Gosl – Go scientific library
Gosl is a set of tools for developing scientific simulations using the Go language. We mainly consider the development of numerical methods and solvers for differential equations but also present some functions for fast Fourier transforms, the generation of random numbers, probability distributions, and computational geometry.
This library contains essential functions for linear algebra computations (operations between all combinations of vectors and matrices, eigenvalues and eigenvectors, linear solvers) and the development of numerical methods (e.g. numerical quadrature).
We link Gosl with existent libraries written in C and Fortran, such as OpenBLAS, LAPACK, UMFPACK, MUMPS, QUADPACK and FFTW3. These existing libraries have been fundamental for the development of high-performant simulations over many years. We believe that it is nearly impossible to rewrite these libraries in native Go and at the same time achieve the same speed delivered by them. Just for reference, a naive implementation of matrix-matrix multiplication in Go is more than 100 times slower than OpenBLAS.
Because of CGO and other libraries, the easiest way to work with Gosl is via Docker. Having Docker and VS Code installed, you can start developing powerful numerical simulations using Gosl in a matter of seconds. Furthermore, the best part of it is that it works on Windows, Linux, and macOS out of the box.
Quick, containerized (recommended)
- Install Docker
- Install Visual Studio Code
- Install the Remote Development extension for VS Code
- Clone https://github.com/cpmech/hello-gosl
- Create your application within a container (see gif below)
Done. And your system will remain "clean."
Our Docker Image also contains Go and the Go Tools for working with VS Code (or not). Below is a video showing the convenience of VS Code + the Go tools + Gosl. Note how fast VS Code is in finding the function ReadLines and the package gosl/io even under a clash with Go's io package. Upon file save, the Go tools automatically add the required imports. Note also the very convenient auto-completion of the callback function given to ReadLines. Also, Code + the Go tools nicely fill the function arguments with default values.
Another great thing about VS Code is the IntelliSense. Here, as soon as we start typing "m comma n two-dot equal T dot", Code immediately offers Size() as the first option because it matches the preceding code. Fantastic!
Debian/Ubuntu GNU Linux
Because we use CGO for linking Gosl with many libraries, we cannot use the so convenient go get functionality for installing Gosl. Moreover, we view Gosl as the most basic set of libraries for high-performance computing and therefore prefer to install Gosl directly alongside Go. In other words, Gosl extends Go with powerful tools for scientific simulations.
Gosl is then linked to
/usr/local/go/src/gosl. We have experimented with GOPATH and the newer Go Modules approach, but both do not work well with CGO (and hence Gosl).
Download and link Gosl
Assuming that you have saved your go code in
$HOME/mygo and that go has been installed in
git clone https://github.com/cpmech/gosl.git $HOME/mygo/gosl ln -s $HOME/mygo/gosl $HOME/go/src/gosl
sudo apt-get install -y --no-install-recommends \ gcc \ gfortran \ libopenmpi-dev \ libhwloc-dev \ liblapacke-dev \ libopenblas-dev \ libmetis-dev \ libsuitesparse-dev \ libmumps-dev \ libfftw3-dev \ libfftw3-mpi-dev
cd $HOME/mygo/gosl bash ./all.bash
Done. Installation completed.
Gosl includes the following essential packages:
- chk. To check numerical results and for unit testing
- io. Input/output including printing to the terminal and handling files
- utl. To generate series (e.g. linspace) and other functions as in pylab/matlab/octave
Gosl includes the following main packages:
- fun. Special functions, DFT, FFT, Bessel, elliptical integrals, orthogonal polynomials, interpolators
- gm. Geometry algorithms and structures
- hb. Pseudo hierarchical binary (hb) data file format
- la. Linear Algebra: vector, matrix, efficient sparse solvers, eigenvalues, decompositions
- mpi. Message Passing Interface for parallel computing
- num. Fundamental numerical methods such as root solvers, non-linear solvers, numerical derivatives and quadrature
- ode. Solvers for ordinary differential equations
- opt. Numerical optimization: Interior Point, Conjugate Gradients, Powell, Grad Descent
- pde. Solvers for partial differential equations (FDM, Spectral, FEM)
- rnd. Random numbers and probability distributions
(see each subdirectory for more information)
Please check out https://github.com/cpmech/gosl-examples
Please check out https://github.com/cpmech/gosl-benchmarks
These other packages, such as machine learning, plotting, etc., have been removed because they do not depend on CGO and may be developed independently. We can now maintain the core of Gosl more efficiently, which has a focus on the foundation for other scientific code.