Advanced Static and Dynamic Scheduling of HPC Applications on Petascale Computer Systems with GPU Accelerators

What has been holding back General Purpose Graphics Processing Units (GPGPUs) until now has been programmability. At present, to take advantage of GPGPUs' potential, application developers must hand-optimize their code, which is expensive and difficult. This project will develop an advanced compiler that, when combined with a GPGPU, will enable the code optimization process to be conducted automatically, for a broad class of compute-intensive applications. Phase I demonstrated the automatic method for some important kernels. In Phase II, the scope of kernels that can be automated via GPGPU computing will be broadened, and the compiler solution will be completed, thereby dramatically reducing the cost of porting massively parallel applications to GPGPUs. Commercial Applications and other Benefits as described by the awardee: The technology should make supercomputing less expensive and much more accessible, facilitating scientific breakthroughs and economic growth. Commercial applications include molecular biology, drug development, cancer research, oil and gas exploration, global warming modeling, etc