Abstract | ||
---|---|---|
We present Bamboo, a custom source-to-source translator that transforms MPI C source into a data-driven form that automatically overlaps communication with available computation. Running on up to 98304 processors of NERSC's Hopper system, we observe that Bamboo's overlap capability speeds up MPI implementations of a 3D Jacobi iterative solver and Cannon's matrix multiplication. Bamboo's generated code meets or exceeds the performance of hand optimized MPI, which includes split-phase coding, the method classically employed to hide communication. We achieved our results with only modest amounts of programmer annotation and no intrusive reprogramming of the original application source. |
Year | DOI | Venue |
---|---|---|
2012 | 10.1109/SC.2012.23 | SC |
Keywords | DocType | ISSN |
matrix multiplication,available computation,application program interfaces,nersc hopper system,c language,98304 processor,mpi application,language translation,mpi implementation,multiprocessing systems,mpi c source,cannon matrix multiplication,latency tolerant,custom source-to-source translator,split phase coding,data-driven form,hopper system,code meet,c source transform,message passing,mpi application translation,jacobian matrices,capability speed,jacobi iterative solver,iterative methods,data driven form,3d jacobi iterative solver,bamboo generated code,translating mpi application,overlaps communication,original application source | Conference | 2167-4329 |
ISBN | Citations | PageRank |
978-1-4673-0805-2 | 11 | 0.65 |
References | Authors | |
20 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Tan Nguyen | 1 | 61 | 5.22 |
Pietro Cicotti | 2 | 101 | 14.52 |
Eric Bylaska | 3 | 11 | 0.65 |
Dan Quinlan | 4 | 54 | 5.12 |
Scott B. Baden | 5 | 519 | 67.52 |