On Tiling Space-Time Mapped Loop Nests [Extended Abstract]
Download:
|
|
by Martin Griebl
http://infosun.fmi.uni-passau.de/cl/papers/Gri01spaa.ps.gz
Add To MetaCart
Abstract:
Automatic parallelization of loop programs based on spacetime mapping typically aims at maximal parallelism. In order to obtain reasonable performance, the granularity of the parallelism must be coarsened, e.g., by applying tiling techniques. In contrast to others, we suggest to apply tiling after the computation of a space-time mapping. This extends the applicability of existing tiling methods and signicantly improves eciency of parallelized loop programs. 1.
Citations
| 204 | Supernode partitioning – Irigoin, Triolet - 1988 |
| 71 | Loop parallelization in the polytope model – Lengauer - 1993 |
| 39 | Synthesizing transformations for locality enhancement of imperfectly-nested loop nests – Ahmed, Mateev, et al. - 2000 |
| 38 | Automatic parallelization in the polytope model – Feautrier - 1996 |
| 10 | On the parallelization of loop nests containing while loops – Lengauer, Griebl - 1995 |
| 8 | A compiler framework for tiling imperfectly-nested loops – Song, Li - 1999 |
| 6 | Optimal semi-oblique tiling – Andonov, Balev, et al. - 2001 |
| 2 | On tiling space-time mapped loop nests – Griebl - 2001 |
| 1 | Minimal number of communication startups by tiling space-time mapped programs. Submitted. Preliminary version available at http://www.fmi.unipassau. de/cl/papers/minstart.html – Griebl |
