Profiling

MAQAO

MAQAO (Modular Assembly Quality Analyzer and Optimizer) is a performance analysis and optimization framework operating at binary level with a focus on core performance. Its main goal of is to guide application developpers along the optimization process through synthetic reports and hints.

The sample JSON files under profiling/ are used to instrument this tool.

First steps and setup

• Grab the tar: wget https://maqao.org/maqao_archive/maqao.x86_64.2.20.1.tar.xz
• Unpack: tar -xf maqao.x86_64.2.20.1.tar.xz
• Export the binary path for ease of use: export MAQ=path/to/maqao.x86_64.2.20.1/bin/maqao

Sample use for MPI

MAQAO is fairly complex in its use. Outlined here is one such command using the bespoke MPI-enabled configuration: template_mpi.json. Consult the documentation for more examples and uses.

Assumptions

The configuration assumes:

• one node of 48 cores
• one thread is expected for each rank
• OMP_NUM_THREADS set to 1
• profiled executable is at path/to/git/fortran-project-template/bin/app_name

You will need to amend these to configure your profiling session.

Usage

First use fpm to compile and install the program with the appropriate flags: -g -fno-omit-frame-pointer and any other optimization options appropriate. Interfacing maqao and fpm run is non-trivial, while installing the binary is.

fpm @installgcc-maqao  # see fpm.rsp for more information

Then invoke maqao, pointing to its config file.

$MAQ oneview -R1 --config=profiling/template_mpi.json -xp=profiling/maqao-test -force-static-analysis  -maximum-threads-per-process=2

This will run the executable as given inside template_mpi.json with the appropriate commands, and generate the output under profiling/maqao-test/.

After profiling is finished, if on a cluster, rsync/scp the maqao-test/ dir locally and open it with a browser:

rsync -avz --partial user@login.hpc.cluster.ac.uk:/path/to/fortran-project-template/profiling/maqao-test .
firefox maqao-test/RESULTS/app_name_one_html/index.html

Caveats

The config is set up for MPI profiling on a single node, and OpenMP threading set to 1. Adjust the config accordingly.