30 June 2009, 15:27

MilePost Compiler: AI optimises machine code

Milepost logo As part of the MilePost (Machine Learning for Embedded Programs Optimisation) project funded by Information Society Technologies (IST) of the European Union, the IBM research lab in Haifa, Israel, has released an open source compiler which automatically learns how to translate source code into machine code suitable for the respective hardware used. The compiler uses suggestions made by an ICI (Interactive Compilation Interface) plug-in.

Software developers can reportedly reduce their development times by up to a factor of ten using MilePost GCC. According to IBM, benchmarks carried out on a System p show that, on average, software compiled with the MilePost compiler also showed a performance increase of 18 per cent. The project, which was started in 2007, considers the technology especially suitable for embedded systems like mobile devices.

The GCC 4.4-based compiler is available to download from the web. Once the MilePost project is complete in August 2009, the software is to be maintained by its user community. In addition, the forthcoming version 4.5 of the GNU Compiler Collection is to have a dedicated plug-in interface for the MilePost extensions.

(crve)

« previous | next »

Print Version | Send by email | Permalink: http://h-online.com/-742269