next up previous
Next: About this document ...

Sherry Li
Performance Analysis of Coarse Solvers for Algebraic Multigrid on Leading Multicore Architectures

Lawrence Berkeley Lab
One Cyclotron Road
MS 50F-1650
Berkeley
CA 94720
xsli@lbl.gov
Alex Druinsky
Pieter Ghysels
Osni Marques
Sam Williams
Andrew Barker
Panayot Vassilevski

We present performance of a novel multigrid code using higher order elements to discretize the PDEs and element-agglomeration AMGe methods which improve the coarsening process over the traditional AMG and have guaranteed approximation property. Another novelty is to use a randomized HSS low-rank sparse factorization method as a coarse grid solver, which provides powerful and flexible coarse solver for numerically challenging problems.

In addition to scaling study, we present a comparative analysis of the impact of the coarse grid solver on the AMG performance. Here, we compare the PCG method and the HSS sparse factorization method. We contrast the performance of our solver on one 12-core socket of a Cray XC30 machine with performance on a 60-core Intel Xeon Phi coprocessor. To obtain peak performance, we optimized the code to take full advantage of fine-grained parallelism and made it thread-friendly for high thread count. We also developed a bounds-and-bottlenecks performance model of the solver which we used to guide us through the optimization effort. We carried out performance tuning in the solver's large parameter space. As a result, signicant speedups were obtained on both machines.




next up previous
Next: About this document ...
root 2016-02-22