Performance tests for QMCkl.

Installation

1. Download and install QMCkl
2. Compile the current tests and run

./autogen.sh
./configure
make
./bench_jastrow

Results

CPU

Intel(R) Xeon(R) CPU E3-1271 v3 @ 3.60GHz, 4cores

GNU options for QMCkl on x86

./configure \
    CC=gcc \
    CFLAGS="-g -O2 -ffast-math  -march=native  -flto -fno-trapping-math -fno-math-errno -ftree-vectorize -fno-stack-protector" \
    FC=gfortran \
    FCFLAGS="-g -O2 -fstack-arrays -ffast-math -march=native  -flto -ftree-vectorize -fno-stack-protector" \
    --enable-hpc --with-openmp --enable-static

GNU options for QMCkl on ARM

./configure \
    CC=gcc \
    CFLAGS="-g -mcpu=native -fno-signaling-nans -fno-trapping-math -freciprocal-math -fno-signed-zeros -fno-math-errno -ffinite-math-only -funroll-loops -O3 -ftree-vectorize -flto -fopenmp-simd" \
    FC=gfortran \
    FCFLAGS="-g -mcpu=native  -O2 -fstack-arrays -ffast-math -flto -ftree-vectorize -fno-stack-protector" \
    --enable-hpc --with-openmp --enable-static

Intel options for QMCkl

./configure --with-icc --with-ifort --enable-hpc --with-openmp --enable-static

# -xHost -ip -Ofast -ftz -finline -g -mkl=sequential -fopenmp

Clang options for QMCkl

./configure \
    CC=clang \
    CFLAGS="-Ofast -g -march=native" \
    FC=flang  \
    FCFLAGS="-Ofast -g -march=native" \
    --enable-hpc --with-openmp --enable-static

Measures

HPC	Compiler	bench_aos (ms)	bench_aos OpenMP (ms)	bench_mos (ms)	bench_mos OpenMP (ms)
No	GCC-8	112.7	113.1	2036.1	2038.7
Yes	GCC-8	68.0	24.8	896.2	250.3
Yes	Intel-2021.5	60.5	24.1	890.0	231.6

Memo

Run MAQAO with

maqao oneview --create-report=three --output-format=html --force-all-loops --external-libraries={\"libqmckl.so.0\"} -- ./bench_aos data/Alz_small.h5