Enhancing the Top-Down Microarchitectural Analysis Method Using Purchasing Power Parity Theory

Yectli A. Huerta, Brent Swartz, David J. Lilja

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The Top-Down method makes it possible to identify bottlenecks as instructions traverse the CPU’s pipeline. Once bottlenecks are identified, incremental changes to the code can be made to mitigate the negative effects bottlenecks might have in performance. This is an iterative process that could potentially result in a more optimal use of CPU resources. It can be difficult to compare bottleneck metrics of the same program generated by different compilers running on the same system. Different compilers could potentially generate different instructions, arrange the instructions in different order, and require different number of cycles to execute the program. Ratios with relatively similar values could hide valuable information that could be used to identify differences in magnitude and influence of bottlenecks. To amplify magnitude differences of bottleneck metrics, we use the cycles required to complete the program as a reference point. We can then quantify the relative difference the effect a bottleneck has when compared with the bottleneck of the reference compiler. This study’s proposed approach is based on the Purchasing Power Parity theory, which is used by economists to compare the purchasing power of different currencies by comparing similar products. We show that this approach can give us more information on how effective each compiler is in using the CPU’s architectural features by comparing their respective bottlenecks. For example, using conventional methods, our measurements show that for the 363.swim benchmark, BackEnd Bound rates for GCC4 was 0.949, and 0.956 for GCC6 and GCC7 respectively. However, using the PPP normalization approach, we showed that there were differences of 55.3 % for GCC6 and 54.9 % for GCC7 over GCC4.

Original languageEnglish (US)
Title of host publicationLanguages and Compilers for Parallel Computing - 33rd International Workshop, LCPC 2020
EditorsBarbara Chapman, José Moreira
PublisherSpringer Science and Business Media Deutschland GmbH
Pages163-177
Number of pages15
ISBN (Print)9783030959524
DOIs
StatePublished - 2022
Event33rd International Workshop on Languages and Compilers for Parallel Computing, LCPC 2020 - Virtual, Online
Duration: Oct 14 2020Oct 16 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13149 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference33rd International Workshop on Languages and Compilers for Parallel Computing, LCPC 2020
CityVirtual, Online
Period10/14/2010/16/20

Bibliographical note

Publisher Copyright:
© 2022, Springer Nature Switzerland AG.

Fingerprint

Dive into the research topics of 'Enhancing the Top-Down Microarchitectural Analysis Method Using Purchasing Power Parity Theory'. Together they form a unique fingerprint.

Cite this