Skip to content

Compilation failures with ROCm >= 5.7 on Frontier #522

Description

@baperry2

From @mrieth1:

I think that might just be incompatible with rocm >= 5.7, when I load:

module load cray-python
module load PrgEnv-cray
module load rocm/6.0.0
module load craype-accel-amd-gfx90a
module load craype-x86-trento
module load cray-libsci/23.12.5
module load cce/17.0.0

and set USE_HIP = TRUE in the FlameSheet case I get Unexpected HIP_PLATFORM: hcc when compiling TPL (that's without MAGMA). I think that's because HIP_PLATFORM=hcc is deprecated in 5.7. I could fix that in a quick and dirty way with export HIP_PLATFORM=amd and including #define __HIP_PLATFORM_HCC__ in sundials/include/sundials/sundials_hip_policies.hpp .
When I set PELE_USE_MAGMA=TRUE I get some additional errors messages during TPL compilation. I think some of that stuff might be related to bug fixes in sundials 7.1 (https://sundials.readthedocs.io/en/latest/Changelog_link.html).

The error with magma is this /lustre/orion/chm210/scratch/mrieth/pele_latest_test_Aug2024/test/PeleLMeX/Submodules/PelePhysics/ThirdParty/BUILD/MAGMA/magma-2.7.1/interface_hip/blas_z_v2.cpp:1853:9: error: no matching function for call to 'hipblasZtrmm'

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to 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