Constrained Precision Tuning

TL;DR

This paper discusses a tool for mixed precision tuning in numerical programs, balancing energy efficiency and performance by optimizing variable precision assignments while managing conversion costs.

Contribution

It introduces enhancements to the POP tool to better limit drawbacks of mixed precision tuning, improving the trade-off between performance and memory usage.

Findings

Effective reduction in energy consumption and memory footprint.

Improved performance through optimized mixed precision assignments.

Demonstrated benefits on FPBench suite tests.

Abstract

Precision tuning or customized precision number representations is emerging, in these recent years, as one of the most promising techniques that has a positive impact on the footprint of programs concerning energy consumption, bandwidth usage and computation time of numerical programs. In contrast to the uniform precision, mixed precision tuning assigns different finite-precision types to each variable and arithmetic operation of a program and offers many additional optimization opportunities. However, this technique introduces new challenge related to the cost of operations or type conversions which can overload the program execution after tuning. In this article, we extend our tool POP (Precision OPtimizer), with efficient ways to limit the number of drawbacks of mixed precision and to achieve best compromise between performance and memory consumption. On a popular set of tests from the FPBench suite, we discuss the results obtained by POP.