high performance computing on graphics processing units: hgpu.org

The advent of multicore systems, joined to the potential acceleration of the graphics processing units, has given us a low cost computation capability unprecedented. The new systems alleviate some well known important architectural problems at the expense of a considerable increment of the programmability wall. The heterogeneity, both at architectural and programming level at the same time, raises the programming difficulties. As a contribution in this context, we propose a development methodology for the automatic source-to-source transformation on specific domains. This methodology is successfully instantiated as a framework to solve Dynamic Programming problems. As a result of applying our framework, the end user (a physicist, a mathematician or a biologist) can express her problem through a latex equation and automatically derive efficient parallel codes for current homogeneous or heterogeneous architectures. The approach allows an easy portability to new potential emergent architectures.