Even faster multi-core processor calculations

Grenoble-based engineers from CEA-List have more than doubled the speed of a matrix multiplication program using a new dynamic compilation method. This method, designed specifically for multi-core processors, works by optimizing the execution of the most calculation-intensive sections of code. It offers an attractive alternative to static compilation, which optimizes the overall code using conservative options.

Today the engineers perform the dynamic compilation manually based on an expert analysis. But they believe that it can be fully automated within the next five years. Applications for their method include multi-core embedded systems and high-performance calculators—unleashing the full potential of such devices.

Contact: henri-pierre.charles@cea.fr
 

Facebook

Twitter