|Name:||(05) Dynamic Parallelization of Computational Code as a Phase of Just-in-Time Compilation|
|Time:||Monday, June 23, 2014
04:50 pm - 04:57 pm
CCL - Congress Center Leipzig
|Speaker:||Artem Lebedev, Rybinsk State Aviation Technical University|
|Abstract:||A solution to improve portability of automatically parallelized code, extend applicability of polyhedron model, organize dynamic load balancing between host system and accelerator is reached via on-the-y transformations within JIT mechanisms of virtual execution system. An architecture of parallelizing optimizer module for ILDJIT is discussed. Transformed for heterogeneous execution LU-decomposition kernel illustrates significant speedup due to ooad acceleration.
Artem Lebedev, Rybinsk State Aviation Technical University