Tuesday, June 30, 2009

Milepost GCC machine learning

Developing code for embedded programs just got a little more easy. When you develop embed software most likely you will have to face things like performance and how much power the device will consume. Tuning such code can be very very complicated. You will have to study the device and your code. Developing a first beta of a embedded program which is not consuming to much power and is performing in a good fashion can take up to months.

Now IBM Research and Milepost Consortium have developed a new GCC open source machine learning compiler. The compiler will optimize your code and make shorter and better performing code so you can shorten your development cycle.

Besides having your compiler learn from your own devices and tricks you can also use the ctuning.org website to collaborate and upload code. This way you will get automatically input on how to improve your code.

If you are a developer on such code,….. or looking into the latest DARPA challenge you might want to give this a go.

Post a Comment