|Name:||Exploiting SIMD & Thread-Level Parallelism in Multiblock CFD|
|Time:||Wednesday, June 25, 2014
10:15 am - 10:30 am
CCL - Congress Center Leipzig
|Breaks:||10:30 am - 11:30 am Coffee Break|
|Speaker:||Ioan Hadade, Imperial College London|
|Abstract:||This paper presents the on-node performance tuning of a multi-block Euler solver for turbomachinery computations. Our work focuses on vertical and horizontal scaling within an x86 multi-socket compute node by exploiting the fine grained parallelism available through SIMD instructions at core level and thread-level parallelism across the die through shared memory. We report on the challenges encountered in enabling effcient vectorization using both compiler directives and intrinsics with an emphasis on the data structure transformations needed for obtaining fully vectorized stencil computations within the solver's fluxes reconstruction phase. Finally, we present the solver performance on different grid fidelity levels running on Intel Sandy Bridge and Ivy Bridge processors.
Ioan Hadade & Luca di Mare, Imperial College London