Runtime library for parallel programming in MPPA-256 manycore processor

Abstract

The demand for high processing power together with the actual concern about power consumption has led the industry towards heterogeneous solutions, which present a good compromise between these two factors. The challenge involving this approach relies on how to develop efficient and portable applications to take advantage of the parallelism available in such architectures. Some software solutions appeared to help programmers developing code for these new class of heterogeneous processors. These solutions aim to provide an easy and flexible way to express parallelism in the applications. Some solutions such as CUDA and OpenCL present a software model and architecture that has proved to be very effective for general purpose programming on these devices. Among the architectures that have been used for heterogeneous programming, manycore processors and CPUs combined with GPGPUs stand out as the most expressive solutions, obtaining the best results. The focus of this work is one specific solution, the MPPA R -256 manycore processor, which targets high parallel processing and low power consumption. This processor integrates 256 calculating cores for general purpose programming providing good processing capabilities. As any other solution for heterogeneous computing available in the market, porting applications and developing new ones for this processor can be difficulty. As MPPA processor has unique hardware and software architectures, it increases the complexity of programming on it. The focus of this work is to study and analyze the viability of implementing a runtime to be linked to the MPPA processor, with some high level functions aiming to help programmers with the task of easily setting up the programming context and necessary features to produce software in a faster way. This study contemplates the implementation and tests realized to validate the proposed solution.