NVIDIA / nvbench

CUDA Kernel Benchmarking Library
Apache License 2.0
510 stars 64 forks source link

Allow nvbench to use an exiting fmt when brought in via find_package #125

Closed robertmaynard closed 1 year ago

robertmaynard commented 1 year ago

LGTM, @robertmaynard feel free to merge after addressing @bdice's question.

Are you okay moving to rapids-cmake 23.06 to get this feature? It would mean moving your minimum CMake version to 3.23.1 compared to the existing 3.20

bdice commented 1 year ago

@robertmaynard Can we finalize this PR?

robertmaynard commented 1 year ago

@robertmaynard Can we finalize this PR?

There is still open questions around the update to using rapids_cpm_fmt requiring nvbench to move to a newer rapids-cmake version, which would bump the CMake minimum version. If we are all fine with dropping that request ( or doing the bump ) I am happy to merge this.

bdice commented 1 year ago

I'd support making this update by bumping the CMake minimum. @allisonvacanti Do you have any reservations about that?

If there are issues with that, let's merge as-is without using rapids_cpm_fmt. This change is in a patch downstream (in cudf) that would be nice to drop, one way or another.

robertmaynard commented 1 year ago

Closing as these changes are now part of https://github.com/NVIDIA/nvbench/pull/140