[FFmpeg-devel] [PATCH 3/3] avutil/hwcontext_qsv: fix D3D11VA<->qsv hwmap errors

[Xiang, Haihao]

On Fri, 2022-04-01 at 17:24 +0800, Tong Wu wrote:
> For hwmap between qsv and d3d11va, The mfxHDLPair information should be
> put into texture_infos when deriving from qsv context. Moreover, when
> uploading from rawvideo, the ways that the textures are created are
> different, bindflag assertions are needed to make sure the right
> textures are derived during the process. Now after this fix,
> d3d_dec->qsv_vpp->qsv_enc, d3d_dec->qsv_vpp->qsv_download->yuv,
> yuv->d3d_upload->qsv_vpp->qsv->download->yuv,
> qsv_dec->qsv_vpp->d3d_download->yuv can all work properly.
> 
> For d3d_dec->qsv_vpp->qsv_enc, one sample command line:
> ffmpeg.exe -hwaccel qsv -c:v h264_qsv -i input.264
> -vf
> "hwmap=derive_device=d3d11va,format=d3d11,hwmap=derive_device=qsv,format=qsv"
> -c:v h264_qsv -y ./output.264


The default child_device_type is dxva2 for option --enable-libmfx, I don't think
it makes sense to derive a QSV device based on dxva2 child device to a d3d11vadevice. 

But even if initializing qsv device with d3d11va child device, the command below
still doesn't work

$ ffmpeg.exe -y -hwaccel qsv -init_hw_device
qsv=qsv:hw,child_device=0,child_device_type=d3d11va -c:v h264_qsv -i input.h264
-vf "hwmap=derive_device=d3d11va,format=d3d11,hwdownload,format=nv12" -f null -

You may try https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=5304 

Thanks
Haihao


> 
> Signed-off-by: Tong Wu <tong1.wu at intel.com>
> ---
>  libavutil/hwcontext_qsv.c | 48 ++++++++++++++++++++++++++++++++-------
>  1 file changed, 40 insertions(+), 8 deletions(-)
> 
> diff --git a/libavutil/hwcontext_qsv.c b/libavutil/hwcontext_qsv.c
> index 95f8071abe..e6a7ac3ef0 100644
> --- a/libavutil/hwcontext_qsv.c
> +++ b/libavutil/hwcontext_qsv.c
> @@ -806,12 +806,23 @@ static int qsv_frames_derive_from(AVHWFramesContext
> *dst_ctx,
>  #if CONFIG_D3D11VA
>      case AV_HWDEVICE_TYPE_D3D11VA:
>          {
> +            dst_ctx->initial_pool_size = src_ctx->initial_pool_size;
>              AVD3D11VAFramesContext *dst_hwctx = dst_ctx->hwctx;
> -            mfxHDLPair *pair = (mfxHDLPair*)src_hwctx-
> >surfaces[i].Data.MemId;
> -            dst_hwctx->texture = (ID3D11Texture2D*)pair->first;
> +            dst_hwctx->texture_infos = av_calloc(src_hwctx->nb_surfaces,
> +                                                 sizeof(*dst_hwctx-
> >texture_infos));
>              if (src_hwctx->frame_type & MFX_MEMTYPE_SHARED_RESOURCE)
>                  dst_hwctx->MiscFlags = D3D11_RESOURCE_MISC_SHARED;
>              dst_hwctx->BindFlags = qsv_get_d3d11va_bind_flags(src_hwctx-
> >frame_type);
> +            for (i = 0; i < src_hwctx->nb_surfaces; i++) {
> +                mfxHDLPair* pair = (mfxHDLPair*)src_hwctx-
> >surfaces[i].Data.MemId;
> +                dst_hwctx->texture_infos[i].texture = (ID3D11Texture2D*)pair-
> >first;
> +                if (dst_hwctx->BindFlags & D3D11_BIND_RENDER_TARGET) {
> +                    dst_hwctx->texture_infos[i].index = 0;
> +                }
> +                else {
> +                    dst_hwctx->texture_infos[i].index = (intptr_t)pair-
> >second;
> +                }
> +            }

>          }
>          break;
>  #endif
> @@ -900,9 +911,16 @@ static int qsv_map_from(AVHWFramesContext *ctx,
>          dst->height  = src->height;
>  
>         if (child_frames_ctx->device_ctx->type == AV_HWDEVICE_TYPE_D3D11VA) {
> +#if CONFIG_D3D11VA
> +            AVD3D11VAFramesContext* child_frames_hwctx = child_frames_ctx-
> >hwctx;
>              mfxHDLPair *pair = (mfxHDLPair*)surf->Data.MemId;
>              dst->data[0] = pair->first;
> -            dst->data[1] = pair->second;
> +            if (child_frames_hwctx->BindFlags & D3D11_BIND_RENDER_TARGET) {
> +                dst->data[1] = 0;
> +            } else {
> +                dst->data[1] = pair->second;
> +            }
> +#endif
>          } else {
>              dst->data[3] = child_data;
>          }
> @@ -930,9 +948,16 @@ static int qsv_map_from(AVHWFramesContext *ctx,
>      dummy->height        = src->height;
>  
>      if (child_frames_ctx->device_ctx->type == AV_HWDEVICE_TYPE_D3D11VA) {
> +#if CONFIG_D3D11VA
> +        AVD3D11VAFramesContext* child_frames_hwctx = child_frames_ctx->hwctx;
>          mfxHDLPair *pair = (mfxHDLPair*)surf->Data.MemId;
>          dummy->data[0] = pair->first;
> -        dummy->data[1] = pair->second;
> +        if (child_frames_hwctx->BindFlags & D3D11_BIND_RENDER_TARGET) {
> +            dst->data[1] = 0;
> +        } else {
> +            dst->data[1] = pair->second;
> +        }
> +#endif
>      } else {
>          dummy->data[3] = child_data;
>      }
> @@ -1287,6 +1312,10 @@ static int qsv_frames_derive_to(AVHWFramesContext
> *dst_ctx,
>          return AVERROR(ENOSYS);
>      }
>  
> +    s->child_frames_ref = av_buffer_ref(dst_ctx->internal->source_frames);
> +    if (!s->child_frames_ref) {
> +        return AVERROR(ENOMEM);
> +    }
>      dst_hwctx->surfaces = s->surfaces_internal;
>  
>      return 0;
> @@ -1314,10 +1343,13 @@ static int qsv_map_to(AVHWFramesContext *dst_ctx,
>          case AV_PIX_FMT_D3D11:
>          {
>              mfxHDLPair *pair = (mfxHDLPair*)hwctx->surfaces[i].Data.MemId;
> -            if (pair->first == src->data[0]
> -                && pair->second == src->data[1]) {
> -                index = i;
> -                break;
> +            if (pair->first == src->data[0]) {
> +                if (hwctx->frame_type &
> MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET
> +                    && pair->second == src->data[1]
> +                    || hwctx->frame_type &
> MFX_MEMTYPE_VIDEO_MEMORY_PROCESSOR_TARGET) {
> +                    index = i;
> +                    break;
> +                }
>              }
>          }
>  #endif