On 25/05/2024 16:18, Sean McGovern wrote: > Hi, > > > On Sat, May 25, 2024, 09:07 Lynne via ffmpeg-devel > wrote: > >> On 25/05/2024 12:30, tong1.wu-at-intel.com@ffmpeg.org wrote: >>> From: Tong Wu >>> >>> Related parameters such as device context, frame context are also moved >>> to base layer. >>> >>> Signed-off-by: Tong Wu >>> --- >>> libavcodec/hw_base_encode.c | 49 ++++++++++++++++++ >>> libavcodec/hw_base_encode.h | 17 +++++++ >>> libavcodec/vaapi_encode.c | 90 +++++++++++---------------------- >>> libavcodec/vaapi_encode.h | 10 ---- >>> libavcodec/vaapi_encode_av1.c | 2 +- >>> libavcodec/vaapi_encode_h264.c | 2 +- >>> libavcodec/vaapi_encode_h265.c | 2 +- >>> libavcodec/vaapi_encode_mjpeg.c | 6 ++- >>> 8 files changed, 102 insertions(+), 76 deletions(-) >>> >>> diff --git a/libavcodec/hw_base_encode.c b/libavcodec/hw_base_encode.c >>> index 16afaa37be..c4789380b6 100644 >>> --- a/libavcodec/hw_base_encode.c >>> +++ b/libavcodec/hw_base_encode.c >>> @@ -592,3 +592,52 @@ end: >>> >>> return 0; >>> } >>> + >>> +int ff_hw_base_encode_init(AVCodecContext *avctx) >>> +{ >>> + FFHWBaseEncodeContext *ctx = avctx->priv_data; >> >> This is the issue I was talking about, this requires that >> FFHWBaseEncodeContext is always the main context. >> >> Could you change it so everything takes FFHWBaseEncodeContext as an >> argument, rather than AVCodecContext (apart from where the function >> absolutely must read some data from it)? >> > > Might this suggestion involve having to do some ugly down-casting? Not at all. Instead of having this context as the main context, just make each encoder have its own context, where this structure is the only element. Vulkan requires extra state, such as an execution context, function pointers from the loader, memory the encoder needs, which, rather than adding to this structure, would be better off being in its own context.