Fast and Efficient Bit-Level Precision Tuning

@inproceedings{Adje2021FastAE,
  title={Fast and Efficient Bit-Level Precision Tuning},
  author={Assal'e Adj'e and Dorra Ben Khalifa and Matthieu Martel},
  booktitle={SAS},
  year={2021}
}
In this article, we introduce a new technique for precision tuning. This problem consists of finding the least data types for numerical values such that the result of the computation satisfies some accuracy requirement. State of the art techniques for precision tuning use a try and fail approach. They change the data types of some variables of the program and evaluate the accuracy of the result. Depending on what is obtained, they change more or less data types and repeat the process. Our… 
1 Citations

Constrained Precision Tuning

  • Dorra Ben KhalifaM. Martel
  • Computer Science
    2022 8th International Conference on Control, Decision and Information Technologies (CoDIT)
  • 2022
This article extends the tool POP, with efficient ways to limit the number of drawbacks of mixed precision and to achieve best compromise between performance and memory consumption.

References

SHOWING 1-10 OF 24 REFERENCES

Precimonious: Tuning assistant for floating-point precision

Premonious is a dynamic program analysis tool to assist developers in tuning the precision of floating-point programs and recommends a type instantiation that uses lower precision while producing an accurate enough answer without causing exceptions.

A Policy Iteration Algorithm for Computing Fixed Points in Static Analysis of Programs

A policy iteration algorithm for monotone self-maps of complete lattices for lattices arising in the interval abstraction of values of variables is introduced and analyzed.

POP: A Tuning Assistant for Mixed-Precision Floating-Point Computations

A static program analysis to determine the lowest floating-point precisions on inputs and intermediate results that guarantees a desired accuracy of the output values is described.

Precision Tuning and Internet of Things

  • Dorra Ben KhalifaM. Martel
  • Computer Science
    2019 International Conference on Internet of Things, Embedded Systems and Communications (IINTEC)
  • 2019
A floating-point precision tuning tool called POP: Precision OPtimizer, which integrates a static forward and backward program analysis, done by abstract interpretation, to determine the minimal precision on the inputs and the intermediary results of a program in order to ensure a desired accuracy on the outputs.

A Study of the Floating-Point Tuning Behaviour on the N-body Problem

The efficiency of POP to tune the classical gravitational N-body problem is demonstrated by considering five bodies that interact under gravitational force from one another, subject to Newton’s laws of motion.

Precision Tuning of an Accelerometer-Based Pedometer Algorithm for IoT Devices

  • Dorra Ben KhalifaM. Martel
  • Computer Science
    2020 IEEE International Conference on Internet of Things and Intelligence System (IoTaIS)
  • 2021
POP provides a mixed precision tuning that finds the instructions and variables that may use lower precision with respect to the user accuracy requirements on the results, and demonstrates the efficiency of POP to tune an accelerometer-based pedometer algorithm for embedded applications.

Tools for Reduced Precision Computation

There is still a gap to close in automation of reduced precision customization, especially for tools based on static analysis rather than profiling, as well as for integration within mainstream, industry-strength compiler frameworks.

AMPT-GA: automatic mixed precision floating point tuning for GPU applications

This paper presents a system called AMPT-GA, a system that selects application-level data precisions to maximize performance while satisfying accuracy constraints, and improves the performance efficiency of the target applications more than the prior state-of-the-art approach called Precimonious.

Error Analysis of ZFP Compression for Floating-Point Data

This paper analyzes the round-off error introduced by ZFP, a state-of-the-art lossy compression algorithm, and defines operators that implement each step of the ZFP compression and decompression to establish a bound on the error caused byZFP.

Exploiting community structure for floating-point precision tuning

This paper exploits the community structure of floating-point variables to devise a scalable hierarchical search for precision tuning, and presents a hierarchical search algorithm that iteratively lowers precision with regard to communities.