goSLP: Globally Optimized Superword Level Parallelism Framework

TL;DR

goSLP is a new auto-vectorization framework that uses integer linear programming to globally optimize superword level parallelism, significantly improving compiler performance on benchmark suites.

Contribution

It introduces a globally optimal approach to statement packing and permutation selection in SLP auto-vectorization using ILP and dynamic programming, surpassing heuristic methods.

Findings

Achieved a 7.58% speedup on SPEC2017fp benchmarks.

Achieved a 2.42% speedup on SPEC2006fp benchmarks.

Achieved a 4.07% speedup on NAS benchmarks.

Abstract

Modern microprocessors are equipped with single instruction multiple data (SIMD) or vector instruction sets which allow compilers to exploit superword level parallelism (SLP), a type of fine-grained parallelism. Current SLP auto-vectorization techniques use heuristics to discover vectorization opportunities in high-level language code. These heuristics are fragile, local and typically only present one vectorization strategy that is either accepted or rejected by a cost model. We present goSLP, a novel SLP auto-vectorization framework which solves the statement packing problem in a pairwise optimal manner. Using an integer linear programming (ILP) solver, goSLP searches the entire space of statement packing opportunities for a whole function at a time, while limiting total compilation time to a few minutes. Furthermore, goSLP optimally solves the vector permutation selection problem using dynamic programming. We implemented goSLP in the LLVM compiler infrastructure, achieving a geometric mean speedup of 7.58% on SPEC2017fp, 2.42% on SPEC2006fp and 4.07% on NAS benchmarks compared to LLVM's existing SLP auto-vectorizer.