Hi All,
I have a code which uses OpenMP to run in parallel, using multiple mesh instances. From time to time, I'm dumping those instances into separate binary .om files, using separate ostream objects with the function
bool OpenMesh::IO::write_mesh ( const Mesh & _mesh, std::ostream & _os, const std::string & _ext, Options _opt = Options::Default, std::streamsize _precision = 6 )
Unfortunately, the execution seem to randomly fail with the error "double free or corruption (fasttop)". This is a backtrace:
======= Backtrace: ========= /lib64/libc.so.6(+0x81489)[0x2b63bd384489] /home1/06589/tg858882/assembly_openmesh_KNL/OpenMesh-9.0/build/Build/bin/../lib/libOpenMeshCore.so.9.0(_Z5omlogv+0x558)[0x2b63bc128488] /home1/06589/tg858882/assembly_openmesh_KNL/OpenMesh-9.0/build/Build/bin/../lib/libOpenMeshCore.so.9.0(_ZNK8OpenMesh2IO10_OMWriter_12write_binaryERSoRNS0_12BaseExporterENS0_7OptionsE+0x49)[0x2b63bc10aaf9] /home1/06589/tg858882/assembly_openmesh_KNL/OpenMesh-9.0/build/Build/bin/../lib/libOpenMeshCore.so.9.0(_ZNK8OpenMesh2IO10_OMWriter_5writeERSoRNS0_12BaseExporterENS0_7OptionsEl+0xe7)[0x2b63bc10aa67] /home1/06589/tg858882/assembly_openmesh_KNL/OpenMesh-9.0/build/Build/bin/../lib/libOpenMeshCore.so.9.0(_ZN8OpenMesh2IO11_IOManager_5writeERSoRKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERNS0_12BaseExporterENS0_7OptionsEl+0xd9)[0x2b63bc0adea9] ./Assembly(_Z7dump_omN8OpenMesh20TriMesh_ArrayKernelTINS_13DefaultTraitsEEElRSo+0x11c)[0x44e33c] ./Assembly(_Z9propagateRN8OpenMesh20TriMesh_ArrayKernelTINS_13DefaultTraitsEEEll13RunParametersRSoS5_R14UmbrellaWindowRSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE+0x36e)[0x4f89ee] ./Assembly(_Z31run_umbrella_parallel_temperingNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x7c4)[0x4f3b84] /opt/intel/compilers_and_libraries_2018.2.199/linux/compiler/lib/intel64_lin/libiomp5.so(__kmp_invoke_microtask+0x93)[0x2b63bc556563] /opt/intel/compilers_and_libraries_2018.2.199/linux/compiler/lib/intel64_lin/libiomp5.so(+0xa93ea)[0x2b63bc51e3ea] /opt/intel/compilers_and_libraries_2018.2.199/linux/compiler/lib/intel64_lin/libiomp5.so(+0xa8aab)[0x2b63bc51daab] /opt/intel/compilers_and_libraries_2018.2.199/linux/compiler/lib/intel64_lin/libiomp5.so(+0xe19c0)[0x2b63bc5569c0] /lib64/libpthread.so.0(+0x7dd5)[0x2b63bc854dd5] /lib64/libc.so.6(clone+0x6d)[0x2b63bd400ead]
As you can see, it traces back to OMWriter and indeed, the error seems to come up upon writing. I was therefore wondering if there is anything that prevents such parallel writing (to separate, binary files, via separate ostream objects) or this is expected to work properly?
Thanks a lot for your help, Botond