Skip to content

Obtain robust CUDA_ARCHITECTURES list when use of all archs is specified #269

@mattgara

Description

@mattgara

Description

While testing builds, I noticed that some architectures were missing when targeting specific builds.

There is a way to retrieve all supported architectures directly from nvcc within the build Docker container. To maintain consistency and avoid similar issues in the future, we should adopt that same approach across all our build scripts.

This issue is meant to track that improvement.

Action items

  • Identify how to extract supported architectures from nvcc within the build container.
  • Apply that method uniformly across all build scripts (including Presto, Gluten & Velox).
  • Verify builds for all expected targets.

Priority

Low (non-blocking)

Context

Follow-up to recent PR affecting build scripts. See original comment here: #261 (comment)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions