OpenBlasSharp

The purpose of this project is to provide a .NET wrapper for OpenBLAS.

Features

• Auto-generated thin wrapper for C# with no special marshalling.
• Covers BLAS, LAPACK, and OpenBLAS-specific functions.
• Most functions and arguments are annotated with doc comments taken from the original FORTRAN code. This is very helpful when working with the BLAS and LAPACK functions, which often require a large number of arguments.

Installation

The NuGet package is available.

Install-Package OpenBlasSharp

This package does not include the native binary. The OpenBlasSharp.Windows package provides the native binary for Windows.

Install-Package OpenBlasSharp.Windows

Or, download the compiled binary and put libopenblas.dll in the same directory as the executable file. Binaries for both x86 and x64 architectures are supported, but the ILP64 build with the x64-64 suffix is not supported.

All the classes are in the OpenBlasSharp namespace.

using OpenBlasSharp;

BLAS Examples

BLAS functions are provided as static methods of the Blas class.

Dot Product

var len = 3;

var rnd = new Random(42);
var x = Enumerable.Range(0, len).Select(i => rnd.NextDouble()).ToArray();
var y = Enumerable.Range(0, len).Select(i => rnd.NextDouble()).ToArray();

fixed (double* px = x)
fixed (double* py = y)
{
    // Calculate x^T * y.
    var result = Blas.Ddot(len, px, 1, py, 1);
}

Matrix Multiplication

var m = 3;
var n = 5;
var k = 4;

var rnd = new Random(42);
var a = Enumerable.Range(0, m * k).Select(i => rnd.NextDouble()).ToArray();
var b = Enumerable.Range(0, k * n).Select(i => rnd.NextDouble()).ToArray();
var c = new double[m * n];

fixed (double* pa = a)
fixed (double* pb = b)
fixed (double* pc = c)
{
    // Calculate c = a * b.
    Blas.Dgemm(
        Order.ColMajor,
        Transpose.NoTrans,
        Transpose.NoTrans,
        m, n, k,
        1.0,
        pa, m,
        pb, k,
        0.0,
        pc, m);
}

LAPACK Examples

LAPACK functions are provided as static methods of the Lapack class.

Inverse Matrix using LU Decomposition

var n = 3;

var random = new Random(42);
var a = Enumerable.Range(0, n * n).Select(i => random.NextDouble()).ToArray();
var piv = new int[n];

fixed (double* pa = a)
fixed (int* ppiv = piv)
{
    Lapack.Dgetrf(
        MatrixLayout.ColMajor,
        n, n,
        pa, n,
        ppiv);

    Lapack.Dgetri(
        MatrixLayout.ColMajor,
        n,
        pa, n, // pa will be the inverse matrix.
        ppiv);
}

Singular Value Decomposition

var m = 4;
var n = 3;

var rnd = new Random(42);
var a = Enumerable.Range(0, m * n).Select(i => rnd.NextDouble()).ToArray();
var s = new double[Math.Min(m, n)];
var u = new double[m * m];
var vt = new double[n * n];
var work = new double[Math.Min(m, n) - 1];

fixed (double* pa = a)
fixed (double* ps = s)
fixed (double* pu = u)
fixed (double* pvt = vt)
fixed (double* pwork = work)
{
    Lapack.Dgesvd(
        MatrixLayout.ColMajor,
        'A', 'A',
        m, n,
        pa, m, // Note that pa will be destroyed.
        ps,
        pu, m,
        pvt, n,
        pwork);
}

OpenBLAS-specific functions

OpenBLAS-specific functions are provided as static methods of the OpenBlas class.

var numThreads = OpenBlas.GetNumThreads();
var numProcs = OpenBlas.GetNumProcs();
var config = OpenBlas.GetConfig();

Limitations

• There are no plans to support low-level LAPACK functions with the work suffix.
• LAPACK functions that require function pointers are currently not supported, but support is planned for the future.
• The doc comments in the original FORTRAN code do not completely align with the C API of OpenBLAS, so the annotations for some APIs are incomplete.

License

OpenBlasSharp is available under the BSD-3-Clause license.