Parallel convolution gridding for Radio Astronomy applications running on KNL and GPU

Oxford e-Research Centre
June 22, 2017 - 13:00 to 14:00
Access Grid Room (277)

7 Keble Road, Oxford, OX1 3QG

  • Seminar
  • No booking required
  • Open to all
  • Many-Core Series
  • Lunch provided

 

Abstract
We present results from recent joint work with the radio astronomy department at Oxford University. The work has relevance to the Square Kilometer Array (SKA). Convolution gridding is one of the first stages in processing raw radio astronomy observations (visibilities) into usable images. The algorithm is totally bandwidth bound and full of stochastic race conditions. We present optimised implementations of the algorithm developed for multicore x86, KNL and P100 GPUs. Through a combination of tiling the grid, bucket sorting the data and keeping local data in cache/registers we manage to obtain satisfying speedups on all platforms over the original serial code. The KNL is consistently the worst performer due to small shared caches, while the P100 is consistently the fastest.

About the speaker
Jacques du Toit studied actuarial science and statistics before seeing the light and switching to probability theory and financial mathematics. He obtained a PhD from Manchester University and joined the Numerical Algorithms Group in 2010 to work on GPU programming.    These days his interests are in multicore/manycore programming, adjoint algorithmic differentiation, code generation, and the combination of all these.