Automatic extraction of heterogenous thread-level parallelism

Affiliated to

Location

Timing

Description

Automatic extraction of heterogenous thread-level parallelism. We would like to explore new ways to automatically extract parallelism and to compile a single source program into multiple threads, each running a separate 'program' on a different processor/ISA (in contrast to SPMD), in the context of GCC. Additional related aspects involve optimizing data layout and data-level parallelism.