Accelerator Exoskeleton

To maximize performance and power efficiency, future multi-core architectures may be heterogeneous, incorporating some accelerator cores alongside the IA cores. Accelerator Exoskeletons provide a shared virtual memory heterogeneous multi-threaded programming paradigm for these accelerators using novel CPU instruction set extensions and software tool chains with an Intel Architecture (IA) look-n-feel. Firstly, we introduce the proposed architectural extensions known as the Exoskeleton Sequencer (EXO), which represents heterogeneous accelerators as ISA-based MIMD architecture resources, and a shared virtual memory heterogeneous multi-threaded program execution model that tightly couples specialized accelerator cores with general-purpose CPU cores. Then we introduce the C for Heterogeneous Integration (CHI) programming environment that includes a compiler, runtime, debugger, and performance-analysis tools. The CHI compiler extends the OpenMP pragma for heterogeneous multi-threading programming, and it produces a single fat binary with code sections corresponding to different instruction sets. The runtime can judiciously spread parallel computation across the heterogeneous cores to optimize performance and power.