Windows vulkan acceleration structure problem

[NeroReflex]

Developing a simple engine with vulkan I notice a problem: I cannot create TLAS with more than 7 BLAS with the windows driver.

The issue is well visible using amd rra tool: only 7 out of 392 meshes are loaded.

The but manifests on windows only, with a 9000 card and a 6000 card.

Same PC with linux the same works flawlessly, as it does with an nvidia card in moth linux and windows.

I managed to reproduce the problem also editing some vulkan example in github: loading more than 7 BLAS is impossibile in windows.

[pperucki-AMD]

We appreciate the report. If you can share a minimal reproducible sample code, it will greatly help us in addressing the issue more quickly.
Additionally, please include the exact GPU cards and driver versions in your report. Have you had a chance to check if the issue persists on the latest drivers?

[pperucki-AMD]

Could you please share a minimal reproducible sample code?

[NeroReflex]

Sorry, I have completely forgotten about this. I might have something that is not minimal, but shows the issue well. Are you up for it? The only problem is the resource file that is several MB: how would you want it to be provided?

[pperucki-AMD]

Are you able to upload the source code and any dependencies to some cloud provider like Google Drive, One Drive or other and share the link here?

[NeroReflex]

Sure, while I upload the resource file the source is on folder engine of this: https://github.com/neroreflex/vulkan-framework as I said this works perfectly on nvidia, and amd on linux. With intel (xe iGPU) is very slow but the model gets loaded correctly. Perhaps the problem might be tied to PREFER_FAST_TRACE ?

[NeroReflex]

https://drive.google.com/file/d/1HlSsiAkLEa5PNmIWvQ8dHEWteU_EGdkw/view?usp=sharing put this into the working directory of the executable. I can provide a prebuilt one if you want, but the secret is just compiling SDL2.lib, and rename the static version into SDL2.lib and copy it to the linker directory...