Translating GPU Binaries to Tiered SIMD Architectures with Ocelot

Parallel Thread Execution ISA (PTX) is a virtual instruction set used by NVIDIA GPUs that explicitly expresses hierarchical MIMD and SIMD style parallelism in an application. In such a programming model, the programmer and compiler are left with the not trivial, but not impossible, task of composing applications from parallel algorithms and data structures. Once this has been accomplished, even simple architectures with low hardware complexity can easily exploit the parallelism in an application. With these applications in mind, this paper presents Ocelot, a binary translation framework designed to allow architectures other than NVIDIA GPUs to leverage the parallelism in PTX programs. Specifically, we show how (i) the PTX thread hierarchy can be mapped to many-core architectures, (ii) translation techniques can be used to hide memory latency, and (iii) GPU data structures can be efficiently emulated or mapped to native equivalents. We describe the low level implementation of our translator, ending with a case study detailing the complete translation process from PTX to SPU assembly used by the IBM Cell Processor.