[chore] cuda-13.1.1 support

[ClSlaid]

• update cudarc to version 19 to support cuda-13.1.1

[Copilot]

Pull request overview

This PR updates the cudarc dependency from version 0.18.0 to 0.19 to enable CUDA 13.1.1 support. The update switches from using a specific CUDA version feature (cuda-12040) to a dynamic build-time version detection feature (cuda-version-from-build-system), allowing the library to work with whichever CUDA version is available on the build system.

Changes:

• Updated cudarc from 0.18.0 to 0.19 with dynamic CUDA version detection
• Updated transitive dependency libloading from 0.8.9 to 0.9.0

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated 1 comment.

File	Description
Cargo.toml	Updated cudarc version to 0.19 and changed feature from cuda-12040 to cuda-version-from-build-system
Cargo.lock	Reflected dependency version updates for cudarc (0.19.1) and libloading (0.9.0)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The PR description claims to add "cuda-13.1.1 support" by updating to cudarc 0.19, but the CI workflows (.github/workflows/ci.yml line 151 and .github/workflows/release.yml line 9) still use CUDA 12.8.1.

The new feature cuda-version-from-build-system will detect the CUDA version at build time from the system where the build occurs. This means:

1. Local builds will use whatever CUDA version is installed locally
2. CI/release builds will still use CUDA 12.8.1 (not 13.1.1 as the PR title suggests)

To truly support CUDA 13.1.1 as stated in the PR description, the CI workflow files should also be updated to install CUDA 13.1.1 instead of 12.8.1. Alternatively, if the intent is to support multiple CUDA versions dynamically, the PR description should be clarified to reflect that this change enables flexible CUDA version support rather than specifically targeting 13.1.1.

[xiaguan]

It seems the CI doesn't have nvcc available. Perhaps we should mention this in the README.