High-performance compute kernels, such as BLAS (Basic Linear Algebra Subprograms), are an essential component of high-performance computing software. This talk will focus on methods for evaluating and tuning the performance of compute kernels. Analytically determining the theoretical peak performance of a kernel for a given hardware platform, robustly measuring the performance of a kernel, and designing hybrid kernels which choose between multiple algorithms, will be discussed.

Lee Killough is a Senior Software Engineer in Technical Computing at Microsoft. Prior to Microsoft he was a developer of high-performance math libraries on the NEC SX vector supercomputer, Itanium and PA-RISC platforms.

Click on this link to add this event to your calendar.