What is it about?
Programmers expect accurate execution of their programs from the processor. We show that selective computations in the program can be carried out inaccurately by the processor, in order to be performance efficient, but keeping the overall result within tolerable error margins. We present methods of inferencing the computations in a program that are error tolerant and demonstrate performance gain that we achieved with our approximate computing scheme.
Featured Image
Why is it important?
Our work proposes new ISA extensions that can be embraced for approximate computing in processors. An application of statistical inference techniques in architecture design, is remarkable. Lastly, considerable performance benefit is demonstrated on many applications from the approximate computing domain with our proposed computation scheme.
Read the Original
This page is a summary of: Enhancing Speculative Execution With Selective Approximate Computing, ACM Transactions on Design Automation of Electronic Systems, March 2019, ACM (Association for Computing Machinery),
DOI: 10.1145/3307651.
You can read the full text:
Contributors
The following have contributed to this page
