X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdxva2_h264.c;h=093db99ec3714eccfacbad81c0444ee97d74b3c8;hb=a4edaeb50fc7510f28a5d79349a7926a182c9930;hp=f9dc7205b8960c7528ddf5d3111c97b99e62c6b6;hpb=2912e87a6c9264d556734e2bf94a99c64cf9b102;p=ffmpeg diff --git a/libavcodec/dxva2_h264.c b/libavcodec/dxva2_h264.c index f9dc7205b89..093db99ec37 100644 --- a/libavcodec/dxva2_h264.c +++ b/libavcodec/dxva2_h264.c @@ -23,6 +23,7 @@ #include "dxva2_internal.h" #include "h264.h" #include "h264data.h" +#include "mpegutils.h" struct dxva2_picture_context { DXVA_PicParams_H264 pp; @@ -44,20 +45,19 @@ static void fill_picture_entry(DXVA_PicEntry_H264 *pic, static void fill_picture_parameters(struct dxva_context *ctx, const H264Context *h, DXVA_PicParams_H264 *pp) { - const MpegEncContext *s = &h->s; - const Picture *current_picture = s->current_picture_ptr; + const H264Picture *current_picture = h->cur_pic_ptr; int i, j; memset(pp, 0, sizeof(*pp)); /* Configure current picture */ fill_picture_entry(&pp->CurrPic, - ff_dxva2_get_surface_index(ctx, current_picture), - s->picture_structure == PICT_BOTTOM_FIELD); + ff_dxva2_get_surface_index(ctx, ¤t_picture->f), + h->picture_structure == PICT_BOTTOM_FIELD); /* Configure the set of references */ pp->UsedForReferenceFlags = 0; pp->NonExistingFrameFlags = 0; for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) { - const Picture *r; + const H264Picture *r; if (j < h->short_ref_count) { r = h->short_ref[j++]; } else { @@ -67,7 +67,7 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context } if (r) { fill_picture_entry(&pp->RefFrameList[i], - ff_dxva2_get_surface_index(ctx, r), + ff_dxva2_get_surface_index(ctx, &r->f), r->long_ref != 0); if ((r->reference & PICT_TOP_FIELD) && r->field_poc[0] != INT_MAX) @@ -88,14 +88,15 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context } } - pp->wFrameWidthInMbsMinus1 = s->mb_width - 1; - pp->wFrameHeightInMbsMinus1 = s->mb_height - 1; + pp->wFrameWidthInMbsMinus1 = h->mb_width - 1; + pp->wFrameHeightInMbsMinus1 = h->mb_height - 1; pp->num_ref_frames = h->sps.ref_frame_count; - pp->wBitFields = ((s->picture_structure != PICT_FRAME) << 0) | - (h->sps.mb_aff << 1) | + pp->wBitFields = ((h->picture_structure != PICT_FRAME) << 0) | + ((h->sps.mb_aff && + (h->picture_structure == PICT_FRAME)) << 1) | (h->sps.residual_color_transform_flag << 2) | - /* sp_for_switch_flag (not implemented by FFmpeg) */ + /* sp_for_switch_flag (not implemented by Libav) */ (0 << 3) | (h->sps.chroma_format_idc << 4) | ((h->nal_ref_idc != 0) << 6) | @@ -108,19 +109,24 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context (h->pps.transform_8x8_mode << 13) | ((h->sps.level_idc >= 31) << 14) | /* IntraPicFlag (Modified if we detect a non - * intra slice in decode_slice) */ + * intra slice in dxva2_h264_decode_slice) */ (1 << 15); pp->bit_depth_luma_minus8 = h->sps.bit_depth_luma - 8; pp->bit_depth_chroma_minus8 = h->sps.bit_depth_chroma - 8; - pp->Reserved16Bits = 3; /* FIXME is there a way to detect the right mode ? */ + if (ctx->workaround & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG) + pp->Reserved16Bits = 0; + else if (ctx->workaround & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO) + pp->Reserved16Bits = 0x34c; + else + pp->Reserved16Bits = 3; /* FIXME is there a way to detect the right mode ? */ pp->StatusReportFeedbackNumber = 1 + ctx->report_id++; pp->CurrFieldOrderCnt[0] = 0; - if ((s->picture_structure & PICT_TOP_FIELD) && + if ((h->picture_structure & PICT_TOP_FIELD) && current_picture->field_poc[0] != INT_MAX) pp->CurrFieldOrderCnt[0] = current_picture->field_poc[0]; pp->CurrFieldOrderCnt[1] = 0; - if ((s->picture_structure & PICT_BOTTOM_FIELD) && + if ((h->picture_structure & PICT_BOTTOM_FIELD) && current_picture->field_poc[1] != INT_MAX) pp->CurrFieldOrderCnt[1] = current_picture->field_poc[1]; pp->pic_init_qs_minus26 = h->pps.init_qs - 26; @@ -146,21 +152,33 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context pp->deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present; pp->redundant_pic_cnt_present_flag= h->pps.redundant_pic_cnt_present; pp->Reserved8BitsB = 0; - pp->slice_group_change_rate_minus1= 0; /* XXX not implemented by FFmpeg */ - //pp->SliceGroupMap[810]; /* XXX not implemented by FFmpeg */ + pp->slice_group_change_rate_minus1= 0; /* XXX not implemented by Libav */ + //pp->SliceGroupMap[810]; /* XXX not implemented by Libav */ } -static void fill_scaling_lists(const H264Context *h, DXVA_Qmatrix_H264 *qm) +static void fill_scaling_lists(struct dxva_context *ctx, const H264Context *h, DXVA_Qmatrix_H264 *qm) { unsigned i, j; memset(qm, 0, sizeof(*qm)); - for (i = 0; i < 6; i++) - for (j = 0; j < 16; j++) - qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][zigzag_scan[j]]; + if (ctx->workaround & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG) { + for (i = 0; i < 6; i++) + for (j = 0; j < 16; j++) + qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][j]; + + for (i = 0; i < 64; i++) { + qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][i]; + qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][i]; + } + } else { + for (i = 0; i < 6; i++) + for (j = 0; j < 16; j++) + qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][zigzag_scan[j]]; - for (i = 0; i < 2; i++) - for (j = 0; j < 64; j++) - qm->bScalingLists8x8[i][j] = h->pps.scaling_matrix8[i][ff_zigzag_direct[j]]; + for (i = 0; i < 64; i++) { + qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][ff_zigzag_direct[i]]; + qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][ff_zigzag_direct[i]]; + } + } } static int is_slice_short(struct dxva_context *ctx) @@ -179,12 +197,22 @@ static void fill_slice_short(DXVA_Slice_H264_Short *slice, slice->wBadSliceChopping = 0; } +static int get_refpic_index(const DXVA_PicParams_H264 *pp, int surface_index) +{ + int i; + for (i = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) { + if ((pp->RefFrameList[i].bPicEntry & 0x7f) == surface_index) + return i; + } + return 0x7f; +} + static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice, - unsigned position, unsigned size) + const DXVA_PicParams_H264 *pp, unsigned position, unsigned size) { const H264Context *h = avctx->priv_data; + H264SliceContext *sl = &h->slice_ctx[0]; struct dxva_context *ctx = avctx->hwaccel_context; - const MpegEncContext *s = &h->s; unsigned list; memset(slice, 0, sizeof(*slice)); @@ -192,42 +220,46 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice, slice->SliceBytesInBuffer = size; slice->wBadSliceChopping = 0; - slice->first_mb_in_slice = (s->mb_y >> FIELD_OR_MBAFF_PICTURE) * s->mb_width + s->mb_x; + slice->first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x; slice->NumMbsForSlice = 0; /* XXX it is set once we have all slices */ - slice->BitOffsetToSliceData = get_bits_count(&s->gb); - slice->slice_type = ff_h264_get_slice_type(h); - if (h->slice_type_fixed) + slice->BitOffsetToSliceData = get_bits_count(&sl->gb); + slice->slice_type = ff_h264_get_slice_type(sl); + if (sl->slice_type_fixed) slice->slice_type += 5; - slice->luma_log2_weight_denom = h->luma_log2_weight_denom; - slice->chroma_log2_weight_denom = h->chroma_log2_weight_denom; - if (h->list_count > 0) - slice->num_ref_idx_l0_active_minus1 = h->ref_count[0] - 1; - if (h->list_count > 1) - slice->num_ref_idx_l1_active_minus1 = h->ref_count[1] - 1; - slice->slice_alpha_c0_offset_div2 = h->slice_alpha_c0_offset / 2 - 26; - slice->slice_beta_offset_div2 = h->slice_beta_offset / 2 - 26; + slice->luma_log2_weight_denom = sl->luma_log2_weight_denom; + slice->chroma_log2_weight_denom = sl->chroma_log2_weight_denom; + if (sl->list_count > 0) + slice->num_ref_idx_l0_active_minus1 = sl->ref_count[0] - 1; + if (sl->list_count > 1) + slice->num_ref_idx_l1_active_minus1 = sl->ref_count[1] - 1; + slice->slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2; + slice->slice_beta_offset_div2 = sl->slice_beta_offset / 2; slice->Reserved8Bits = 0; for (list = 0; list < 2; list++) { unsigned i; for (i = 0; i < FF_ARRAY_ELEMS(slice->RefPicList[list]); i++) { - if (list < h->list_count && i < h->ref_count[list]) { - const Picture *r = &h->ref_list[list][i]; + if (list < sl->list_count && i < sl->ref_count[list]) { + const H264Picture *r = sl->ref_list[list][i].parent; unsigned plane; - fill_picture_entry(&slice->RefPicList[list][i], - ff_dxva2_get_surface_index(ctx, r), + unsigned index; + if (ctx->workaround & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO) + index = ff_dxva2_get_surface_index(ctx, &r->f); + else + index = get_refpic_index(pp, ff_dxva2_get_surface_index(ctx, &r->f)); + fill_picture_entry(&slice->RefPicList[list][i], index, r->reference == PICT_BOTTOM_FIELD); for (plane = 0; plane < 3; plane++) { int w, o; - if (plane == 0 && h->luma_weight_flag[list]) { - w = h->luma_weight[i][list][0]; - o = h->luma_weight[i][list][1]; - } else if (plane >= 1 && h->chroma_weight_flag[list]) { - w = h->chroma_weight[i][list][plane-1][0]; - o = h->chroma_weight[i][list][plane-1][1]; + if (plane == 0 && sl->luma_weight_flag[list]) { + w = sl->luma_weight[i][list][0]; + o = sl->luma_weight[i][list][1]; + } else if (plane >= 1 && sl->chroma_weight_flag[list]) { + w = sl->chroma_weight[i][list][plane-1][0]; + o = sl->chroma_weight[i][list][plane-1][1]; } else { - w = 1 << (plane == 0 ? h->luma_log2_weight_denom : - h->chroma_log2_weight_denom); + w = 1 << (plane == 0 ? sl->luma_log2_weight_denom : + sl->chroma_log2_weight_denom); o = 0; } slice->Weights[list][i][plane][0] = w; @@ -243,16 +275,16 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice, } } } - slice->slice_qs_delta = 0; /* XXX not implemented by FFmpeg */ - slice->slice_qp_delta = s->qscale - h->pps.init_qp; - slice->redundant_pic_cnt = h->redundant_pic_count; - if (h->slice_type == FF_B_TYPE) - slice->direct_spatial_mv_pred_flag = h->direct_spatial_mv_pred; - slice->cabac_init_idc = h->pps.cabac ? h->cabac_init_idc : 0; - if (h->deblocking_filter < 2) - slice->disable_deblocking_filter_idc = 1 - h->deblocking_filter; + slice->slice_qs_delta = 0; /* XXX not implemented by Libav */ + slice->slice_qp_delta = sl->qscale - h->pps.init_qp; + slice->redundant_pic_cnt = sl->redundant_pic_count; + if (sl->slice_type == AV_PICTURE_TYPE_B) + slice->direct_spatial_mv_pred_flag = sl->direct_spatial_mv_pred; + slice->cabac_init_idc = h->pps.cabac ? sl->cabac_init_idc : 0; + if (sl->deblocking_filter < 2) + slice->disable_deblocking_filter_idc = 1 - sl->deblocking_filter; else - slice->disable_deblocking_filter_idc = h->deblocking_filter; + slice->disable_deblocking_filter_idc = sl->deblocking_filter; slice->slice_id = h->current_slice - 1; } @@ -261,12 +293,12 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx, DXVA2_DecodeBufferDesc *sc) { const H264Context *h = avctx->priv_data; - const MpegEncContext *s = &h->s; - const unsigned mb_count = s->mb_width * s->mb_height; + const unsigned mb_count = h->mb_width * h->mb_height; struct dxva_context *ctx = avctx->hwaccel_context; - const Picture *current_picture = h->s.current_picture_ptr; + const H264Picture *current_picture = h->cur_pic_ptr; struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private; DXVA_Slice_H264_Short *slice = NULL; + void *dxva_data_ptr; uint8_t *dxva_data, *current, *end; unsigned dxva_size; void *slice_data; @@ -276,9 +308,11 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx, /* Create an annex B bitstream buffer with only slice NAL and finalize slice */ if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder, - DXVA2_BitStreamDateBufferType, - &dxva_data, &dxva_size))) + DXVA2_BitStreamDateBufferType, + &dxva_data_ptr, &dxva_size))) return -1; + + dxva_data = dxva_data_ptr; current = dxva_data; end = dxva_data + dxva_size; @@ -354,13 +388,13 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx, } -static int start_frame(AVCodecContext *avctx, - av_unused const uint8_t *buffer, - av_unused uint32_t size) +static int dxva2_h264_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) { const H264Context *h = avctx->priv_data; struct dxva_context *ctx = avctx->hwaccel_context; - struct dxva2_picture_context *ctx_pic = h->s.current_picture_ptr->hwaccel_picture_private; + struct dxva2_picture_context *ctx_pic = h->cur_pic_ptr->hwaccel_picture_private; if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0) return -1; @@ -370,7 +404,7 @@ static int start_frame(AVCodecContext *avctx, fill_picture_parameters(ctx, h, &ctx_pic->pp); /* Fill up DXVA_Qmatrix_H264 */ - fill_scaling_lists(h, &ctx_pic->qm); + fill_scaling_lists(ctx, h, &ctx_pic->qm); ctx_pic->slice_count = 0; ctx_pic->bitstream_size = 0; @@ -378,12 +412,14 @@ static int start_frame(AVCodecContext *avctx, return 0; } -static int decode_slice(AVCodecContext *avctx, - const uint8_t *buffer, uint32_t size) +static int dxva2_h264_decode_slice(AVCodecContext *avctx, + const uint8_t *buffer, + uint32_t size) { const H264Context *h = avctx->priv_data; + const H264SliceContext *sl = &h->slice_ctx[0]; struct dxva_context *ctx = avctx->hwaccel_context; - const Picture *current_picture = h->s.current_picture_ptr; + const H264Picture *current_picture = h->cur_pic_ptr; struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private; unsigned position; @@ -400,38 +436,40 @@ static int decode_slice(AVCodecContext *avctx, position, size); else fill_slice_long(avctx, &ctx_pic->slice_long[ctx_pic->slice_count], - position, size); + &ctx_pic->pp, position, size); ctx_pic->slice_count++; - if (h->slice_type != FF_I_TYPE && h->slice_type != FF_SI_TYPE) + if (sl->slice_type != AV_PICTURE_TYPE_I && sl->slice_type != AV_PICTURE_TYPE_SI) ctx_pic->pp.wBitFields &= ~(1 << 15); /* Set IntraPicFlag to 0 */ return 0; } -static int end_frame(AVCodecContext *avctx) +static int dxva2_h264_end_frame(AVCodecContext *avctx) { H264Context *h = avctx->priv_data; - MpegEncContext *s = &h->s; + H264SliceContext *sl = &h->slice_ctx[0]; struct dxva2_picture_context *ctx_pic = - h->s.current_picture_ptr->hwaccel_picture_private; + h->cur_pic_ptr->hwaccel_picture_private; + int ret; if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) return -1; - return ff_dxva2_common_end_frame(avctx, s, - &ctx_pic->pp, sizeof(ctx_pic->pp), - &ctx_pic->qm, sizeof(ctx_pic->qm), - commit_bitstream_and_slice_buffer); + ret = ff_dxva2_common_end_frame(avctx, &h->cur_pic_ptr->f, + &ctx_pic->pp, sizeof(ctx_pic->pp), + &ctx_pic->qm, sizeof(ctx_pic->qm), + commit_bitstream_and_slice_buffer); + if (!ret) + ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height); + return ret; } AVHWAccel ff_h264_dxva2_hwaccel = { .name = "h264_dxva2", .type = AVMEDIA_TYPE_VIDEO, - .id = CODEC_ID_H264, - .pix_fmt = PIX_FMT_DXVA2_VLD, - .capabilities = 0, - .start_frame = start_frame, - .decode_slice = decode_slice, - .end_frame = end_frame, - .priv_data_size = sizeof(struct dxva2_picture_context), + .id = AV_CODEC_ID_H264, + .pix_fmt = AV_PIX_FMT_DXVA2_VLD, + .start_frame = dxva2_h264_start_frame, + .decode_slice = dxva2_h264_decode_slice, + .end_frame = dxva2_h264_end_frame, + .frame_priv_data_size = sizeof(struct dxva2_picture_context), }; -