X then as mpi reference implementations v2x then

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ed performance CSEP 524: Parallel ComputaIon Winter 2013: Chamberlain 13 Beyond MPI- 1 MPI- 2 (1990’s): –  support for coordinated parallel I/O –  (poor support for) single- sided communicaIon* –  dynamic process creaIon (“add a new MPI rank now”) MPI- 3 (circa 2012): –  bemer support for single- sided communicaIon –  bemer support for mulIthreading within MPI –  acIve messages*, ** –  bemer support for GPUs/accelerators** –  bemer compiler support for MPI** * = we’ll be defining this term next week; ** = a work- in- progress? CSEP 524: Parallel ComputaIon Winter 2013: Chamberlain 14 The Stencil Ramp: (a series of increasingly complex Stencil- based algorithms) Stencils we have known Stencils we have known (and loved!): –  The 3- point stencil A: –  The Jacobi iteraIon n 1.0 –  The 9- point stencil from homework Some more advanced uses of stencils –  The MulIgrid method (MG) –  The Fast MulIpole Method (FMM) CSEP 524: Parallel ComputaIon Winter 2013: Chamberlain 16 A Distributed Memory Algorithm: The Mul3grid Method (as told through the NAS MG benchmark) NAS Parallel Benchmarks (NPB) •...
View Full Document

Ask a homework question - tutors are online