Static Nonconcurrency Analysis of OpenMP Programs

Lin, Yaping

doi:10.1007/978-3-540-68555-5_4

Cited by 25 publications

(28 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Current static tools have been proved to work properly on specific subsets of OpenMP such as having a fixed number of threads [22] or using only affine constructs [6]. A more general approach can be used to determine the regions of code that are definitely non-concurrent [21]. Although it is not an accurate solution, it does not produce false negatives, which is paramount in the safety-critical domain.…”

Section: Race Conditionsmentioning

confidence: 99%

A Functional Safety OpenMP $$^{*}$$ for Critical Real-Time Embedded Systems

Royuela

Durán

Serrano

et al. 2017

Scaling OpenMP for Exascale Performance and Portability

View full text Add to dashboard Cite

Abstract. OpenMP* has recently gained attention in the embedded domain by virtue of the augmentations implemented in the last specification. Yet, the language has a minimal impact in the embedded real-time domain mostly due to the lack of reliability and resiliency mechanisms. As a result, functional safety properties cannot be guaranteed. This paper analyses in detail the latest specification to determine whether and how the compliant OpenMP implementations can guarantee functional safety. Given the conclusions drawn from the analysis, the paper describes a set of modifications to the specification, and a set of requirements for compiler and runtime systems to qualify for safety critical environments. Through the proposed solution, OpenMP can be used in critical real-time embedded systems without compromising functional safety.

show abstract

Section: Race Conditionsmentioning

confidence: 99%

A Functional Safety OpenMP $$^{*}$$ for Critical Real-Time Embedded Systems

Royuela

Durán

Serrano

et al. 2017

Scaling OpenMP for Exascale Performance and Portability

View full text Add to dashboard Cite

show abstract

“…Current mechanisms have been proved to work properly on specific subsets of OpenMP such as having a fixed number of threads [22] or avoiding the use of non-affine constructs 10 [10]. A more general approach can be used to determine the regions of code that are definitely non-concurrent [20]. Although it is not an accurate solution, it will never deliver false negatives.…”

Section: Correctness: Data Races and Synchronizationmentioning

confidence: 99%

OpenMP Tasking Model for Ada: Safety and Correctness

Royuela

Martorell

Quiñones

et al. 2017

Reliable Software Technologies – Ada-Europe 2017

View full text Add to dashboard Cite

The safety-critical real-time embedded domain increasingly demands the use of parallel architectures to fulfill performance requirements. Such architectures require the use of parallel programming models to exploit the underlying parallelism. This paper evaluates the applicability of using OpenMP, a widespread parallel programming model, with Ada, a language widely used in the safety-critical domain.Concretely, this paper shows that applying the OpenMP tasking model to exploit fine-grained parallelism within Ada tasks does not impact on programs safeness and correctness, which is vital in the environments where Ada is mostly used. Moreover, we compare the OpenMP tasking model with the proposal of Ada extensions to define parallel blocks, parallel loops and reductions. Overall, we conclude that the OpenMP tasking model can be safely used in such environments, being a promising approach to exploit fine-grain parallelism in Ada tasks, and we identify the issues which still need to be further researched. Abstract. The safety-critical real-time embedded domain increasingly demands the use of parallel architectures to fulfill performance requirements. Such architectures require the use of parallel programming models to exploit the underlying parallelism. This paper evaluates the applicability of using OpenMP, a widespread parallel programming model, with Ada, a language widely used in the safety-critical domain. Concretely, this paper shows that applying the OpenMP tasking model to exploit fine-grained parallelism within Ada tasks does not impact on programs safeness and correctness, which is vital in the environments where Ada is mostly used. Moreover, we compare the OpenMP tasking model with the proposal of Ada extensions to define parallel blocks, parallel loops and reductions. Overall, we conclude that the OpenMP tasking model can be safely used in such environments, being a promising approach to exploit fine-grain parallelism in Ada tasks, and we identify the issues which still need to be further researched.

show abstract

“…This tool is restricted to program fragments called Affine Control Loops but it has the advantage of reporting accurate errors to the user. Lin [8] describes a concurrency analysis technique to detect whether two statements will not be executed concurrently by different threads in a team. The method is an intraprocedural analysis based on phase partioning using an OpenMP Control Flow Graph (OMPCFG) that models the transfer of control flow in an OpenMP program.…”

Section: Related Workmentioning

confidence: 99%

“…Unlike Zhang et al, our analysis is language independent and verifies woksharing-construct placements in a program. To detect possible deadlocks we use the graph representation defined in [8]. Potential errors are automatically returned to the user with the line of the erroneous conditionals by a simple analysis of the OMPCFG.…”

Section: Related Workmentioning

confidence: 99%

Static Validation of Barriers and Worksharing Constructs in OpenMP Applications

Saillard

Carribault

Barthou

2014

Using and Improving OpenMP for Devices, Tasks, and More

View full text Add to dashboard Cite

Abstract. The OpenMP specification requires that all threads in a team execute the same sequence of worksharing and barrier regions. An improper use of such directive may lead to deadlocks. In this paper we propose a static analysis to ensure this property is verified. The well-defined semantic of OpenMP programs makes compiler analysis more effective. We propose a new compile-time method to identify in OpenMP codes the potential improper uses of barriers and worksharing constructs, and the execution paths that are responsible for these issues. We implemented our method in a GCC compiler plugin and show the small impact of our analysis on performance for NAS-OMP benchmarks and a test case for a production industrial code. IntroductionOpenMP is a popular parallel programming model for shared memory machines. While OpenMP aims at making parallel programming easier, there are a number of improper uses of worksharing constructs and barriers that are not statically detected by compilers and may lead to deadlock or unspecified behavior. Indeed, the OpenMP specification requires that all threads of a team must execute the same sequence of worksharing constructs and barriers [16]. However in practice no error occurs when all threads of a team do not execute exactly the same barrier. That is why we authorize threads synchronizations with different barriers and defined two verbosity levels (0 and 1) defining soft and hard barriers verifications. Throughout the rest of the paper, examples are presented with verbosity level 0.To show the difficulty to enforce this constraint in OpenMP codes, consider the motivating examples in Figure 1. In function f of Listing 1.1, each thread may or may not encounter the single construct line 9, depending on the control flow (line 6). According to the OpenMP specification, all threads in a team should encounter the same single, or none of them. However, compiling this code and executing it does not lead to a syntactic error but to a deadlock. Indeed, if the result of the conditional is not the same among all threads, the first barrier executed will be for some threads the implicit barrier line 12 (end of single) while for others, it will be the explicit barrier line 14. Then the first group of threads will stop at the explicit barrier line 14 while the second group will stop at the barrier related to the end of the parallel region. Finally, the first set of threads will be released and eventually deadlock at this last barrier. Note that if Listing 1.1.

show abstract

Static Nonconcurrency Analysis of OpenMP Programs

Cited by 25 publications

References 7 publications

A Functional Safety OpenMP $$^{*}$$ for Critical Real-Time Embedded Systems

A Functional Safety OpenMP $$^{*}$$ for Critical Real-Time Embedded Systems

OpenMP Tasking Model for Ada: Safety and Correctness

Static Validation of Barriers and Worksharing Constructs in OpenMP Applications

Contact Info

Product

Resources

About