From 92c6a099bafd10e5f2fc8a157f033692f930f998 Mon Sep 17 00:00:00 2001 From: Laurent Aimar Date: Wed, 20 Jan 2010 18:54:51 +0000 Subject: [PATCH] H264 DXVA2 implementation It allows VLD H264 decoding using DXVA2 (GPU assisted decoding API under VISTA and Windows 7). It is implemented by using AVHWAccel API. It has been tested successfully for some time in VLC using an nvidia card on Windows 7. To compile it, you need to have the system header dxva2api.h (either from microsoft or using http://downloads.videolan.org/pub/videolan/testing/contrib/dxva2api.h) The generated libavcodec.dll does not depend directly on any new lib as the necessary objects are given by the application using FFmpeg. Originally committed as revision 21353 to svn://svn.ffmpeg.org/ffmpeg/trunk --- configure | 5 + libavcodec/Makefile | 3 +- libavcodec/allcodecs.c | 1 + libavcodec/dxva2.h | 68 +++++ libavcodec/dxva2_h264.c | 553 ++++++++++++++++++++++++++++++++++++++++ libavcodec/mpegvideo.c | 1 + libavutil/pixdesc.c | 6 + libavutil/pixfmt.h | 1 + 8 files changed, 637 insertions(+), 1 deletion(-) create mode 100644 libavcodec/dxva2.h create mode 100644 libavcodec/dxva2_h264.c diff --git a/configure b/configure index 0f136054a8..7a8fc2497a 100755 --- a/configure +++ b/configure @@ -107,6 +107,7 @@ Configuration options: --disable-rdft disable RDFT code --disable-vaapi disable VAAPI code --disable-vdpau disable VDPAU code + --disable-dxva2 disable DXVA2 code --enable-runtime-cpudetect detect cpu capabilities at runtime (bigger binary) --enable-hardcoded-tables use hardcoded tables instead of runtime generation --enable-memalign-hack emulate memalign, interferes with memory debuggers @@ -865,6 +866,7 @@ CONFIG_LIST=" bzlib dct doc + dxva2 fastdiv ffmpeg ffplay @@ -1174,6 +1176,8 @@ h263_vaapi_hwaccel_select="vaapi h263_decoder" h263i_decoder_select="h263_decoder" h263p_encoder_select="h263_encoder" h264_decoder_select="golomb" +h264_dxva2_hwaccel_deps="dxva2api_h" +h264_dxva2_hwaccel_select="dxva2 h264_decoder" h264_vaapi_hwaccel_deps="va_va_h" h264_vaapi_hwaccel_select="vaapi" h264_vdpau_decoder_deps="vdpau_vdpau_h vdpau_vdpau_x11_h" @@ -2399,6 +2403,7 @@ check_func_headers windows.h VirtualAlloc check_header conio.h check_header dlfcn.h +check_header dxva2api.h check_header malloc.h check_header poll.h check_header sys/mman.h diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 3888589511..24518479e8 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -3,7 +3,7 @@ include $(SUBDIR)../config.mak NAME = avcodec FFLIBS = avutil -HEADERS = avcodec.h opt.h vaapi.h vdpau.h xvmc.h +HEADERS = avcodec.h dxva2.h opt.h vaapi.h vdpau.h xvmc.h OBJS = allcodecs.o \ audioconvert.o \ @@ -135,6 +135,7 @@ OBJS-$(CONFIG_H263_ENCODER) += mpegvideo_enc.o mpeg4video.o mpeg4vide OBJS-$(CONFIG_H264_DECODER) += h264.o h264idct.o h264pred.o h264_loopfilter.o h264_direct.o cabac.o \ h264_sei.o h264_ps.o h264_refs.o h264_cavlc.o h264_cabac.o\ mpegvideo.o error_resilience.o +OBJS-$(CONFIG_H264_DXVA2_HWACCEL) += dxva2_h264.o OBJS-$(CONFIG_H264_ENCODER) += h264enc.o h264dspenc.o OBJS-$(CONFIG_H264_VAAPI_HWACCEL) += vaapi_h264.o OBJS-$(CONFIG_HUFFYUV_DECODER) += huffyuv.o diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index 43e49d1792..b775b04f77 100644 --- a/libavcodec/allcodecs.c +++ b/libavcodec/allcodecs.c @@ -55,6 +55,7 @@ void avcodec_register_all(void) /* hardware accelerators */ REGISTER_HWACCEL (H263_VAAPI, h263_vaapi); + REGISTER_HWACCEL (H264_DXVA2, h264_dxva2); REGISTER_HWACCEL (H264_VAAPI, h264_vaapi); REGISTER_HWACCEL (MPEG2_VAAPI, mpeg2_vaapi); REGISTER_HWACCEL (MPEG4_VAAPI, mpeg4_vaapi); diff --git a/libavcodec/dxva2.h b/libavcodec/dxva2.h new file mode 100644 index 0000000000..5c5fe21e2f --- /dev/null +++ b/libavcodec/dxva2.h @@ -0,0 +1,68 @@ +/* + * DXVA2 HW acceleration + * + * copyright (c) 2009 Laurent Aimar + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_DXVA_H +#define AVCODEC_DXVA_H + +#include + +#include + +/** + * This structure is used to provides the necessary configurations and data + * to the DXVA2 FFmpeg HWAccel implementation. + * + * The application must make it available as AVCodecContext.hwaccel_context. + */ +struct dxva_context { + /** + * DXVA2 decoder object + */ + IDirectXVideoDecoder *decoder; + + /** + * DXVA2 configuration used to create the decoder + */ + const DXVA2_ConfigPictureDecode *cfg; + + /** + * The number of surface in the surface array + */ + unsigned surface_count; + + /** + * The array of Direct3D surfaces used to create the decoder + */ + LPDIRECT3DSURFACE9 *surface; + + /** + * A bit field configuring the workarounds needed for using the decoder + */ + uint64_t workaround; + + /** + * Private to the FFmpeg AVHWAccel implementation + */ + unsigned report_id; +}; + +#endif /* AVCODEC_DXVA_H */ diff --git a/libavcodec/dxva2_h264.c b/libavcodec/dxva2_h264.c new file mode 100644 index 0000000000..9a832af392 --- /dev/null +++ b/libavcodec/dxva2_h264.c @@ -0,0 +1,553 @@ +/* + * DXVA2 H264 HW acceleration. + * + * copyright (c) 2009 Laurent Aimar + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "dxva2.h" +#include "avcodec.h" + +#include "mpegvideo.h" +#include "h264.h" +#include "h264data.h" + +struct dxva2_picture_context { + DXVA_PicParams_H264 pp; + DXVA_Qmatrix_H264 qm; + unsigned slice_count; + DXVA_Slice_H264_Short slice_short[MAX_SLICES]; + DXVA_Slice_H264_Long slice_long[MAX_SLICES]; + const uint8_t *bitstream; + unsigned bitstream_size; +}; + +static void *get_surface(const Picture *picture) +{ + return picture->data[3]; +} +static unsigned get_surface_index(const struct dxva_context *ctx, + const Picture *picture) +{ + void *surface = get_surface(picture); + unsigned i; + + for (i = 0; i < ctx->surface_count; i++) + if (ctx->surface[i] == surface) + return i; + + assert(0); + return 0; +} + +static void fill_picture_entry(DXVA_PicEntry_H264 *pic, + unsigned index, unsigned flag) +{ + assert((index&0x7f) == index && (flag&0x01) == flag); + pic->bPicEntry = index | (flag << 7); +} + +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; + int i; + + memset(pp, 0, sizeof(*pp)); + /* Configure current picture */ + fill_picture_entry(&pp->CurrPic, + get_surface_index(ctx, current_picture), + s->picture_structure == PICT_BOTTOM_FIELD); + /* Configure the set of references */ + pp->UsedForReferenceFlags = 0; + pp->NonExistingFrameFlags = 0; + for (i = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) { + if (i < h->short_ref_count + h->long_ref_count) { + const Picture *r; + if (i < h->short_ref_count) { + r = h->short_ref[i]; + assert(!r->long_ref); + } else { + r = h->long_ref[i - h->short_ref_count]; + assert(r->long_ref); + } + fill_picture_entry(&pp->RefFrameList[i], + get_surface_index(ctx, r), + r->long_ref != 0); + + if ((r->reference & PICT_TOP_FIELD) && r->field_poc[0] != INT_MAX) + pp->FieldOrderCntList[i][0] = r->field_poc[0]; + if ((r->reference & PICT_BOTTOM_FIELD) && r->field_poc[1] != INT_MAX) + pp->FieldOrderCntList[i][1] = r->field_poc[1]; + + pp->FrameNumList[i] = r->long_ref ? r->pic_id : r->frame_num; + if (r->reference & PICT_TOP_FIELD) + pp->UsedForReferenceFlags |= 1 << (2*i + 0); + if (r->reference & PICT_BOTTOM_FIELD) + pp->UsedForReferenceFlags |= 1 << (2*i + 1); + } else { + pp->RefFrameList[i].bPicEntry = 0xff; + pp->FieldOrderCntList[i][0] = 0; + pp->FieldOrderCntList[i][1] = 0; + pp->FrameNumList[i] = 0; + } + } + + pp->wFrameWidthInMbsMinus1 = s->mb_width - 1; + pp->wFrameHeightInMbsMinus1 = s->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) | + (h->sps.residual_color_transform_flag << 2) | + /* sp_for_switch_flag (not implemented by FFmpeg) */ + (0 << 3) | + (h->sps.chroma_format_idc << 4) | + ((h->nal_ref_idc != 0) << 6) | + (h->pps.constrained_intra_pred << 7) | + (h->pps.weighted_pred << 8) | + (h->pps.weighted_bipred_idc << 9) | + /* MbsConsecutiveFlag */ + (1 << 11) | + (h->sps.frame_mbs_only_flag << 12) | + (h->pps.transform_8x8_mode << 13) | + ((h->sps.level_idc >= 31) << 14) | + /* IntraPicFlag (Modified if we detect a non + * intra slice in 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 ? */ + pp->StatusReportFeedbackNumber = 1 + ctx->report_id++; + pp->CurrFieldOrderCnt[0] = 0; + if ((s->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) && + 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; + pp->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0]; + pp->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1]; + pp->ContinuationFlag = 1; + pp->pic_init_qp_minus26 = h->pps.init_qp - 26; + pp->num_ref_idx_l0_active_minus1 = h->pps.ref_count[0] - 1; + pp->num_ref_idx_l1_active_minus1 = h->pps.ref_count[1] - 1; + pp->Reserved8BitsA = 0; + pp->frame_num = h->frame_num; + pp->log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4; + pp->pic_order_cnt_type = h->sps.poc_type; + if (h->sps.poc_type == 0) + pp->log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4; + else if (h->sps.poc_type == 1) + pp->delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag; + pp->direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag; + pp->entropy_coding_mode_flag = h->pps.cabac; + pp->pic_order_present_flag = h->pps.pic_order_present; + pp->num_slice_groups_minus1 = h->pps.slice_group_count - 1; + pp->slice_group_map_type = h->pps.mb_slice_group_map_type; + 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 */ +} + +static void fill_scaling_lists(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]]; + + 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]]; +} + +static int is_slice_short(struct dxva_context *ctx) +{ + assert(ctx->cfg->ConfigBitstreamRaw == 1 || + ctx->cfg->ConfigBitstreamRaw == 2); + return ctx->cfg->ConfigBitstreamRaw == 2; +} + +static void fill_slice_short(DXVA_Slice_H264_Short *slice, + unsigned position, unsigned size) +{ + memset(slice, 0, sizeof(*slice)); + slice->BSNALunitDataLocation = position; + slice->SliceBytesInBuffer = size; + slice->wBadSliceChopping = 0; +} + +static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice, + unsigned position, unsigned size) +{ + H264Context *h = avctx->priv_data; /* FIXME Can't use const because of get_bits_count */ + struct dxva_context *ctx = avctx->hwaccel_context; + MpegEncContext *s = &h->s; + unsigned list; + + memset(slice, 0, sizeof(*slice)); + slice->BSNALunitDataLocation = position; + 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->NumMbsForSlice = 0; /* XXX it is set once we have all slices */ + slice->BitOffsetToSliceData = get_bits_count(&s->gb) + 8; + slice->slice_type = ff_h264_get_slice_type(h); + if (h->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; + slice->slice_beta_offset_div2 = h->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]; + unsigned plane; + fill_picture_entry(&slice->RefPicList[list][i], + get_surface_index(ctx, r), + 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[list][i]; + o = h->luma_offset[list][i]; + } else if (plane >= 1 && h->chroma_weight_flag[list]) { + w = h->chroma_weight[list][i][plane-1]; + o = h->chroma_offset[list][i][plane-1]; + } else { + w = 1 << (plane == 0 ? h->luma_log2_weight_denom : + h->chroma_log2_weight_denom); + o = 0; + } + slice->Weights[list][i][plane][0] = w; + slice->Weights[list][i][plane][1] = o; + } + } else { + unsigned plane; + slice->RefPicList[list][i].bPicEntry = 0xff; + for (plane = 0; plane < 3; plane++) { + slice->Weights[list][i][plane][0] = 0; + slice->Weights[list][i][plane][1] = 0; + } + } + } + } + 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; + else + slice->disable_deblocking_filter_idc = h->deblocking_filter; + slice->slice_id = h->current_slice - 1; +} + +static int commit_buffer(AVCodecContext *avctx, + struct dxva_context *ctx, + DXVA2_DecodeBufferDesc *dsc, + unsigned type, const void *data, unsigned size, + unsigned mb_count) +{ + void *dxva_data; + unsigned dxva_size; + int result; + + if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder, type, + &dxva_data, &dxva_size))) { + av_log(avctx, AV_LOG_ERROR, "Failed to get a buffer for %d\n", type); + return -1; + } + if (size <= dxva_size) { + memcpy(dxva_data, data, size); + + memset(dsc, 0, sizeof(*dsc)); + dsc->CompressedBufferType = type; + dsc->DataSize = size; + dsc->NumMBsInBuffer = mb_count; + + result = 0; + } else { + av_log(avctx, AV_LOG_ERROR, "Buffer for type %d was too small\n", type); + result = -1; + } + if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder, type))) { + av_log(avctx, AV_LOG_ERROR, "Failed to release buffer type %d\n", type); + result = -1; + } + return result; +} + +static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx, + struct dxva_context *ctx, + struct dxva2_picture_context *ctx_pic, + DXVA2_DecodeBufferDesc *bs, + DXVA2_DecodeBufferDesc *sc, + unsigned mb_count) +{ + DXVA_Slice_H264_Short *slice = NULL; + uint8_t *dxva_data, *current, *end; + unsigned dxva_size; + void *slice_data; + unsigned slice_size; + unsigned padding; + unsigned i; + + /* 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))) + return -1; + current = dxva_data; + end = dxva_data + dxva_size; + + for (i = 0; i < ctx_pic->slice_count; i++) { + static const uint8_t start_code[] = { 0, 0, 1 }; + static const unsigned start_code_size = sizeof(start_code); + unsigned position, size; + + assert(offsetof(DXVA_Slice_H264_Short, BSNALunitDataLocation) == + offsetof(DXVA_Slice_H264_Long, BSNALunitDataLocation)); + assert(offsetof(DXVA_Slice_H264_Short, SliceBytesInBuffer) == + offsetof(DXVA_Slice_H264_Long, SliceBytesInBuffer)); + + if (is_slice_short(ctx)) + slice = &ctx_pic->slice_short[i]; + else + slice = (DXVA_Slice_H264_Short*)&ctx_pic->slice_long[i]; + + position = slice->BSNALunitDataLocation; + size = slice->SliceBytesInBuffer; + if (start_code_size + size > end - current) { + av_log(avctx, AV_LOG_ERROR, "Failed to build bitstream"); + break; + } + + slice->BSNALunitDataLocation = current - dxva_data; + slice->SliceBytesInBuffer = start_code_size + size; + + if (!is_slice_short(ctx)) { + DXVA_Slice_H264_Long *slice_long = (DXVA_Slice_H264_Long*)slice; + if (i < ctx_pic->slice_count - 1) + slice_long->NumMbsForSlice = + slice_long[1].first_mb_in_slice - slice_long[0].first_mb_in_slice; + else + slice_long->NumMbsForSlice = mb_count - slice_long->first_mb_in_slice; + } + + memcpy(current, start_code, start_code_size); + current += start_code_size; + + memcpy(current, &ctx_pic->bitstream[position], size); + current += size; + } + padding = FFMIN(128 - ((current - dxva_data) & 127), end - current); + if (slice && padding > 0) { + memset(current, 0, padding); + current += padding; + + slice->SliceBytesInBuffer += padding; + } + if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(ctx->decoder, + DXVA2_BitStreamDateBufferType))) + return -1; + if (i < ctx_pic->slice_count) + return -1; + + memset(bs, 0, sizeof(*bs)); + bs->CompressedBufferType = DXVA2_BitStreamDateBufferType; + bs->DataSize = current - dxva_data; + bs->NumMBsInBuffer = mb_count; + + if (is_slice_short(ctx)) { + slice_data = ctx_pic->slice_short; + slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_short); + } else { + slice_data = ctx_pic->slice_long; + slice_size = ctx_pic->slice_count * sizeof(*ctx_pic->slice_long); + } + assert((bs->DataSize & 127) == 0); + return commit_buffer(avctx, ctx, sc, + DXVA2_SliceControlBufferType, + slice_data, slice_size, mb_count); +} + + +static int 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; + + if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0) + return -1; + assert(ctx_pic); + + /* Fill up DXVA_PicParams_H264 */ + fill_picture_parameters(ctx, h, &ctx_pic->pp); + + /* Fill up DXVA_Qmatrix_H264 */ + fill_scaling_lists(h, &ctx_pic->qm); + + ctx_pic->slice_count = 0; + ctx_pic->bitstream_size = 0; + ctx_pic->bitstream = NULL; + return 0; +} + +static int decode_slice(AVCodecContext *avctx, + const uint8_t *buffer, uint32_t size) +{ + H264Context *h = avctx->priv_data; /* FIXME Can't use const because of get_bits_count */ + struct dxva_context *ctx = avctx->hwaccel_context; + const Picture *current_picture = h->s.current_picture_ptr; + struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private; + unsigned position; + + if (ctx_pic->slice_count >= MAX_SLICES) + return -1; + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + position = buffer - ctx_pic->bitstream; + if (is_slice_short(ctx)) + fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count], + position, size); + else + fill_slice_long(avctx, &ctx_pic->slice_long[ctx_pic->slice_count], + position, size); + ctx_pic->slice_count++; + + if (h->slice_type != FF_I_TYPE && h->slice_type != FF_SI_TYPE) + ctx_pic->pp.wBitFields &= ~(1 << 15); /* Set IntraPicFlag to 0 */ + return 0; +} + +static int end_frame(AVCodecContext *avctx) +{ + H264Context *h = avctx->priv_data; + MpegEncContext *s = &h->s; + const unsigned mb_count = s->mb_width * s->mb_height; + struct dxva_context *ctx = avctx->hwaccel_context; + const Picture *current_picture = h->s.current_picture_ptr; + struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private; + unsigned buffer_count = 0; + DXVA2_DecodeBufferDesc buffer[4]; + DXVA2_DecodeExecuteParams exec; + int result; + + if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + if (FAILED(IDirectXVideoDecoder_BeginFrame(ctx->decoder, + get_surface(current_picture), + NULL))) { + av_log(avctx, AV_LOG_ERROR, "Failed to begin frame\n"); + return -1; + } + + result = commit_buffer(avctx, ctx, &buffer[buffer_count], + DXVA2_PictureParametersBufferType, + &ctx_pic->pp, sizeof(ctx_pic->pp), 0); + if (result) { + av_log(avctx, AV_LOG_ERROR, + "Failed to add picture parameter buffer\n"); + goto end; + } + buffer_count++; + + result = commit_buffer(avctx, ctx, &buffer[buffer_count], + DXVA2_InverseQuantizationMatrixBufferType, + &ctx_pic->qm, sizeof(ctx_pic->qm), 0); + if (result) { + av_log(avctx, AV_LOG_ERROR, + "Failed to add inverse quantization matrix buffer\n"); + goto end; + } + buffer_count++; + + result = commit_bitstream_and_slice_buffer(avctx, ctx, ctx_pic, + &buffer[buffer_count + 0], + &buffer[buffer_count + 1], + mb_count); + if (result) { + av_log(avctx, AV_LOG_ERROR, + "Failed to add bitstream or slice control buffer\n"); + goto end; + } + buffer_count += 2; + + /* TODO Film Grain when possible */ + + assert(buffer_count == 4); + + memset(&exec, 0, sizeof(exec)); + exec.NumCompBuffers = buffer_count; + exec.pCompressedBuffers = buffer; + exec.pExtensionData = NULL; + if (FAILED(IDirectXVideoDecoder_Execute(ctx->decoder, &exec))) { + av_log(avctx, AV_LOG_ERROR, "Failed to execute\n"); + result = -1; + } + +end: + if (FAILED(IDirectXVideoDecoder_EndFrame(ctx->decoder, NULL))) { + av_log(avctx, AV_LOG_ERROR, "Failed to end frame\n"); + result = -1; + } + + if (!result) + ff_draw_horiz_band(s, 0, s->avctx->height); + return result; +} + +AVHWAccel h264_dxva2_hwaccel = { + .name = "h264_dxva2", + .type = CODEC_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), +}; + diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c index 7f77d00d57..0ad5718c69 100644 --- a/libavcodec/mpegvideo.c +++ b/libavcodec/mpegvideo.c @@ -81,6 +81,7 @@ const enum PixelFormat ff_pixfmt_list_420[] = { }; const enum PixelFormat ff_hwaccel_pixfmt_list_420[] = { + PIX_FMT_DXVA2_VLD, PIX_FMT_VAAPI_VLD, PIX_FMT_YUV420P, PIX_FMT_NONE diff --git a/libavutil/pixdesc.c b/libavutil/pixdesc.c index b689519fde..972da4578f 100644 --- a/libavutil/pixdesc.c +++ b/libavutil/pixdesc.c @@ -646,6 +646,12 @@ const AVPixFmtDescriptor av_pix_fmt_descriptors[PIX_FMT_NB] = { }, .flags = PIX_FMT_BE, }, + [PIX_FMT_DXVA2_VLD] = { + .name = "dxva2_vld", + .log2_chroma_w = 1, + .log2_chroma_h = 1, + .flags = PIX_FMT_HWACCEL, + }, }; int av_get_bits_per_pixel(const AVPixFmtDescriptor *pixdesc) diff --git a/libavutil/pixfmt.h b/libavutil/pixfmt.h index a18da18af4..6eddde02bb 100644 --- a/libavutil/pixfmt.h +++ b/libavutil/pixfmt.h @@ -126,6 +126,7 @@ enum PixelFormat { PIX_FMT_YUV444P16LE, ///< planar YUV 4:4:4, 48bpp, (1 Cr & Cb sample per 1x1 Y samples), little-endian PIX_FMT_YUV444P16BE, ///< planar YUV 4:4:4, 48bpp, (1 Cr & Cb sample per 1x1 Y samples), big-endian PIX_FMT_VDPAU_MPEG4, ///< MPEG4 HW decoding with VDPAU, data[0] contains a vdpau_render_state struct which contains the bitstream of the slices as well as various fields extracted from headers + PIX_FMT_DXVA2_VLD, ///< HW decoding through DXVA2, Picture.data[3] contains a LPDIRECT3DSURFACE9 pointer PIX_FMT_NB, ///< number of pixel formats, DO NOT USE THIS if you want to link with shared libav* because the number of formats might differ between versions };