Run HPL Benchmark with AMD BLIS library

High Performance Linpack (HPL) is a software package that solves a (random) dense linear system in double precision (64 bits) arithmetic on distributed-memory computers. HPL is a Linpack benchmark which measures the floating point rate of execution for solving a linear system of equations.

We have created an utility that builds and runs HPL. It also builds AMD BLIS library that is used to link to HPL as the dependent BLAS library.

The utility can be downloaded here https://github.com/amd/blis/releases/download/1.2/install_run_hpl-blis.tar.gz

The utility package contains following files

1. README.txt
2. build_run_hpl_blis.sh - Script to build and run HPL. Also pulls AMD BLIS library from github and builds and links to HPL
3. HPL.dat - Sample HPL dat file
4. Make.Linux_AMD_BLIS - Configuration file with AMD BLIS library paths

---

Instructions to run the script

---

1. Edit the HPL.dat file to required tuning parameters
2. Edit the MPI path variable 'MPdir' in Make.Linux_AMD_BLIS to appropriate path on the target machine
3. Edit the 'MPI_PATH' in build_run_hpl_blis.sh to appropriate MPI library path on the target machine
4. Set NUM_ITER to the required number of iterations
5. Set BLIS_MT to use either multi-threaded or single threaded version of BLIS
6. Set NUM_MPI_PROCESS_MT to required MPI Rank in case of multi-thread BLIS run or Set NUM_MPI_PROCESS_ST to required MPI Rank in case of single-thread BLIS run
7. Run ./build_run_hpl_blis.sh

The script on successful run, installs BLIS library under $HOME/amd/blis. HPL is built and installed under $HOME/hpl. It then runs binary xhpl for the set number of iterations