POSC—a partitioning and optimizing SISAL compiler

Abstract

Single-assignment languages like SISAL offer parallelism at all levels—among arbitrary operations, conditionals, loop iterations, and function calls. All control and data dependencies are local, and can be easily determined from the program. Various studies of SISAL programs have shown that they contain massive amounts of potential parallelism . There are two major challenges in converting this potential parallelism into real speedup on multiprocessor systems. First, it is important to carefully select the useful parallelism in a SISAL program, so as to obtain good speedup by trading off parallelism with overhead. Second, it is important to do sequential optimizations , so that the sequential components (tasks) of the SISAL program have comparable execution times with sequential languages such as Fortran, Pascal and C. The POSC compiler system described in this paper addresses both issues by integrating previous work on efficient sequential implementation of SISAL programs with previous work on selecting the useful parallelism in a SISAL program. The combined approach is validated by real speedup measurements on a Sequent Balance multiprocessor.

Predictability of load/store instruction latencies

“…The Partitioning and Optimizing SISAL compiler [SC90] uses profiling to estimate the complexity of large, variable sized, chunks of work and partitions the chunks in order to achieve optimal run time. This is different from the work presented here as we partition the code into medium "fixed" sized clusters.…”

Compiler Techniques for Massively Scalable Implicit Task Parallelism

“…Hardware data-flow-based languages and execution models received significant attention in the past, but there is a resurgence of interest in hardware-based [15], [23] and softwarebased [6], [7], [10], [25], [29] dataflow models because of their ability to expose parallelism, mask latency, and enable fault tolerance. Previous work has optimized data flow languages with arrays: SISAL [26] and Id [32]. Both have similarities to Swift/T; but emphasize generating low-level machine code.…”

Paralelização parcial de programas SISAL utilizando a biblioteca MPI

“…O método de particionamento escolhido foi o s/ice dos /oops [oral/ [SC90] apenas no seu nível mais externo. O método slice consiste na divisão de todas as os instânci:.s do }oral/ em grupos de instâncias consecutivas, de modo a formar foralls de menor número de itemções (divisão dos índices do fora/[).…”

“…O modelo de programação SPMD (Single Program Multiple Data) [FC95,AG94], com processos executando no estilo mestre/escravo [HB93] foi escolhido para receber as rotinas de comunicação coletiva de MP/. Para a transformação de um programa SJSAL para o modelo SPMD realizou-se o particionamento s/ice, [SC90], da estrutura fora//, [Ske91], de modo a inserir cada fatia obtida em um processo diferente. O cb.tigo fonte mnnipulndo foi o do formato /FI, fSGR5, SWR51.…”