Segmentation fault in LAPACKE_dgesv for large matrices

szp — Sat, 06 Apr 2024 03:19:30 GMT

Hello! I'm learning usage of MKL these days. When I tried to use LAPACKE_dgesv to solve linear equations, I find that it works for small matrices but fails for large matrices (segmentation fault (core dumped)). My CPU is AMD, operating system is Rockylinux 9, and OneMKL version is 2023.2.0. I have also tried 2024.1, but it has the same problem. I have also set ulimit to unlimit. Here is my code.

#include <iostream> #include <string> #include <iomanip> #include <random> #include <mkl.h> using std::cout; using std::endl; using std::string; using std::setw; using std::setprecision; using std::fixed; double generateRandomDouble(double min, double max) { static std::random_device rd; static std::mt19937 gen(rd()); std::uniform_real_distribution<double> dis(min, max); return dis(gen); } int linsol(double *A,double *B, int lda, int ldb, int i) { lapack_int *ipiv = new lapack_int[lda]; int info = 100; if (i == 0) info = LAPACKE_dgesv(LAPACK_ROW_MAJOR, lda, ldb, A, lda, ipiv, B, ldb); else { info = LAPACKE_dgesv(LAPACK_COL_MAJOR, lda, ldb, A, lda, ipiv, B, lda); } delete[] ipiv; return info; } void print(double *A, int row, int col) { for(int i = 0; i < row; ++i) { for(int j = 0; j < col; ++j) cout << setw(8) << fixed << setprecision(4) << A[i*col+j] << " "; cout << endl; } } int main(int argc, char *argv[]) { int n = std::stoi(argv[1]); double *matrix_a = (double*)mkl_malloc(n * n * sizeof(double), 64); double *matrix_b = (double*)mkl_malloc(n * n * sizeof(double), 64); for (int i = 0; i < n; ++i) { for (int j = 0; j < n; ++j) matrix_a[i * n + j] = generateRandomDouble(0.0,1.0); } for (int i = 0; i < n; ++i) { for (int j = 0; j < n; ++j) matrix_b[i * n + j] = generateRandomDouble(0.0, 1.0); } print(matrix_a, n, n); cout << endl; print(matrix_b, n, n); cout << linsol(matrix_a, matrix_b, n, n, 1) << endl; print(matrix_b, n, n); mkl_free(matrix_a); mkl_free(matrix_b); return 0; }

And my CMakelists.txt is as follows (MKLROOT has been set as environment variable).

cmake_minimum_required(VERSION 3.10) project(mkl LANGUAGES CXX) file(GLOB SRC ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp) set(CMAKE_CXX_STANDARD 17) list(APPEND CMAKE_CXX_FLAGS -O3) find_package(MKL REQUIRED) set(EXECUTABLE_OUTPUT_PATH ${CMAKE_CURRENT_SOURCE_DIR}) add_executable(1 ${SRC}) target_link_libraries(1 iomp5 ${MKL_LIBRARIES})

The configuration is as follows.

-- The CXX compiler identification is GNU 11.4.1 -- Detecting CXX compiler ABI info -- Detecting CXX compiler ABI info - done -- Check for working CXX compiler: /usr/bin/c++ - skipped -- Detecting CXX compile features -- Detecting CXX compile features - done -- MKL_ARCH: None, set to ` intel64` by default -- MKL_LINK: None, set to ` dynamic` by default -- MKL_INTERFACE_FULL: None, set to ` intel_ilp64` by default -- MKL_THREADING: None, set to ` intel_thread` by default -- Found MKL: /opt/intel/oneapi/mkl/2023.2.0 -- Found MKL: /opt/intel/oneapi/mkl/2023.2.0/lib/intel64/libmkl_intel_ilp64.so -- Found MKL: /opt/intel/oneapi/mkl/2023.2.0/lib/intel64/libmkl_intel_thread.so -- Found MKL: /opt/intel/oneapi/mkl/2023.2.0/lib/intel64/libmkl_core.so -- Found MKL: /opt/intel/oneapi/compiler/latest/linux/compiler/lib/intel64_lin/libiomp5.so -- Configuring done (0.3s) -- Generating done (0.0s) -- Build files have been written to: /path/to/my/dir

On my computer, when n >=25, it appears "Segmentation fault (core dumped)", and the returned value of LAPACKE_dgesv is not printed on the screen, so I think the error occurs in the function. So what is the problem? How to solve it? Very thanks!

Re: Segmentation fault in LAPACKE_dgesv for large matrices

Gennady_F_Intel — Wed, 10 Apr 2024 07:14:35 GMT

At the first glance the problem is that you link against ilp64 libraries (libmkl_intel_ilp64.so ) without using compiler option -DMKL_ILP64. Please check the MKL Linker adviser recommendation.

You might try to relink against libmkl_intel_lp64.so without modification you code and check how is will go.

Re: Segmentation fault in LAPACKE_dgesv for large matrices

szp — Wed, 10 Apr 2024 07:51:44 GMT

Very thanks! The problem is solved after I link my program to libmkl_intel_lp64.so.

topic Re: Segmentation fault in LAPACKE_dgesv for large matrices in Intel® oneAPI Math Kernel Library

Segmentation fault in LAPACKE_dgesv for large matrices

Re: Segmentation fault in LAPACKE_dgesv for large matrices

Re: Segmentation fault in LAPACKE_dgesv for large matrices