programming

BLAS and LAPACK on Windows

Using BLAS / LAPACK with Visual Studio on Windows

Many open source math libraries will depend on a BLAS / LAPACK provider being available on your system. While for Linux it’s just a matter of installing one of the providers maintained by your distribution, for Windows users this had been a hassle for a long time. In recent years it has become a lot easier with three major providers offering both binary redistributables and dev packages. This post describes where to obtain the relevant packages for Windows. You might also want to read about how to manage C/C++ library dependencies on Windows.

OpenBLAS

OpenBLAS is an open source library distributed under 3-clause BSD license. You can obtain precompiled versions from github.com/OpenMathLib/OpenBLAS/releases. The latest version (as of July 2025) is 0.3.30.

The shared link library libopenblas.lib might not work for Visual Studio users, so you will have to create a new import library:

  • Open Visual Studio x64 Native Tools Command Prompt
  • cd to the directory that contains libopenblas.def,
  • Delete existing libopenblas.lib (you can actually delete all files except libopenblas.def)
  • Run lib /def:libopenblas.def /machine:x64 /out:libopenblas.lib

Optionally, you can extract the exported symbols with dumpbin /EXPORTS libopenblas.dll > libopenblas.exp and then copy the symbol names to libopenblas.def.

You can use OpenBLAS with CMake by setting the BLA_VENDOR option to OpenBLAS. If the libraries cannot be found automatically, specify them explicitly using

cmake -DBLA_VENDOR=OpenBLAS
      -DBLAS_LIBRARIES=C:/usr/lib/libopenblas.lib
      -DLAPACK_LIBRARIES=C:/usr/lib/libopenblas.lib
      [... other options]

Adjust the paths as needed.

Intel MKL

Intel’s oneAPI Math Kernel Library is distributed under the Intel Simplified Software License, also see license FAQ. To get started, use the following Nuget packages:

You can use those packages with CMake by specifying the BLA_VENDOR option, for example Intel10_64_dyn to link against the single, dynamic mkl_rt DLL. In case you are not using CMake, add the dependencies manually to your Visual Studio project. Always link mkl_core.lib. For 32-bit integer type add mkl_intel_lp64.lib, for 64-bit integers mkl_intel_ilp64.lib. Then add one of the following libraries:

  • mkl_sequential.lib for the sequential version
  • mkl_intel_thread.lib for the OpenMP version
  • mkl_tbb_thread.lib for the TBB version

The last two options will result in additional dependencies on Intel OpenMP or Intel TBB. You can also download a stand-alone installer, which might help integrating MKL into Visual Studio.

Note that Intel MKL libraries will work on AMD processors, but there have been reports of considerable slowdown (15-20%). Though workarounds exist, see for example https://danieldk.eu/Posts/2020-08-31-MKL-Zen, you might want to try the BLAS/LAPACK libraries provided by AMD, see next section.

AMD Optimizing CPU Libraries

The AMD Optimizing CPU Libraries (AOCL) are a set of (mostly) open source libraries compiled for AMD processors. The BLAS/LAPACK libraries are based on BLIS and libflame, both distributed under 3-clause BSD license.

You can use those packages with CMake by specifying the BLA_VENDOR option AOCL or AOCL_mt.

Building BLAS and LAPACK

Another option is to compile the libraries by yourself.

CLAPACK for Windows

A non-optimized version of BLAS and LAPACK is available at https://icl.cs.utk.edu/lapack-for-windows/clapack/index.html. This LAPACK version (3.2.1) was released in 2009, so it’s rather outdated and I recommend using one of the providers presented above. But it still might be useful to get started quickly without depending on foreign DLLs.

The FORTRAN code is already translated to C, so we can use Visual Studio to compile the sources. Download the source code archive (direct link) and extract it to a folder of your choice. We will use CMake, so if you don’t have it installed already, go to https://cmake.org/download/ and download the latest version. Make sure that cmake is in your PATH, open a terminal in the CLAPACK root folder and run the following build commands (adjust the install prefix):

cmake -B build -DBUILD_TESTING=OFF
      -DCMAKE_BUILD_TYPE=Release
      -DCMAKE_INSTALL_PREFIX="path/to/install"
cmake --build build
cmake --install build

This will build static libraries libf2c.lib, blas.lib and lapack.lib. f2c is the Fortran-to-C compiler that was used to translate the code. All three libraries have to be linked.

You can use those libraries with CMake by setting the following options:

cmake -DBLA_VENDOR=Generic
      -DBLA_STATIC=ON
      -DBLAS_LIBRARIES=C:/usr/lib/libf2c.lib;C:/usr/lib/blas.lib
      -DLAPACK_LIBRARIES=C:/usr/lib/libf2c.lib;C:/usr/lib/lapack.lib
      [... other options]

Adjust the paths as needed.

Please use the contact form for comments.