[BUG] accelerate-llvm-ptx won't compile

[noahmartinwilliams]

Hi! Thanks for trying out Accelerate! We are sorry that you ran into trouble ): Please fill in the following report to help us fix the problem.

Description
When running "stack build" on a (slightly modified) version of accelerate-llvm the compiler complains about a missing function (called "clockRate").

Steps to reproduce
Attached are tarballs containing the slightly modified code for accelerate-llvm and its dependencies (modifications were done to try and make it closer to a successful compile).

accelerate-llvm.tar.gz

c2hs.tar.gz

cuda.tar.gz

hashtables.tar.gz

Steps to reproduce the behaviour:

1. Unzip the attached folders all in the same directory.
2. go to accelerate-llvm directory
3. run "stack build" and it should fail with an error message about a missing function.

Expected behaviour
It should compile.

Your environment

• Accelerate:
• Accelerate backend(s):
• GHC: 9.14.1
• OS: Arch Linux up to date as of February 21st, 2026.

Additional context
CUDA version 13.1

[tomsmeding]

Thank you for the missing language extensions for cuda; I've pushed them on the PR.

What you're running into here is that my cuda-13 PR to cuda introduces a breaking change, for annoying technical reasons that can be read in the PR, and Accelerate has not been patched to support that breaking change. There are various ways to resolve this, but the most desirable outcome (it seems, for now) is that cuda is restructured to remove support for the CUDA runtime bindings, becoming a driver-bindings-only package; this allows CUDA 13 support to be added without breaking changes other than, of course, dropping the runtime API. To live in that world already, you can update your cuda package commit from my cuda-13 branch to my runtime-split branch. With that branch, Accelerate should work fine.

Apologies for the mess.