10:15 Ashay Rane
Title: Unification of static analyses and runtime measurements for improving vectorization
Modern compilers execute extremely sophisticated static analyses to enable to optimize across a wide spectrum of code patterns. However, there are many cases where even the most sophisticated static analysis is insufficient, or where the computation complexity makes complete static analysis impractical. It is often possible to leverage runtime measurements to discover further opportunities for optimization. This session will describe our work centered around gathering relevant dynamic profiling metrics using a low-overhead profiling system and feeding these details back to the compiler in terms of specific pragmas and options. The study demonstrates that opportunities for additional vectorization are frequent and the performance gains obtained are substantial. For the applications used in our study, we observed a geometric mean of 1.5x in speedup on the Intel Xeon Phi coprocessor.
10:37 Carson Brownlee
Title: Ray Tracing using Intel’s Embree for Scientific Visualization
In this talk I will cover performance of Intel’s Embree ray tracer and implications for large scale scientific computing. Ray tracing provides many benefits for scientific visualization for both performant image generation and physically based analysis. Simulations across many domains continue to produce ever expanding amounts of data and acceleration methods present in ray tracing methods provide a straightforward path for handling large amounts of geometry efficiently. By tracing light rays, we can efficiently replicate many visualization techniques that require physically based computation of light paths.