LAPACKE Matrix Factorization Routines (sgetrf, sgeqrf) Crash with Exit Code 0xC06D007E

[draush]

I'm encountering crashes specifically with LAPACKE matrix factorization routines in Intel oneAPI MKL 2025.0 on Windows. The program crashes with exit code 0xC06D007E when calling factorization routines like LAPACKE_sgetrf and LAPACKE_sgeqrf. Importantly, basic LAPACK operations and CBLAS functions work correctly.

I've attempted a reinstall of oneAPI MKL through this link to no avail.

Environment:
- Windows 10
- Intel CPU
- Intel oneAPI MKL 2025.0
- CMake 3.28
- C++17

Working Operations:
- Basic CBLAS operations
- LAPACKE_slacpy (matrix copy)
- LAPACKE_slantr (matrix norm)
- LAPACKE_slange (general matrix norm)
- LAPACKE_slaset (matrix initialization)

Failing Operations:
- LAPACKE_sgetrf (LU factorization)
- LAPACKE_sgeqrf (QR factorization)
- LAPACKE_sgetri (matrix inversion, fails at the required sgetrf step)

Steps to reproduce:
1. Create a new project with the attached CMakeLists.txt and main.cpp:

#include <iostream>
#include <mkl.h>
#include <mkl_lapacke.h>
#include <vector>

void print_matrix(const std::vector<float>& matrix, int n) {
    for (int i = 0; i < n; i++) {
        for (int j = 0; j < n; j++) {
            std::cout << matrix[i * n + j] << " ";
        }
        std::cout << std::endl;
    }
    std::cout << std::endl;
}

int main() {
    const int N = 3;

    // First demonstrate a working LAPACK operation
    std::cout << "Testing working LAPACK operation (matrix copy)..." << std::endl;
    std::vector<float> test_matrix(N * N, 1.0f);
    std::vector<float> matrix_copy(N * N);
    MKL_INT info = LAPACKE_slacpy(LAPACK_ROW_MAJOR, 'A', N, N, test_matrix.data(), N, matrix_copy.data(), N);
    std::cout << "LAPACKE_slacpy completed successfully with info = " << info << std::endl << std::endl;

    // Now demonstrate the failing operation
    std::cout << "Testing failing operation (LU factorization)..." << std::endl;
    std::vector<float> matrix(N * N, 0.0f);
    for (int i = 0; i < N; i++) {
        matrix[i * N + i] = 1.0f;
    }

    std::cout << "Original matrix:" << std::endl;
    print_matrix(matrix, N);

    std::vector<MKL_INT> ipiv(N);
    std::cout << "Beginning LAPACKE_sgetrf routine..." << std::endl;
    info = LAPACKE_sgetrf(LAPACK_ROW_MAJOR, N, N, matrix.data(), N, ipiv.data());

    // Note: The following lines will not be reached due to the crash
    if (info != 0) {
        std::cerr << "LU factorization failed with error " << info << std::endl;
        return 1;
    }

    std::cout << "Matrix after LU factorization:" << std::endl;
    print_matrix(matrix, N);

    return 0;
}

2. Build and run

3. Output:

```

Testing working LAPACK operation (matrix copy)...
LAPACKE_slacpy completed successfully with info = 0

Testing failing operation (LU factorization)...
Original matrix:
1 0 0
0 1 0
0 0 1

Beginning LAPACKE_sgetrf routine...

Process finished with exit code -1066598274 (0xC06D007E)

```

[Shiquan_Su]

Hi,

I tested your example against our coming oneapi 2025.1.0 release. I don't see the error message on the second case anymore. Here is what I see:

shiquans@orcsle163:/cts/projects/shiquans/osc_06489336$ cat runme.sh

#!/bin/bash

module purge

module load utils/cmake-3.28.4

module load oneapi/2025.1.0-preview

cmake .

make

./mkl_bug

shiquans@orcsle163:/cts/projects/shiquans/osc_06489336$ . ./runme.sh

/cts/tools/utils/cmake-3.28.4/cmake-3.28.4-linux-x86_64//bin/cmake

/nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/compiler/2025.0/bin/icx

-- MKL_VERSION: 2025.0.0

-- MKL_ROOT: /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0

-- MKL_ARCH: intel64

-- MKL_LINK: dynamic

-- MKL_INTERFACE_FULL: intel_ilp64

-- MKL_THREADING: intel_thread

-- MKL_MPI: intelmpi

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_scalapack_ilp64.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_cdft_core.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_intel_ilp64.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_intel_thread.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_core.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/mkl/2025.0/lib/libmkl_blacs_intelmpi_ilp64.so

-- Found /nfs/pdx/disks/cts2/tools/oneapi/2025.0.0-preview/compiler/2025.0/lib/libiomp5.so

-- MKL::mkl_scalapack_ilp64;MKL::mkl_cdft_core;MKL::mkl_intel_ilp64;MKL::mkl_intel_thread;MKL::mkl_core;MKL::mkl_blacs_intelmpi_ilp64;MKL::MKL

-- Configuring done (0.1s)

-- Generating done (0.1s)

-- Build files have been written to: /cts/projects/shiquans/osc_06489336

[100%] Built target mkl_bug

Testing working LAPACK operation (matrix copy)...

LAPACKE_slacpy completed successfully with info = 0

Testing failing operation (LU factorization)...

Original matrix:

1 0 0

0 1 0

0 0 1

Beginning LAPACKE_sgetrf routine...

Matrix after LU factorization:

1 0 0

0 1 0

0 0 1

The oneAPI 2025.1.0 will be released around end of Q1 2025.

[draush · Accepted solution]

Thank you for your reply.

As I needed a solution quicker than the end of Q1 2025, I found a workaround for this problem by using the intel-mkl port through vcpkg.

[Fengrui]

Hi,

I tried the code with the current oneMKL 2025.0.1 release and observed no issues. This exit code seems to be related to something else.

Thanks,

Fengrui