Skip to content

Conversation

@castigli
Copy link
Contributor

@castigli castigli commented Dec 2, 2025

Extending CoralGemm test for BF16.

@jgphpc
Copy link
Collaborator

jgphpc commented Dec 4, 2025

cscs-ci run alps-beverin-uenv;MY_UENV=prgenv-gnu/25.07-6.3.3:v10

@jgphpc
Copy link
Collaborator

jgphpc commented Dec 4, 2025

cscs-ci run alps-santis-uenv;MY_UENV=prgenv-gnu/25.11:v1

Copilot finished reviewing on behalf of jgphpc December 4, 2025 08:20
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

This PR extends the CoralGemm GPU benchmark test to support BF16 (brain floating point 16-bit) precision in addition to the existing FP32 and FP64 support. BF16 is a reduced-precision format commonly used in AI/ML workloads to achieve higher throughput while maintaining sufficient numerical accuracy.

Key changes:

  • Added 'R_16B' (BF16) to the supported precision types
  • Configured BF16 to use FP32 for output matrix and computation precision (mixed-precision GEMM)
  • Added performance reference values for BF16 across three GPU architectures (mi200, mi300, gh200)

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

@jgphpc jgphpc merged commit e18c3a8 into eth-cscs:main Dec 4, 2025
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants