[tools] memory_test_crosscorrelate fails with mkl

[junghans]

see votca/tools#267 (comment) for details

[JensWehner]

I would just disable the tests

[junghans]

We excluded them for now in votca/tools@e553a43, but we could also disable them completely.

[junghans]

@JoshuaSBrown any opinion?

The full backtrace is:

==11438== Memcheck, a memory error detector
==11438== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==11438== Using Valgrind-3.16.0 and LibVEX; rerun with -h for copyright info
==11438== Command: /__w/tools/tools/builddir/tools/src/tests/unit_test_crosscorrelate
==11438== 
==11438== Syscall param sched_setaffinity(mask) points to unaddressable byte(s)
==11438==    at 0xC9571BD: syscall (in /usr/lib64/libc-2.30.so)
==11438==    by 0xC02153C: __kmp_affinity_determine_capable (z_Linux_util.cpp:189)
==11438==    by 0xBFC28B2: __kmp_env_initialize(char const*) (kmp_settings.cpp:5747)
==11438==    by 0xBFA9B8B: __kmp_do_serial_initialize (kmp_runtime.cpp:7093)
==11438==    by 0xBFA9B8B: __kmp_do_middle_initialize (kmp_runtime.cpp:7236)
==11438==    by 0xBFA9B8B: __kmp_middle_initialize (kmp_runtime.cpp:7345)
==11438==    by 0xBF873D1: omp_get_num_procs@@VERSION (kmp_ftn_entry.h:616)
==11438==    by 0x9B8F8ED: mkl_serv_get_dynamic (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_intel_thread.so)
==11438==    by 0xE28723D: static_dfti_commit (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_avx2.so)
==11438==    by 0x825B17F: DftiCommitDescriptor (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_core.so)
==11438==    by 0x947013B: fftw_plan_guru64_dft_r2c (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_intel_lp64.so)
==11438==    by 0x946EFCA: fftw_plan_dft_r2c (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_intel_lp64.so)
==11438==    by 0x946F00A: fftw_plan_dft_r2c_1d (in /opt/intel/compilers_and_libraries_2020.1.217/linux/mkl/lib/intel64_lin/libmkl_intel_lp64.so)
==11438==    by 0x4894410: void Eigen::FFT<double, Eigen::default_fft_impl<double> >::fwd<Eigen::Map<Eigen::Matrix<double, -1, 1, 0, -1, 1>, 0, Eigen::Stride<0, 0> >, Eigen::Matrix<std::complex<double>, -1, 1, 0, -1, 1> >(Eigen::MatrixBase<Eigen::Matrix<std::complex<double>, -1, 1, 0, -1, 1> >&, Eigen::MatrixBase<Eigen::Map<Eigen::Matrix<double, -1, 1, 0, -1, 1>, 0, Eigen::Stride<0, 0> > > const&, long) (in /__w/tools/tools/builddir/tools/src/libtools/libvotca_tools.so.7)
==11438==    by 0x4893CA7: votca::tools::CrossCorrelate::AutoCorrelate(votca::tools::DataCollection<double>::selection&) (in /__w/tools/tools/builddir/tools/src/libtools/libvotca_tools.so.7)
==11438==    by 0x4043E3: crosscorrelate_test::crosscorrelate::test_method() (in /__w/tools/tools/builddir/tools/src/tests/unit_test_crosscorrelate)
==11438==    by 0x403D64: crosscorrelate_test::crosscorrelate_invoker() (in /__w/tools/tools/builddir/tools/src/tests/unit_test_crosscorrelate)
==11438==    by 0x494A501: ??? (in /usr/lib64/libboost_unit_test_framework.so.1.69.0)
==11438==    by 0x494956C: boost::execution_monitor::catch_signals(boost::function<int ()> const&) (in /usr/lib64/libboost_unit_test_framework.so.1.69.0)
==11438==    by 0x49495F7: boost::execution_monitor::execute(boost::function<int ()> const&) (in /usr/lib64/libboost_unit_test_framework.so.1.69.0)
==11438==    by 0x49496CD: boost::execution_monitor::vexecute(boost::function<void ()> const&) (in /usr/lib64/libboost_unit_test_framework.so.1.69.0)
==11438==    by 0x497391E: boost::unit_test::unit_test_monitor_t::execute_and_translate(boost::function<void ()> const&, unsigned int) (in /usr/lib64/libboost_unit_test_framework.so.1.69.0)
==11438==  Address 0x0 is not stack'd, malloc'd or (recently) free'd
==11438== 

I am a bit confused why fftw_plan_dft_r2c_1d is coming from libmkl_intel_lp64.so and not libfftw3.so.

[junghans]

I can reproduce this locally in side the votca/buildenv:intel container.

docker run -it votca/buildenv:intel /bin/bash
git clone --recursive https://github.com/votca/votca
cd votca
cmake -B build -DENABLE_TESTING=ON
cd build
make
ctest -V -R  memory_test_crosscorrelate

[JensWehner]

yes that mkl thing is weird. Somehow the eigen interface seems to work with the mkl fftw. The unit test for correlate does not fail. it seems that the fftw and mkl interface for fft are the same. Really weird.

[junghans]

Yeah, mkl has an internal fftw implementation, so there are duplicated symbols.

[junghans]

Revert 314aa1e when issue was found.