COMET: Neural Cost Model Explanation Framework

TL;DR

COMET is a framework that explains neural cost models for CPU assembly code, improving interpretability and understanding of their predictions, which can enhance their adoption in compiler workflows.

Contribution

It introduces the first explanation framework for neural cost models, enabling faithful and intuitive insights into their decision-making process.

Findings

COMET's explanations reveal correlations between feature granularity and model errors.

Ithemal's higher prediction errors are linked to less granular features in explanations.

COMET helps identify potential reasons for neural model inaccuracies.

Abstract

Cost models predict the cost of executing given assembly code basic blocks on a specific microarchitecture. Recently, neural cost models have been shown to be fairly accurate and easy to construct. They can replace heavily engineered analytical cost models used in mainstream compiler workflows. However, their black-box nature discourages their adoption. In this work, we develop the first framework, COMET, for generating faithful, generalizable, and intuitive explanations for neural cost models. We generate and compare COMET's explanations for the popular neural cost model, Ithemal against those for an accurate CPU simulation-based cost model, uiCA. Our empirical findings show an inverse correlation between the prediction errors of Ithemal and uiCA and the granularity of basic block features in COMET's explanations for them, thus indicating potential reasons for the higher error of Ithemal with respect to uiCA.