TYPE_VIDEO,
530 .id = AV_CODEC_ID_H264,
531 .pix_fmt = AV_PIX_FMT_D3D11VA_VLD,
532 .start_frame = dxva2_h264_start_frame,
533 .decode_slice = dxva2_h264_decode_slice,
534 .end_frame = dxva2_h264_end_frame,
535 .frame_priv_data_size = sizeof(struct dxva2_picture_context),