Resource-constrained software pipelining

Abstract

This paper presents a software pipelining algorithm for the automatic extraction of ne-grain parallelism in general loops. The algorithm accounts for machine resource constraints in a way that smoothly integrates the management of resource constraints with software pipelining. Furthermore, generality in the software pipelining algorithm is not sacri ced to handle resource constraints, and scheduling choices are made with truly global information. Proofs of correctness and the results of experiments with an implementation are also presented.

Global Software Pipelining with Iteration Preselection

“…Unrolling kernel recognition algorithms [AN90] are theoretically very powerful. In practice, however, they cause enormous code growth and their controlling heuristics restrict their power [NN97].…”

Creating Converged Trace Schedules Using String Matching

“…Broadly these algorithms can be classified under the following three categories: Enhanced Pipeline Scheduling [10,12], Modulo Scheduling [21,22] and Kernel Recognition [2,4]. A detailed survey on these algorithms can be found in [3].…”

“…Kernel Recognition algorithms [2,4] simultaneously unroll and schedule the loop. This process is continued until a repeating pattern is detected in the schedule.…”

“…Software pipelining [17] was proposed as a means to expose more ILP to these architectures. Even though there have been numerous approaches [2,3,10,12,21,4] to software pipelining, it has remained a complex problem.…”

“…Kernel Recognition [2,3,4] algorithms are one class of solutions to do software pipelining. These algorithms simultaneously unroll and schedule the loops.…”