Register-Pressure-Aware Instruction Scheduling Using Ant Colony Optimization

This paper describes a new approach to register-pressure-aware instruction scheduling, using Ant Colony Optimization (ACO) . ACO is nature-inspired optimization technique that researchers have successfully applied NP-hard sequencing problems like the Traveling Salesman Problem (TSP) and its derivatives. In this work, we describe an algorithm for solving long-standing compiler problem of balancing Instruction-Level Parallelism (ILP) Register Pressure (RP) in pre-allocation scheduling. Three different cost functions are studied estimating RP during The proposed implemented LLVM open-source compiler, performance evaluated experimentally on three machines with instruction-set architectures: Intel x86, ARM, AMD GPU. compared exact Branch-and-Bound (B&B) previous work. On x86 both algorithms relative LLVM's generic scheduler, while GPU, AMD's production scheduler. experimental results show SPECrate 2017 Floating Point, gives geometric-mean improvements 1.13% 1.25% execution speed respectively, Using PlaidML it improvement 7.14% approximately same execution-time as B&B algorithm, each outperforming other substantial number hard scheduling regions. better than many large instances times out on. Both outperform CPU