Sushi JavaScript Library

The Fastest Matrix Library for JavaScript

Sushi is being developed to be the fastest matrix library for JavaScript. Tempura and Soba go well with Sushi.

• Tempura (Machine Learning Library) http://github.com/mil-tokyo/tempura
• Soba (Visualization Library) http://github.com/mil-tokyo/soba

Related papers are available ( http://mil-tokyo.github.io/miljs.html ).

Technical Features

Use of Typed Array

Faster than other matrix libraries which use javascript array.

Support of WebCL

You can use WebCL and get at most 30 times faster calculation speed. Three implementations

• Nokia WebCL for Firefox : http://webcl.nokiaresearch.com/
• Motorola-Mobility Node WebCL for node.js : https://github.com/mikeseven/node-webcl
• Chromium-WebCL : https://github.com/amd/Chromium-WebCL

are supported.

How to Try

node.js

cd test
node main

If you want to use WebCL, install node-webcl https://github.com/mikeseven/node-webcl

Currently, WebCL is supported by node v0.10.29, node-webcl v0.8.3 or v0.9.2

Try

npm install node-webcl

However, current nvidia driver only supports OpenCL 1.1, which is not compatible with node-webcl 0.9.0 or later. OpenCL 1.1 is supported by node-webcl v0.8.3, but unfortunately it has a small bug. So we provide bugfix version here: https://github.com/mil-tokyo/node-webcl/tree/opencl11_fix

Simple installation:

npm install https://github.com/mil-tokyo/node-webcl/archive/opencl11_fix.tar.gz

Browsers

Open test/main.html with your browser. Chromium-WebCL supports WebCL without any plugins. Firefox requires Nokia WebCL ( http://webcl.nokiaresearch.com/ ) to use WebCL with it.

Please Contribute to the Fastest Matrix Library

Contact me miljs@mi.t.u-tokyo.ac.jp