phi3-results.txt

Compile with different options on Skylake Gold.

Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GH

———————————————————————————————————————————————————————————

which icc
/opt/intel/compilers_and_libraries_2019.0.117/linux/bin/intel64/icc

icc -O3 -I /home/users/cerati/miniconda2/include/ xtensor-test.cc -o xtensor-test.exe
time for NN=10000000 multiplications is 7.02 s

-O3 -no-vec
time for NN=10000000 multiplications is 7.02 s

-O3 -xCOMMON-AVX512
time for NN=10000000 multiplications is 6.88 s

-O3 -mavx2 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 3.08 s

-O3 -mavx -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 3.06 s

-O3 -xCORE-AVX512 -qopt-zmm-usage=high -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.91 s

-O3 -xCORE-AVX512 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.89 s

-O3 -xCOMMON-AVX512 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.88 s

-O3 -march=skylake-avx512 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.87 s

———————————————————————————————————————————————————————————

which g++
/cvmfs/cms.cern.ch/slc7_amd64_gcc630/external/gcc/6.3.0/bin/g++

g++ -O3 -I /home/users/cerati/miniconda2/include/ xtensor-test.cc -o xtensor-test.exe
time for NN=10000000 multiplications is 3.79 s

-O3 -fno-tree-loop-vectorize
time for NN=10000000 multiplications is 3.99 s

-O3 -mavx
time for NN=10000000 multiplications is 3.79 s

-O3 -fno-tree-loop-vectorize -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 3.66 s

-O3 -mavx -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 3.15 s

-O3 -mavx2 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.9 s

-O3 -march=skylake -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.83 s

-O3 -march=skylake-avx512 -DXTENSOR_USE_XSIMD
time for NN=10000000 multiplications is 2.79 s