We are trying to build neuron from source, to be imported in Python, using the Intel compilers 2017, and Intel MPI, on the Stampede KNL (TACC) system that has a Haswell login.
Our steps are 1. configure, 2. make, 3. make install
Code: Select all
./configure --prefix=$IDIR/nrn75 --without-iv --without-nrniv --without-nrnoc-x11 --without-x -with-paranrn CC=icc CXX=icpc CPP='icc -E' CXXCPP='icpc -E' CFLAGS='-xCORE-AVX2 -axMIC-AVX512' LDFLAGS='-xCORE-AVX2 -axMIC-AVX512' CXXFLAGS='-xCORE-AVX2 -axMIC-AVX512' MPICC=mpiicc MPICXX=mpiicpc LD=xild AR=xiar --with-nrnpython=python3.6
We also used the '-no-ipo', '-O0' flags separately and together, but we got the same errors.mpiicpc -shared -ipo -Xlinker -export-dynamic build/temp.linux-x86_64-3.6/ctng.o -L/opt/apps/intel17/python3/3.6.0/lib -lpython3.6m -o /work/04119/pmoolcha/Build/nrn-7.5/share/lib/python/neuron/rxd/geometry3d/ctng.cpython-36m-x86_64-linux-gnu.so
": internal error: 010101_(5000 + 54)
icpc: error #10014: problem during multi-file optimization compilation (code 4)
1. Ideally, we would like to have the code compiled with the Intel compilers and Intel MPI whilst allowing any potential optimization to proceed.
2. In case that is not possible, please let us know how to turn off the optimization for that particular step since it ignores both flags and yields the error.