Skip to content

Conversation

@christiangnrd
Copy link
Member

No description provided.

These always run on the CPU backend which is currently broken in 1.12, creating false negatives for the GPU tests
@christiangnrd
Copy link
Member Author

christiangnrd commented Dec 18, 2025

id for AMDGPU is used to identify whether we need to launch hostcalls, based on the specific id of the global variable. It is not needed for KA, so can be dropped as in https://github.com/JuliaGPU/AMDGPU.jl/pull/868/files#diff-082b94339c8f038178ee472ca9b6feec6f27f434c469138f168031f248f223f9R197.

(From JuliaGPU/KernelAbstractions.jl#666)

Is my changing static local code generation from LLVMExternalLinkage to LLVMInternalLinkage going to cause any unintended effects? I had to do so to prevent local memory allocations from being treated as the same one if there happen to be more than one in a kernel. Otherwise I think we’ll have to readd the id parameter.

@pxl-th
Copy link
Member

pxl-th commented Dec 19, 2025

Is my changing static local code generation from LLVMExternalLinkage to LLVMInternalLinkage going to cause any unintended effects?

I don't think. I think it is a remnant from pre-HIP times.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants