2013 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) 2013
DOI: 10.1109/ispass.2013.6557172
|View full text |Cite
|
Sign up to set email alerts
|

Non-determinism and overcount on modern hardware performance counter implementations

Abstract: Ideal hardware performance counters provide exact deterministic results. Real-world performance monitoring unit (PMU) implementations do not always live up to this ideal. Events that should be exact and deterministic (such as retired instructions) show run-to-run variation and overcount on x86 64 machines, even when run in strictly controlled environments. These effects are non-intuitive to casual users and cause difficulties when strict determinism is desirable, such as when implementing deterministic replay … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
74
0
2

Year Published

2014
2014
2023
2023

Publication Types

Select...
3
3
3

Relationship

0
9

Authors

Journals

citations
Cited by 103 publications
(76 citation statements)
references
References 17 publications
0
74
0
2
Order By: Relevance
“…Real-world hardware counters usually do not live up to the ideal ones. The undesired deviation from the expected result is usually due to non-determinism (different values for identical runs) and overcount (counting some instructions multiple times) [14]. There are various external sources of these variations including program layout, measurement overhead, multi-processor variations and uncertainty of compilers that may leave many unexplored corner cases.…”
Section: Related Workmentioning
confidence: 99%
“…Real-world hardware counters usually do not live up to the ideal ones. The undesired deviation from the expected result is usually due to non-determinism (different values for identical runs) and overcount (counting some instructions multiple times) [14]. There are various external sources of these variations including program layout, measurement overhead, multi-processor variations and uncertainty of compilers that may leave many unexplored corner cases.…”
Section: Related Workmentioning
confidence: 99%
“…This is because performance counters report slightly different number of instructions even for identical instruction sequences, as reported in [39]. However, our framework is unaffected by this issue.…”
Section: Schedulermentioning
confidence: 93%
“…Hardware counters are limited on mobile devices. For example, L2 memory counters are not available on many ARM processors [Weaver et al 2013]. This prevents porting analytical methodologies which are relying on hardware performance counters.…”
Section: Methodology Restrictionsmentioning
confidence: 99%