Learning to Combine Instructions in LLVM Compiler

TL;DR

This paper introduces a Neural Instruction Combiner (NIC) that uses neural sequence-to-sequence models to replace traditional instruction combining in LLVM, reducing manual effort and improving optimization efficiency.

Contribution

First implementation of a neural instruction combiner integrated into a full compiler pipeline, with a new dataset and demonstrating high accuracy and feasibility.

Findings

NIC achieves 72% exact match with traditional IC

BLEU score of 0.94 indicating high translation quality

Feasible integration into production compiler pipeline

Abstract

Instruction combiner (IC) is a critical compiler optimization pass, which replaces a sequence of instructions with an equivalent and optimized instruction sequence at basic block level. There can be thousands of instruction-combining patterns which need to be frequently updated as new coding idioms/applications and novel hardware evolve over time. This results in frequent updates to the IC optimization pass thereby incurring considerable human effort and high software maintenance costs. To mitigate these challenges associated with the traditional IC, we design and implement a Neural Instruction Combiner (NIC) and demonstrate its feasibility by integrating it into the standard LLVM compiler optimization pipeline. NIC leverages neural sequence-to-sequence (Seq2Seq) models for generating optimized encoded IR sequence from the unoptimized encoded IR sequence. To the best of our knowledge, ours is the first work demonstrating the feasibility of a neural instruction combiner built into a full-fledged compiler pipeline. Given the novelty of this task, we built a new dataset for training our NIC neural model. We show that NIC achieves exact match results percentage of 72% for optimized sequences as compared to traditional IC and neural machine translation metric Bleu precision score of 0.94, demonstrating its feasibility in a production compiler pipeline.