2. Change versions
You can install other versions by compiling from source, the version installed from source will be newer.
# For OpenMPI
cd ~ && wget https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-4.1.5.tar.gz
tar -xzvf openmpi-4.1.5.tar.gz && cd openmpi-4.1.5
./configure --prefix=/usr/local/openmpi
sudo make all install -j8
echo "export PATH=/usr/local/openmpi/bin:$PATH" >> ~/.bashrc
echo "export LD_LIBRARY_PATH=/usr/local/openmpi/lib:$LD_LIBRARY_PATH" >> ~/.bashrc
source ~/.bashrc
# For OpenBLAS
cd ~ && wget https://github.com/xianyi/OpenBLAS/releases/download/v0.3.23/OpenBLAS-0.3.23.tar.gz # If there is a network problem, please switch to https://ghproxy.com/https://github.com/xianyi/OpenBLAS/releases/download/v0.3.23/OpenBLAS-0.3.23.tar.gz
tar -xzvf OpenBLAS-0.3.23.tar.gz && cd OpenBLAS-0.3.23
make FC=gfortran -j8
sudo make PREFIX=/usr/local/openblas install
# For MPL
cd ~ && wget https://www.netlib.org/benchmark/hpl/hpl-2.0.tar.gz
tar -xzvf hpl-2.0.tar.gz && cd hpl-2.0
You need to change the MPdir and LAdir to the corresponding installation paths, like this:
TOPdir = $(HOME)/hpl-2.0
MPdir = /usr/local/openmpi
MPlib = $(MPdir)/lib/libmpi.so
LAdir = /usr/local/openblas
LAlib = $(LAdir)/lib/libopenblas.a
CC = /usr/local/openmpi/bin/mpicc
LINKER = /usr/local/openmpi/bin/mpif77
Then recompile MPL and test again.
cp ~/hpl-2.3/bin/Linux_PII_CBLAS/HPL-Tuning.dat ~/hpl-2.0/bin/Linux_PII_CBLAS/HPL.dat
cd ~/hpl-2.0/bin/Linux_PII_CBLAS
mpirun -np 8 xhpl
Output:
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR13L2L2 16384 192 1 1 15.90 1.844e+02
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 0.0022339 ...... PASSED
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR13L2L2 20352 192 1 1 35.05 1.604e+02
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 0.0019320 ...... PASSED
================================================================================
It seems the results are even worse:>
5. In Windows
If you are on windows, you can simply install Intel® oneAPI Base Toolkit to get benchmarks and Intel® oneAPI HPC Toolkit to get mpiexec.
After installation, Run the terminal in administrator mode and go to the directory <oneAPI Install path>\mkl\2023.1.0\benchmarks\mp_linpack.
Then run runme_intel64_dynamic.bat Or run xhpl_intel64_dynamic.exe directly.
This is a SAMPLE run script. Change it to reflect the correct
number of CPUs/threads, number of nodes, MPI processes per node, etc..
This run started on:
05/20/2023 Sat
01:02
Capturing output into: xhpl_intel64_dynamic_outputs.txt
Done:
05/20/2023 Sat
01:02
And you will get a HPL report xhpl_intel64_dynamic_outputs.txt in the current directory.
Then you can refer to the performance tuning guide above to improve your score.
