mcanthony/cl.js

A lightweight middleware library for WebCL.

CL.js

CL.js is a lightweight utility library for WebCL. It relieves the developer from writing most of the boilerplate code that is otherwise required to get anything running, while preserving the look and feel of WebCL and providing easy access to the underlying raw API. See http://toaarnio.github.io/CL.js for documentation.

Please visit http://webcl.nokiaresearch.com for demos, tutorials, and more information on WebCL.

Example

To get started, we need to include CL.js from our HTML source:

<script type="text/javascript" src="/path/to/CL.js"></script>

Now, let's initialize the global CL object and populate it with information about the WebCL platforms and devices that are available in this system:

CL.setup({ debug: true });

With debug mode enabled, any exceptions will be reported on the console together with helpful information about the function and arguments that triggered the exception. Next, let's create the CL resources that we're going to need in this example:

var src = CL.loadSource('kernels/random.cl');
for (var d=0; d < CL.devices.length; d++) {
  var ctx = CL.createContext({ device: CL.devices[d], name: 'device'+d });
  var buffer1M = ctx.createBuffer({ size: 1024*1024, name: 'results' });
  var kernel = ctx.buildKernel({ source: src });
  kernel.setArgs(buffer1M);
  ctx.createCommandQueue({ name: 'theQueue' });
}

Note that we assigned a plain-text name for each of our contexts, buffers and queues. This makes it easy to find the resources that we need later on. The Kernel object that we created for each Context is assigned the same name as the kernel function in random.cl. For the purposes of this example, let's assume that the kernel function is called generateRandomNumbers. Now, let's proceed to run it on each Device in turn, reading back the results into an ArrayBuffer:

var randomNumbers = new Uint8Array(1024*1024);
for (var d=0; d < CL.devices.length; d++) {
   var ctx = CL.getContext('device'+d);
   var queue = ctx.getQueue('theQueue');
   var kernel = ctx.getKernel('generateRandomNumbers');
   var buffer1M = ctx.getBuffer('results');
   queue.enqueueKernel(kernel, [randomNumbers.length]);
   queue.enqueueReadBuffer(buffer1M, randomNumbers);
   queue.finish();
   /* Do something with randomNumbers */
}