search

rapidsai / raft

RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.
https://docs.rapids.ai/api/raft/stable/
Apache License 2.0
683 stars 180 forks source link

add gpuarch arg to allow for setting specific gpu architectures #2301

Closed jperez999 closed 2 months ago

jperez999 commented 2 months ago

This PR aims to add a new argument to the build script. --gpuarch gives the user the ability to set the desired build architectures. Currently there are only 2 options available to build script users, NATIVE and RAPIDS. However, NATIVE builds can fail if users have multiple GPUs that are not all the same architecture on their systems. In order to prevent this causing a failure, I added this argument that will override and set RAFT_CMAKE_CUDA_ARCHITECTURES to the values set by the user.

Examples of usage: --gpuarch="70;89"

Values are expected to be integers, separated by semicolons ;.

copy-pr-bot[bot] commented 2 months ago

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

robertmaynard commented 2 months ago

Please file an issue against https://github.com/rapidsai/rapids-cmake that NATIVE doesn't work properly with multiple GPU output, since it is designed to do so.

This change isn't needed once we correct NATIVE to work as intended.

jperez999 commented 2 months ago

I am closing in lieu of a ticket filed against rapids-cmake directly https://github.com/rapidsai/rapids-cmake/issues/605 as @robertmaynard suggested.