Automatic parallelization of loop
- Automatic generation of parallel (multithreaded) code for specific loop constructions, using the gomp library.
- Zdenek Dvorak
- Sebastian Pop
- Diego Novillo (reviewer)
- The first version of the code is finished, allowing parallelization of simple loops (that however appear quite frequently in practice), and tested on gcc bootstrap and microbenchmarks. Further changes are planned to be done incrementally, once this basic framework is merged.
- Performance gains on multiprocessor machines.
- The implementation is basically self-contained, requiring almost no changes to unrelated code (except for several improvements to the code shared with Open MP implementation).