Automatically exploiting implicit parallelism in Java

Abstract

In this paper we show how implicit parallelism in Java programs can be made explicit by a restructuring compiler using the multi‐threading mechanism of the language. In particular, we focus on automatically exploiting implicit parallelism in loops and multi‐way recursive methods. Expressing parallelism in Java itself clearly has the advantage that the transformed program remains portable. After compilation of the transformed Java program into byte‐code, speedup can be obtained on any platform on which the Java byte‐code interpreter supports the true parallel execution of threads. Moreover, we will see that the transformations presented in this paper only induce a slight overhead on uni‐processors. © 1997 John Wiley & Sons, Ltd.

Automatic Parallelization and Optimization of Programs by Proof Rewriting

“…Practical approaches for parallelizing programs include parallelizing compilers [9,3]. Parallelizing compilers focuses on loop parallelization and do not consider arbitrary pieces of code.…”

JavaSpMT: A speculative thread pipelining parallelization model for Java programs

“…Exceptions that may be thrown during the parallel execution of the JavaSpMT code are handled by catching the exception in the corresponding Java thread's run method [1]. This caught exception may be thrown back to the main thread of execution if necessary.…”

“…JAVAR parallelization tool. JAVAR [1] parallelizes serial Java programs using standard Java multithreading and synchronization primitives. It can automatically parallelize loops with static data-dependences as well as multiway recursive methods.…”

“…Other parallelization tools. Among the different existing Java parallelization techniques [1,2,3,6,7,8,10,11], High Performance Java [1], Do! [10], and the Tiny Data-Parallel language [6] target parallelization on shared-memory architectures.…”

“…The other approaches [2,3,7,8,11] attempt to improve performance of Java programs in a distributed en- vironment. JavaSpMT targets shared-memory architectures and is similar to the JAVAR tool [1] in that the parallelization is achieved by a Java source code transformation using Java multithreading and a library that is implemented entirely in Java. However, the speculative execution combined with run-time data-dependence checking make our approach unique.…”