Skip to content

ALCF.md

Latest commit

 

History

History
95 lines (80 loc) · 3.65 KB

ALCF.md

File metadata and controls

95 lines (80 loc) · 3.65 KB
Raw

Build instructions for machines available at ALCF

Aurora @ ALCF

  • Below shows typical configure and make commands for building PnetCDF and its installation, which will produces both static and shared libraries. 
    ./configure --prefix=/path/to/install \
            MPICC=mpicc \
            MPICXX=mpicxx \
            MPIF77=mpifort \
            MPIF90=mpifort \
  • At the time of this writing (June 2026), the default MPI compilers installed on Aurora are Intel oneAPI based compilers. 
    % mpicc --version
Intel(R) oneAPI DPC++/C++ Compiler 2025.3.2 (2025.3.2.20260112)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/aurora/26.26.0/oneapi/compiler/2025.3/bin/compiler
Configuration file: /opt/aurora/26.26.0/oneapi/compiler/2025.3/bin/compiler/../icx.cfg

Note on testing PnetCDF, i.e. running "make check" or "make ptest".

Aurora is in a cross-compile environment, i.e. programs are compiled on the login nodes and run on compute nodes (the hardware and software may not be the same between the two nodes). The executable compiled for compute nodes must be through submitting a job to a batch scheduler. At the time of this writing (June 2026), the batch job scheduler is PBS (Portable Batch System).

There are two make command options for running the test. One is "make check" to run test programs sequentially and "make ptest" for parallel runs.

  • When running test programs developed for running on a single MPI process, use below lines to your batch script file which allocates a single compute node and appends two environment variables. Variable TESTSEQRUN sets the mpiexec command-line option to use 1 process only and variable `TESTOUTDIR set to the file directory for storing the test output files. 
    #PBS -l select=1
#PBS -l walltime=01:00:00

export LD_LIBRARY_PATH="/path/to/pnetcdf/installation:$LD_LIBRARY_PATH"

make check TESTSEQRUN="mpiexec -n 1" \
           TESTOUTDIR="/output/folder"
  • When run test programs developed for running in parallel, use below lines to your batch script file which allocates 4 compute nodes (a less number of nodes is also fine). "make ptest" requires at least 4 MPI processes. "make ptests" will run the tests using 2, 4, 6, and 8 processes. Variable TESTMPIRUN sets the mpiexec command-line option to use NP MPI processes, where NP will be adjusted internally by PnetCDF. Similarly, variable TESTOUTDIR set to the file directory for storing the test output files. 
    #SBATCH -N 4
#SBATCH -t 00:30:00
make ptest TESTMPIRUN="mpiexec -n NP" \
           TESTOUTDIR="/output/folder"

Polaris @ ALCF

  • Below shows typical configure and make commands for building PnetCDF and its installation, which will produces both static and shared libraries.

    ./configure --prefix=/path/to/install \
            cross_compiling="yes" \
            MPICC=cc \
            MPICXX=CC \
            MPIF77=ftn \
            MPIF90=ftn

  • At the time of this writing (June 2026), the default MPI compilers installed on Polaris are NVIDIA based compilers.

    % cc --version

nvc 25.5-0 64-bit target on x86-64 Linux -tp zen3-64
NVIDIA Compilers and Tools
Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES.  All rights reserved.

  • The batch job scheduler on Polaris is also PBS (Portable Batch System), same as on Aurora. Please refer to the above note for running test on Polaris.

Other machines historically available at ALCF

Please see README.CRAY.md.