Branch: refs/heads/main
Revision: ae2e545
Author: e-kayrakli
Link: Add an extern declaration for chpl_nodeID for GPU by e-kayrakli · Pull Request #19149 · chapel-lang/chapel · GitHub
Log Message:
Merge pull request #19149 from e-kayrakli/chpl-nodeid-gpu
Add an extern declaration for chpl_nodeID for GPU
This adds an extern __constant__ for chpl_nodeID in the GPU part of the
runtime.
In summary, we need the GPU "version" of that constant so that the device
functions have a symbol to access to. Before switching to LLVM 12, we weren't
bumping into this issue because we were adding an extern definition in the
generated code. But with LLVM 12, we somehow stopped doing that.
This is definitely not the solution we need for this in the long term, and the
issue has already been captured in
https://github.com/Cray/chapel-private/issues/2758
[Trivial, fix for ongoing nightly failures]
Modified Files:
M runtime/include/gpu/cuda/chpl-gpu-gen-includes.h
Compare: https://github.com/chapel-lang/chapel/compare/d3f3a55c2964...ae2e545127c5