Open mobin-2008 opened 2 years ago
I am ready for any help and information
@Venemo That's one for you I guess ^^
To debug this issue, what would be helpful is either a trace of the game taken with wined3d (see wiki for how to make a trace), or the log with NINE_TGSI_DUMP=1 (needs mesa built with ndebug set to false, or debug build)
Thanks for the bug report. I think the relevant part is this:
tgsi_to_nir: unhandled TGSI property 23 = 1
I'll look into it
I cannot replicate the issue when running the trace.
I was doing something for a while and I did not pursue this issue. I'm sorry. I do not understand! I recorded the game performance with wined3d9 and sent it to you. Where was my work wrong?
What a coincidence. I ran today into the exact same tgsi_to_nir: unhandled TGSI property 23 = 1
problem described here.
I tried to test the new nir-to-tgsi path of the r600 Mesa driver (via R600_DEBUG=nir
) in conjunction with the Xbox "emulator" CXBX-R and Gallium Nine.
Long story short, it looks that the underlying problem here has to do with the incompleteness of tgsi-to-nir, please see "ttn doesn't translate const decls into ubo vars, so then ntt doesn't make a const decl" in Mesa MR 14883.
Mesa developer Emma Anholt suggest therefore, I quote:
wine nine is basically the only user of tgsi_to_nir (unless you're on a nir-backend-only driver and using it to take in u_blitter shaders). I think we need to just make nine stop using ttn on preferred-nir drivers, given that ttn is not complete.
More information can be found at the corresponding WIP: r600: Use tgsi-to-nir instead of TGSI when the NIR debug opt is disabled MR on Mesa.
@axeldavy (Really sorry to bother you again) and @Venemo (Because you are here the "owner" of the bug)
@mobin-2008 You haven't done here anything wrong. It's just so that the problem must be reproducible by the devs, otherwise it is almost impossible to fix it. :wink:
Whatever, if you have the time and motivation then it would be interesting to know if you are able to reproduce this tgsi_to_nir: unhandled TGSI property 23 = 1
error on your nvc0 nouveau driver based hardware with one of my Crash Tag Team Racing apitraces from #5966 or #5967.
Have you tried enforcing tgsi then ?
Yes, Gallium Nine works on the r600 driver perfectly well with TGSI. This tgsi_to_nir: unhandled TGSI property 23 = 1
problem here is only present for me when I enforce NIR with R600_DEBUG=nir
.
PS It is somehow interesting and strange why we have no such problem with the RadeonSI driver.
https://gitlab.freedesktop.org/anholt/mesa/-/commit/8392711ef663139b5cc46ce48f8dcb3a461c6b51
I think this problem has been completely solved by upgrading this issue in the upstream project (mesa) when version 22.2 is released. In this case, I close this issue.
Yes, there was a fix (or workaround) in Mesa MR !14883 which landed recently.
However, according to the comment here of Timur Kristóf it looks that the tgsi_to_nir: unhandled TGSI property 23 = 1
problem will be resolved finally when some additional D3D9 functionality is added to TTN. :wink:
Addition: For whatever reason I still get this tgsi_to_nir: unhandled TGSI property 23 = 1
bug with the r600 driver.
Related issue on Freedesktop's gitlab:#5678
Hello Im cannot run some games with d3d9-nine for example : Call of Duty Modern Warfare 2 : log of wine
system info AMD Ryzen 5 3500x NVIDIA Geforce GTX 1650 (with nouveau & mesa) 2x8GB ram, DDR4, 3200,CL16