topic Re:omp warning when using pardiso and metis threaded version in Intel® oneAPI Math Kernel Library

omp warning when using pardiso and metis threaded version

may_ka — Fri, 25 Feb 2022 00:20:44 GMT

Hi,

when running pardiso and setting iparm[1]=3 (C interface) I get the warning message:

```

OMP: Warning #97: Requested number of active parallel levels "-2147483648" is negative; ignored

```

The relevant omp setting is "OMP_MAX_ACTIVE_LEVELS=2147483647", which should be perfectly fine for a 32-bit integer.

Mkl version is oneapi 2021-2, operation system linux. The mkl interface is ilp64.

Any idea?

Best

Re:omp warning when using pardiso and metis threaded version

Gennady_F_Intel — Fri, 25 Feb 2022 05:10:18 GMT

Which OpenMP threading do you use? checking with mkl_intel_thread.so linked with the testing examples (mkl 2022), I see no such messages.

Re: omp warning when using pardiso and metis threaded version

may_ka — Fri, 25 Feb 2022 06:52:00 GMT

the link line is:

-Wl,--start-group \ /opt/oneapi/mkl/2021.2.0/lib/intel64/libmkl_intel_ilp64.a \ /opt/oneapi/mkl/2021.2.0/lib/intel64/libmkl_core.a \ /opt/oneapi/mkl/2021.2.0/lib/intel64/libmkl_intel_thread.a -l iomp5 -l pthread -lm -ldl \ -Wl,--end-group

Re:omp warning when using pardiso and metis threaded version

Gennady_F_Intel — Fri, 25 Feb 2022 07:38:40 GMT

what is the matrix type?

Checking the problem right now with -2 matrix type on the real workload, I see no problems here.

Re: Re:omp warning when using pardiso and metis threaded version

may_ka — Fri, 25 Feb 2022 10:26:51 GMT

The matrix type is 2 (real and symmetric positive definite), and iparm[1]=3 (C-Interface, ".. The parallel (OpenMP) version of the nested dissection algorithm..")

Re:omp warning when using pardiso and metis threaded version

Gennady_F_Intel — Fri, 25 Feb 2022 15:26:23 GMT

Karl,

checking with mtype=2. I have had no success reproducing:

icc --version : icc (ICC) 2021.5.0 20211109

echo $MKLROOT : /opt/intel/oneapi/mkl/2022.0.2

We need the test to go further.

Re: omp warning when using pardiso and metis threaded version

may_ka — Fri, 25 Feb 2022 23:47:38 GMT

Hi,

thank's for the update.

I use g++

g++ --version g++ (GCC) 11.2.0

Compiling a stand-alone executable which calls pardiso with this

g++ -O3 -std=c++20 -c main.cpp -o main.o -I /opt/intel/oneapi/mkl/2021.2.0/include/intel64/ilp64 -I /opt/intel/oneapi/mkl/2021.2.0/include/ g++ -static -L /opt/intel/oneapi/compiler/2021.2.0/linux/compiler/lib/intel64_lin -o exe main.o -Wl,--start-group /opt/intel/oneapi/mkl/2021.2.0/lib/intel64/libmkl_intel_ilp64.a /opt/intel/oneapi/mkl/2021.2.0/lib/intel64/libmkl_core.a /opt/intel/oneapi/mkl/2021.2.0/lib/intel64/libmkl_intel_thread.a -l iomp5 -l pthread -lm -ldl -Wl,--end-group /usr/bin/ld: /opt/intel/oneapi/mkl/2021.2.0/lib/intel64/libmkl_core.a(mkl_memory_patched.o): in function `mkl_serv_set_memory_limit': mkl_memory.c:(.text+0x5d1): warning: Using 'dlopen' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking

and this omp setting:

printenv | egrep "OMP|KMP" OMP_PLACES=cores OMP_PROC_BIND=true OMP_DYNAMIC=FALSE KMP_AFFINITY=granularity=core,scatter OMP_MAX_ACTIVE_LEVELS=2147483647 OMP_NUM_THREADS=18 OMP_STACKSIZE=2000M

produces this screen output:

OMP: Warning #182: OMP_PLACES: ignored because KMP_AFFINITY has been defined OMP: Warning #182: OMP_PROC_BIND: ignored because KMP_AFFINITY has been defined OMP: Warning #97: Requested number of active parallel levels "-2147483648" is negative; ignored.

Hope this helps.

Re: omp warning when using pardiso and metis threaded version

may_ka — Sat, 26 Feb 2022 02:51:18 GMT

Hi,

thanks for looking into this.

I am using g++

g++ --version g++ (GCC) 11.2.0

an executable calling pardiso, when compiled and linked with

running with environment variables

printenv | egrep "OMP|KMP|MKL" OMP_PLACES=cores OMP_PROC_BIND=true OMP_DYNAMIC=FALSE KMP_AFFINITY=granularity=core,scatter MKLROOT=/opt/intel/mkl OMP_MAX_ACTIVE_LEVELS=2147483647 OMP_NUM_THREADS=18 OMP_STACKSIZE=2000M MKL_NUM_THREADS=18

produces these massages:

Hope that helps.