- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello.
I am writing a driver for the Intel Arc A750 GPU for my own operating system.
I am already able to execute some opencl kernels, but when I try to execute a kernel that reads an untyped surface, the GPU hangs as soon as the surface access happens. It also happens with no access when I set the number of binding table entries to prefetch in the Interface Descriptor to 1.
Things I have already verified:
- I know that my Binding table and part of the surface state works, as I get a pagefault when I don't map the surface and don't get a pagefault when I do map the surface.
- I've also verified that the surface is correctly mapped by writing to it via an bindless scattered A32 write.
- Surface Type is BUFFER, surface format is RAW, AUX is None, MOCS is 6, pitch is 0, which should be correct according to the PRM
- The size should not be a problem, because a) it fails even with just prefetching the surface state and b) I've tested it with an absurdly high size and it still failed.
- All base addresses and sizes are properly aligned
Does anybody has any idea on what the reason for the hang could be and how I could resolve it?
I would be thankfull for all explaination about hangs in such a scenario or a valid surface state for untyped surface reads with x.
With kind regards
Joel Marker
Link Copied
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page