Math Library Optimizations

Repository for testing and analyzing the performance of low-level math libraries in RHEL

Current Code

src/libgomp: Contains C code used for testing the performance of libgomp

Compiling the Code

Under each subdirectory (e.g., src/libgomp), there is a Makefile. To build the code,

$ make -C src/<library>

To clean the code,

$ make -C src/<library> clean

Replace <library> with an existing library -- e.g., libgomp

Dockerfiles

This repository contains Dockerfiles for running various benchmarks. To build,

$ podman build -f Dockerfiles/Dockerfile.<name> .

The WORKDIR value is set to the directory where the benchmarks should be run. You shouldn't need to change directories.

Note that if you're using the custom libgomp podman build, you should run scripts/setup_gcc_volume.sh first to prepare the custom libraries for mounting in the container.

Once that is done, you can run the image

$ podman run -v /tmp/gcc-podman-mnt:/gcc:z ${IMAGE_NAME_OR_ID} /bin/bash

Make sure to run run_me_first.sh before continuing with anything. (This script is the only script in the current working directory.) This script will prepare the libgomp benchmarks build for you. i.e., you should run the script like so:

$ . ./run_me_first.sh

Running this script will setup your gcc/libgomp build, and it will change you to the appropriate directory for running the benchmarks.

libgomp

To run libgomp ULHPC benchmarks,

$ ./bin/xhpcg <nx> <ny> <nz> <desired-runtime>
$ #e.g., ./bin/xhpcg 280 280 280 1800

The nx, ny, and nz represent the global problem size in the x, y, and z dimensions. Ideally, you want official runtimes to be >1800s. (For more information, see this pdf from AMD.)