Implementation And Optimization Of OpenMP On Heterogeneous Mutil-core Processor

Cell processor is a representative of high performance heterogeneous multi-core processor today. Its programming model and compilation technique has become an important research topic recently. The OpenMP specification as a predominant parallel program language on shared memory systems supports incremental development of parallel application. Therefore providing OpenMP compilation environment for Cell processor could make it easy to program and improve programming productivity,We discusse the key points of implementing OpenMP compiler on a heterogeneous multi-core processor, and break through in OpenMP execution model, single source compilation, and shared memory abstract for OpenMP compilation techniques on Cell processor. We implement an OpenMP C compiler for Cell processor based on the GNU OpenMP compiler, and verify the feasibility of porting an OpenMP compiler to a heterogeneous multi-core processor.We put two optimization techniques software cache and code overlay in the compiler for high performance. We design and implement a software cache supporting data consistency which reduces the latency of OpenMP shared data access saliently. We also design and implement a practical automatic code partition algorithm which reduces the swap of code blocks and resolves the overflow problem of OpenMP parallel code on Cell's SPE effectively.Finally, we test the compiler's function and performance on Cell processor. The result indicates that it supports the OpenMP C syntax in good completeness and provides good speedup on Cell processor. We find it fully utilizing the parallel computing ability of Cell processor while coming up with good programmability.