Enabling user-driven Checkpointing strategies in Reverse-mode Automatic Differentiation

TL;DR

This paper enhances the Tapenade AD tool by enabling user-driven checkpointing strategies, allowing for tailored memory and computation trade-offs, and demonstrates significant memory and time savings in scientific applications.

Contribution

It introduces user-selectable checkpointing options in Tapenade and provides experimental insights towards achieving optimal checkpointing strategies.

Findings

Memory savings up to 35%

Execution time reduction up to 90%

Guidelines for optimal checkpointing strategies

Abstract

This paper presents a new functionality of the Automatic Differentiation (AD) tool Tapenade. Tapenade generates adjoint codes which are widely used for optimization or inverse problems. Unfortunately, for large applications the adjoint code demands a great deal of memory, because it needs to store a large set of intermediates values. To cope with that problem, Tapenade implements a sub-optimal version of a technique called checkpointing, which is a trade-off between storage and recomputation. Our long-term goal is to provide an optimal checkpointing strategy for every code, not yet achieved by any AD tool. Towards that goal, we first introduce modifications in Tapenade in order to give the user the choice to select the checkpointing strategy most suitable for their code. Second, we conduct experiments in real-size scientific codes in order to gather hints that help us to deduce an optimal checkpointing strategy. Some of the experimental results show memory savings up to 35% and execution time up to 90%.