Automatic Discovery of Coarse-Grained Parallelism in Media Applications

Abstract

Abstract. With the increasing use of multi-core microprocessors and hardware accelerators in embedded media processing systems, there is an increasing need to discover coarse-grained parallelism in media applications written in C and C++. Common versions of these codes use a pointer-heavy, sequential programming model to implement algorithms with high levels of inherent parallelism. The lack of automated tools capable of discovering this parallelism has hampered the productivity of parallel programmers and application-specific hardware designers, as well as inhibited the development of automatic parallelizing compilers. Automatic discovery is challenging due to shifts in the prevalent programming languages, scalability problems of analysis techniques, and the lack of experimental research in combining the numerous analyses necessary to achieve a clear view of the relations among memory accesses in complex programs. This paper is based on a coherent prototype system designed to automatically find multiple levels of coarse-grained parallelism. It visits several of the key analyses that are necessary to discover parallelism in contemporary media applications, distinguishing those that perform satisfactorily at this time from those that do not yet have practical, scalable solutions. We show that, contrary to common belief, a compiler with a strong, synergistic portfolio of modern analysis capabilities can automatically discover a very substantial amount of coarse-grained parallelism in complex media applications such as an MPEG-4 encoder. These results suggest that an automatic coarse-grained parallelism discovery tool can be built to greatly enhance the software and hardware development processes of future embedded media processing systems.

VEBoC: Variation and error-aware design for billions of devices on a chip

“…Extracting and mapping these coarse level tasks in different cores of the system can lead to remarkable improvements in performance and efficiency of the application execution. During recent and ongoing work, the IMPACT compiler is enhanced with a strong synergistic portfolio of different analysis techniques that can help expose the coarse-grain parallelism [21] 9D-2 today. In particular, a combination of different analysis techniques focusing on 1) pointers, 2) arrays and structures and 3) variable constraints and relationships, are effectively applied to discover different forms of coarse-grain parallelism.…”

Implicitly parallel programming models for thousand-core microprocessors

“…The programmer, who understands the high-level parallelism, can then assert that the two frames do not overlap (user assert), allowing the Concurrency Discovery module to determine that the loop is parallelizable. Previous work [17] has discussed the battery of analyses that can be used to expose the parallelism in an MPEG-4 encoder, including a context sensitive, interprocedural pointer analysis capable of distinguishing among heap objects allocated from different malloc call chains, the Omega test [16], and accurate analysis of non-affine index expressions for cross-iteration dependences. Mileage can vary significantly across different types of analysis techniques.…”

Commutativity analysis for software parallelization

“…In contrast to this popular opinion, several recent papers [2,18,25] have shown that recent advances in compiler analysis and hardware support for speculation have enabled compiler-based parallelization to be surprisingly effective in domains conventionally thought beyond its reach. For example, Bridges et al [2] report 454% average speedup over single threaded code on the SPECint2000 benchmarks, using a mixture of automatic parallelization with small code annotations.…”