nvbench
nvbench copied to clipboard
NVIDIA
Add multi-cuda wheel build
This PR introduces "multi-CUDA" wheels builds, exactly similar to the ones we build for CCCL, as described here.
Instead of shipping multiple packages (pynvbench-cu12, pynvbench-cu13); we ship a single one (pynvbench) that contains both CUDA 12 and 13 binaries (bindings). The resulting wheel size is just ~2.2MB, so this is not a problem. At runtime, depending on the available CUDA version, the appropriate binary is loaded.
At install time, we still need the user to specify which CUDA major version to install for - because the version of required dependencies (nvidia-cuda-cupti-cu12) depend on the CUDA major version. This specification of the CUDA version is done using "extras":
pip install pynvbench[cu12] # or cu13
![copy-pr-bot[bot] avatar](https://avatars.githubusercontent.com/in/354886?v=4 "Published by a pub.dev verified publisher"){kind=small}