Title: The PGI Accelerator Programming Model: Design and Implementation Speaker: Michael Wolfe, PGI Compiler Engineer and Brent Leback, PGI Engineering Manager Date/Time: Wednesday, March 18, 2009, 1:30 pm Location: CSRI Building, Room 90 (Sandia NM) Brief Abstract: PGI has developed a Kernels programming model for accelerators, such as GPUs, where a kernel roughly corresponds to a set of compute-intensive parallel loops with rectangular limits. We have designed directives for Fortran and C to target this model, similar in design to the well-known and widely used OpenMP directives. We are implementing the directives and programming model in the PGI Fortran and C compilers initially to target X64 + NVIDIA platforms. In this talk, we describe the target system architecture. We introduce the Kernels programming model and show how it maps to the target system. We describe the directives and our experiences using them to port programs to accelerators. We also discuss the generality and limitations of the programming model, and how it can be mapped onto more general target systems.CSRI POC: Doug Doerfler, (505) 844-9528 |