CURD: a dynamic CUDA race detector

Peng, Yuanfeng; Grover, Vinod; Devietti, Joseph

doi:10.1145/3192366.3192368

Cited by 16 publications

(12 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such techniques are better suited for highly detailed analysis in smaller kernels, and typically are unable to prove DRF. Dynamic data-race detection executes a kernel to find data-races on a fixed input, e.g., [14,18,19,28,32,38,39]. This technique only reports real data-races, but suffers from a slowdown of at least 10× compared to the non-instrumented program, and requires the kernel input data, which might be unavailable or unknown.…”

Section: Related Workmentioning

confidence: 99%

Checking Data-Race Freedom of GPU Kernels, Compositionally

Cogumbreiro

Lange

Rong

et al. 2021

Computer Aided Verification

View full text Add to dashboard Cite

GPUs offer parallelism as a commodity, but they are difficult to program correctly. Static analyzers that guarantee data-race freedom (DRF) are essential to help programmers establish the correctness of their programs (kernels). However, existing approaches produce too many false alarms and struggle to handle larger programs. To address these limitations we formalize a novel compositional analysis for DRF, based on access memory protocols. These protocols are behavioral types that codify the way threads interact over shared memory.Our work includes fully mechanized proofs of our theoretical results, the first mechanized proofs in the field of DRF analysis for GPU kernels. Our theory is implemented in , a tool that outperforms the state-of-the-art. Notably, it can correctly verify at least $$1.42{\times }$$ 1.42 × more real-world kernels, and it exhibits a linear growth in 4 out of 5 experiments, while others grow exponentially in all 5 experiments.

show abstract

Section: Related Workmentioning

confidence: 99%

Checking Data-Race Freedom of GPU Kernels, Compositionally

Cogumbreiro

Lange

Rong

et al. 2021

Computer Aided Verification

View full text Add to dashboard Cite

show abstract

“…Given its importance in fields such as machine learning and highperformance computing, CUDA has gained a fair amount of attention in the program analysis literature in recent years. There exist a number of static [Li and Gopalakrishnan 2010;Pereira et al 2016;Zheng et al 2011] and dynamic [Boyer et al 2008;Eizenberg et al 2017;Peng et al 2018;Wu et al 2019] analyses for verifying certain properties of CUDA programs, but much of this work focused on functional properties, e.g., freedom from data races. Wu et al [2019] investigate several classes of bugs, one of which is łnon-optimal implementationž, including several types of performance problems.…”

Section: Related Workmentioning

confidence: 99%

Modeling and analyzing evaluation cost of CUDA kernels

Muller

Hoffmann

2021

Proc. ACM Program. Lang.

View full text Add to dashboard Cite

General-purpose programming on GPUs (GPGPU) is becoming increasingly in vogue as applications such as machine learning and scientific computing demand high throughput in vector-parallel applications. NVIDIA's CUDA toolkit seeks to make GPGPU programming accessible by allowing programmers to write GPU functions, called kernels, in a small extension of C/C++. However, due to CUDA's complex execution model, the performance characteristics of CUDA kernels are difficult to predict, especially for novice programmers. This paper introduces a novel quantitative program logic for CUDA kernels, which allows programmers to reason about both functional correctness and resource usage of CUDA kernels, paying particular attention to a set of common but CUDA-specific performance bottlenecks. The logic is proved sound with respect to a novel operational cost semantics for CUDA kernels. The semantics, logic and soundness proofs are formalized in Coq. An inference algorithm based on LP solving automatically synthesizes symbolic resource bounds by generating derivations in the logic. This algorithm is the basis of RaCuda, an end-to-end resource-analysis tool for kernels, which has been implemented using an existing resource-analysis tool for imperative programs. An experimental evaluation on a suite of CUDA benchmarks shows that the analysis is effective in aiding the detection of performance bugs in CUDA kernels.

show abstract

“…Data race detection analyses have been proposed for parallel programming models such as OpenMP [3,11,30,74] and GPUs [19,33,34,42,58,86,87].…”

Section: Related Workmentioning

confidence: 99%

Efficient Data Race Detection of Async-Finish Programs Using Vector Clocks

Kumar,

Agrawal,

Biswas

2021

Preprint

View full text Add to dashboard Cite

Existing data race detectors for task-based programs incur large run time and space overheads. The overheads arise because of frequent lookups in fine-grained tree data structures to check whether two accesses can happen in parallel.This work shows how to efficiently apply vector clocks for dynamic race detection of async-finish programs with locks. Our proposed technique, FastRacer, builds on the Fast-Track algorithm with per-task and per-variable optimizations to reduce the size of vector clocks. FastRacer also exploits the structured parallelism of async-finish programs to use a coarse-grained encoding of the dynamic task inheritance to limit the metadata in the presence of many concurrent readers. Our evaluation shows that FastRacer substantially improves time and space overheads over FastTrack and is competitive with the state-of-the-art data race detectors for async-finish programs with locks.

show abstract

CURD: a dynamic CUDA race detector

Cited by 16 publications

References 36 publications

Checking Data-Race Freedom of GPU Kernels, Compositionally

Checking Data-Race Freedom of GPU Kernels, Compositionally

Modeling and analyzing evaluation cost of CUDA kernels

Efficient Data Race Detection of Async-Finish Programs Using Vector Clocks

Contact Info

Product

Resources

About