emlapack

Porting BLAS / LAPACK to JavaScript.

CLAPACK is compiled with emscripten.

Install

npm install emlapack

Example

Computing all eigenvalues of a real (upper triangular) symmetric matrix A

var emlapack = require('emlapack'),
    n = 5,
    dsyev = emlapack.cwrap('dsyev_', null, ['number', 'number', 'number', 'number', 'number', 'number', 'number', 'number', 'number']),
    pjobz = emlapack._malloc(1),
    puplo = emlapack._malloc(1),
    pn = emlapack._malloc(4),
    pa = emlapack._malloc(n * n * 8),
    plda = emlapack._malloc(4),
    pw = emlapack._malloc(n * 8),
    plwork = emlapack._malloc(4),
    pinfo = emlapack._malloc(4),
    pworkopt = emlapack._malloc(4);

emlapack.setValue(pjobz, 'V'.charCodeAt(0), 'i8');
emlapack.setValue(puplo, 'U'.charCodeAt(0), 'i8');
emlapack.setValue(pn, n, 'i32');
emlapack.setValue(plda, n, 'i32');
emlapack.setValue(plwork, -1, 'i32');

var a = new Float64Array(emlapack.HEAPF64.buffer, pa, n * n);
var w = new Float64Array(emlapack.HEAPF64.buffer, pw, n);
a.set([1.96, 0, 0, 0, 0, -6.49, 3.8, 0, 0, 0, -0.47, -6.39, 4.17, 0, 0, -7.2, 1.5, -1.51, 5.7, 0, -0.65, -6.34, 2.67, 1.8, -7.1]);

dsyev(pjobz, puplo, pn, pa, plda, pw, pworkopt, plwork, pinfo);

var workopt = emlapack.getValue(pworkopt, 'double'),
    pwork = emlapack._malloc(workopt * 8);
emlapack.setValue(plwork, workopt, 'i32');

dsyev(pjobz, puplo, pn, pa, plda, pw, pwork, plwork, pinfo);

console.log(w);
// { '0': -11.065575263268391,
//   '1': -6.22874693239854,
//   '2': 0.8640279752720624,
//   '3': 8.865457108365518,
//   '4': 16.09483711202934 }

Building

To build, you'll need to install and set up Emscripten, then download clapack into ./clapack.

First, configure clapack by creating make.inc:

$ cd clapack
$ cp make.inc.example make.inc

Next, follow the libf2c installation instructions. You'll need to tell it to create *.h from *.h0 files. The easiest way is to just make all:

$ cd F2CLIBS/libf2c
$ make all

Now from the main emlapack directory, proceed with gulp build:

$ gulp build

This should take a while. Coffee, perhaps?

Customizing Builds

Manual

You can modify src/export-functions.js with the functions you are using to reduce the bundle size significantly.

Custom

If the bundle is too large and you know the specific set of functions you would like to use, you can create a file called custom/export-functions.js which exports the list of functions you want.

You could copy the src/export-functions.js file to the custom directory and modify it to your needs.

The custom folder also contains a export-functions.example.js you could copy and rename to export-functions.js and modify.

Docker

For convenience, this repo also includes a Dockerfile so you can build everything in Docker.

If you need a custom build, just add your export-functions.js file to the custom folder and execute the scripts/docker_build.sh script. All it does is build the container and then runs it.

NOTE: Make sure you run this script in the project root directory.

./scripts/docker_build.sh

After the script completes, you will see asmjs.js, emlapack.wasm, and wasm.js outputted to the root of the project directory.

References

For more detail, see:

• http://www.netlib.org/blas/
• http://www.netlib.org/lapack/
• http://emscripten.org/

License

© 2015 Yosuke Onoue, BSD-3-Clause.