EXOCHI

摘要: Future mainstream microprocessors will likely integrate specialized accelerators, such as GPUs, onto a single die to achieve better performance and power efficiency. However, it remains keen challenge program heterogeneous multicore platform, since these accelerators feature ISAs functionality that are significantly different from the general purpose CPU cores. In this paper, we present EXOCHI: (1) Exoskeleton Sequencer(EXO), an architecture represent acceleratorsas ISA-based MIMD resources, shared virtual memory multithreaded execution model tightly couples accelerator cores with generalpurpose cores, (2) C for Heterogeneous Integration(CHI), integrated C/C++ programming environment supports accelerator-specific inline assembly domain-specific languages. The CHI compiler extends OpenMP pragma multithreading programming, produces fat binary code sections corresponding instruction sets. runtime can judiciously spread parallel computation across optimize power.We have prototyped EXO on physical platform consisting of Intel® Core™ 2 Duo processor 8-core 32-thread Graphics Media Accelerator X3000. addition, implemented C++ Compiler, toolset, debugger. On prototype system, enhanced suite production-quality media kernels video image processing utilize through interface, achieving significant speedup (1.41X to10.97X) over IA32 alone.