Advanced Compiler Technologies - Exascale
As the Petascale supercomputing milestone has been reached first by the Cell-based Roadrunner supercomputer, we are now working on steps to achieve the next milestone, the Exascale computing milestone.
It is no secret that to achieve such high rates of computation, future computers will need to efficiently exploit massive amount of parallelism at all levels of the hierarchy, from concurrently running applications loosely coupled in a multi-model simulation, to massively parallelized applications, possibly using MPI on top of OpenMP. In addition to traditional parallelism, various forms of speculative executions may also play a significant role, from computational speculation to memory-level speculation. Finally, thread-level, SIMD-level, and instruction level parallelism will need to deliver the actual performance within strict power budget.
We are actively evaluating approaches to reach the Exascale Milestone based on compiler analysis and optimizations for the benchmark suites that will be relevant for such platforms.