Skip to content

Conversation

@theonewolf
Copy link

This commit checks at compile time if we are using CUDA 13 headers and will link to CUDA 13, at which point we need to use a different function signature for cuCreateCtx as it changed in CUDA 13 (it added a parameter).

For CUDA versions below 13, we continue using the old function signature (a 3 parameter version).

This compiles and runs on my CUDA 13 system. I have not directly tested this with CUDA 12 and below yet.

@anomixer
Copy link

anomixer commented Aug 18, 2025

Sorry! My typo.
Using your repo and "make COMPUTE=100" works. Thanks!

@karthikeyan1729
Copy link

Since the new argument is a pointer, will just passing NULL work?

checkError(cuCtxCreate(&d_ctx, NULL, 0, d_dev));

The CUDA toolkit doc does not clarify this.

@karthikeyan1729
Copy link

#if CUDA_VERSION >= 13000

But the Makefile does not define CUDA_VERSION as an integer. It has

CUDA_VERSION ?= 11.8.0

@karthikeyan1729
Copy link

Please ignore my comment about CUDA_VERSION. It comes from cuda.h and not from Makefile.

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.

3 participants