Hello .I have been experimenting with the recent Windows port of RADV, and as I noticed by looking at the code, the current experimental implementation is kind of hard-coded for gfx11. On my Radeon 890M (gfx1150) I couldn't get SashaWillem's triangle example running as radv_wddm2_bo_from_handle() doesn't produce correct results for shared handles sourced from DXGI's swapchian images, giving pitch_bytes of zero, and then image view creation fails or this image.
I assume this means that AMD has changed the internal pPrivateData structures and they have to be reserve-engineered again. I am willing to do that, but I would really like to know if it would be possible for someone here to contact AMD's Windows kernel team and ask them if they are willing to assist by providing a stable way to communicate with their Windows kernel driver. These exact issues have been addressed in Faith's XDC 2024 talk and the recent Collabora post, but there wasn't any information about actually asking AMD for their opinion. If this isn't the proper place to ask, I would be grateful for any pointers, since I'm not very well-connected in the driver community. P.S. There is also a bug in https://gitlab.freedesktop.org/lfrb/mesa/-/blob/wddm2/src/amd/vulkan/winsys/wddm2/radv_wddm2_bo.c?ref_type=heads#L675 where because of making one allocation instead of four separate ones the structs in the private runtime data overlap with the allocation info structs, which causes the hAllocation handle to be invalid, but after fixing that and using the (seemingly) correct hAllocation handle I still couldn't get any further because of the pPrivateData issues.
