Detecting Data-Races in High-Performance Computing

Technical Report · Wed Oct 30 00:00:00 EDT 2019

DOI:https://doi.org/10.2172/1573167· OSTI ID:1573167

Schordan, Markus ^[1]; Liao, Chunhua ^[1]; Lin, Pei-Hung ^[1]; Karlin, Ian ^[1]

---

1. Lawrence Livermore National Laboratory (LLNL), Livermore, CA (United States)

High-performance computing is susceptible to data-race bugs. Data races occur when sequences of computed events do not happen in the intended order, a significant barrier to the integrity of parallel programs using shared-memory architectures. Many approaches to data-race detection are based on traditional static-analysis techniques that typically execute on a program’s source code or control-flow graph (the graphical representation of all possible orders of events in the execution of a program). Similar to classical data-flow analyses, such tools can produce false alarms that report spurious race conditions on error free programs. We have developed a software verification technique that requires as input only a program’s source code and can automatically detect a data-race bug which definitely exists, determine that a program is definitely data-race free, or report that it cannot determine either of the two cases.

• Research Organization: Lawrence Livermore National Laboratory (LLNL), Livermore, CA (United States)
• Sponsoring Organization: USDOE National Nuclear Security Administration (NNSA)
• DOE Contract Number: AC52-07NA27344
• OSTI ID: 1573167
• Report Number(s): LLNL-TR-795673; 997076
• Country of Publication: United States
• Language: English

Related Subjects

Computer science