Fence placement for legacy Data-Race-Free programs via synchronization read detection

Andrew J McPherson; Vijay Nagarajan; Susmit Sarkar; Marcelo Cintra

doi:10.1145/2835179

Fence placement for legacy Data-Race-Free programs via synchronization read detection

Andrew J McPherson, Vijay Nagarajan, Susmit Sarkar, Marcelo Cintra

School of Computer Science

Research output: Contribution to journal › Article › peer-review

Abstract

Shared-memory programmers traditionally assumed Sequential Consistency (SC), but modern systems have relaxed memory consistency. Here, the trend in languages is towards Data-Race-Free (DRF) models, where, assuming annotated synchronizations and the program being well-synchronized by those synchronizations, the hardware and compiler guarantee SC. However, legacy programs lack annotations, so even well-synchronized (legacy DRF) programs aren’t recognized. For legacy DRF programs, we can significantly prune the set of memory orderings determined by automated fence placement, by automatically identifying synchronization reads. We prove our rules for identifying them conservative, implement them within LLVM, and observe a 30% average performance improvement over previous techniques.

Original language	English
Article number	46
Number of pages	23
Journal	ACM Transactions on Architecture and Code Optimization (TACO)
Volume	12
Issue number	4
DOIs	https://doi.org/10.1145/2835179
Publication status	Published - 7 Jan 2016

Keywords

Fence placement
Relaxed memory models

Access to Document

10.1145/2835179

TACOFINAL
© 2015, Publisher / the Author(s). This work is made available online in accordance with the publisher’s policies. This is the author created, accepted version manuscript following peer review and may differ slightly from the final published version. The final published version of this work is available at dl.acm.org / https://dx.doi.org/10.1145/2835179
Accepted author manuscript, 304 KB

Cite this

@article{a056d73c8f2149cda332657620f5cff2,

title = "Fence placement for legacy Data-Race-Free programs via synchronization read detection",

abstract = "Shared-memory programmers traditionally assumed Sequential Consistency (SC), but modern systems have relaxed memory consistency. Here, the trend in languages is towards Data-Race-Free (DRF) models, where, assuming annotated synchronizations and the program being well-synchronized by those synchronizations, the hardware and compiler guarantee SC. However, legacy programs lack annotations, so even well-synchronized (legacy DRF) programs aren{\textquoteright}t recognized. For legacy DRF programs, we can significantly prune the set of memory orderings determined by automated fence placement, by automatically identifying synchronization reads. We prove our rules for identifying them conservative, implement them within LLVM, and observe a 30% average performance improvement over previous techniques.",

keywords = "Fence placement, Relaxed memory models",

author = "McPherson, {Andrew J} and Vijay Nagarajan and Susmit Sarkar and Marcelo Cintra",

note = "Funding: This research is supported by EPSRC grant EP/L000725/1 and an Intel early career faculty award to the University of Edinburgh.",

year = "2016",

month = jan,

day = "7",

doi = "10.1145/2835179",

language = "English",

volume = "12",

journal = "ACM Transactions on Architecture and Code Optimization (TACO)",

issn = "1544-3566",

publisher = "ACM",

number = "4",

}

TY - JOUR

T1 - Fence placement for legacy Data-Race-Free programs via synchronization read detection

AU - McPherson, Andrew J

AU - Nagarajan, Vijay

AU - Sarkar, Susmit

AU - Cintra, Marcelo

N1 - Funding: This research is supported by EPSRC grant EP/L000725/1 and an Intel early career faculty award to the University of Edinburgh.

PY - 2016/1/7

Y1 - 2016/1/7

N2 - Shared-memory programmers traditionally assumed Sequential Consistency (SC), but modern systems have relaxed memory consistency. Here, the trend in languages is towards Data-Race-Free (DRF) models, where, assuming annotated synchronizations and the program being well-synchronized by those synchronizations, the hardware and compiler guarantee SC. However, legacy programs lack annotations, so even well-synchronized (legacy DRF) programs aren’t recognized. For legacy DRF programs, we can significantly prune the set of memory orderings determined by automated fence placement, by automatically identifying synchronization reads. We prove our rules for identifying them conservative, implement them within LLVM, and observe a 30% average performance improvement over previous techniques.

AB - Shared-memory programmers traditionally assumed Sequential Consistency (SC), but modern systems have relaxed memory consistency. Here, the trend in languages is towards Data-Race-Free (DRF) models, where, assuming annotated synchronizations and the program being well-synchronized by those synchronizations, the hardware and compiler guarantee SC. However, legacy programs lack annotations, so even well-synchronized (legacy DRF) programs aren’t recognized. For legacy DRF programs, we can significantly prune the set of memory orderings determined by automated fence placement, by automatically identifying synchronization reads. We prove our rules for identifying them conservative, implement them within LLVM, and observe a 30% average performance improvement over previous techniques.

KW - Fence placement

KW - Relaxed memory models

U2 - 10.1145/2835179

DO - 10.1145/2835179

M3 - Article

SN - 1544-3566

VL - 12

JO - ACM Transactions on Architecture and Code Optimization (TACO)

JF - ACM Transactions on Architecture and Code Optimization (TACO)

IS - 4

M1 - 46

ER -

Fence placement for legacy Data-Race-Free programs via synchronization read detection

Abstract

Keywords

Access to Document

Fingerprint

Cite this