Telcon: 2023 03 15

Held Wednesday March 15th, 9am PT

(Note: daylight saving time is in effect: this is now UTC-7 rather than UTC-8)

Attendees

• Peter Scheibel (host)
• Jack Morrison
• Massimiliano Culpo
• Mark Krentel
• Wileam Phan
• Dom Heinzeller

Agenda

This meeting will be Q&A: there are no pre-planned general Spack topics.

• Jack: https://github.com/spack/spack/pull/34991
  • Installation with multiple compilers/targets
  • spack load is reporting an ambiguity when requesting a package since it is built multiple times
  • Using a tool called reframe to generate Spack environments "on the fly"
  • https://spackpm.slack.com/archives/CMHK7MF51/p1678744085160509
  • Higher-level question: how to manage compilers that are typically only used as part of an environment?
    • e.g. you want to install a compiler that is used to build packages in an environment, and generally the compiler doesn't exist before you create the environment
      • Possible way to do this:
        • One environment spec mentions the compiler: openmpi %gcc
        • You can set compilers: to an empty section
        • Set config:install_missing_compilers:true
      • Jack: I tried this earlier, it seemed to hang during install time (not concretize time)
• Dom: https://github.com/spack/spack/pull/35601
• Wileam: https://github.com/spack/spack/pull/35949
  • Error building gcc with CUDA 12
    • conftest.c >&5 cc1: error: PTX version ('-mptx') needs to be at least 4.2 to support selected '-misa' (sm_53)
    • There's a similar patch in the Spack Gcc package for CUDA 11

        if spec.satisfies("+nvptx"):
            # backport of 383400a6078d upstream to allow support of cuda@11:
            filter_file(
                '#define ASM_SPEC "%{misa=*:-m %*}"',
                '#define ASM_SPEC "%{misa=*:-m %*; :-m sm_35}"',
                "gcc/config/nvptx/nvptx.h",
                string=True,
            )