On the Determination of Inlining Vectors for Program Optimization

Cammarota, Rosario; Nicolau, Alexandru; Veidenbaum, Alexander V.; Kejariwal, Arun; Donato, Debora; Madhugiri, Mukund

doi:10.1007/978-3-642-37051-9_9

Cited by 6 publications

(2 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We plan to more intelligently choose when to inline and when not on work-item loop basis. A method similar to the one presented in [13] could be used. All of the worst-performing cases presented in Section 6 would benefit from these.…”

Section: Discussionmentioning

confidence: 99%

pocl: A Performance-Portable OpenCL Implementation

Jääskeläinen

Lama²,

Schnetter

et al. 2014

Int J Parallel Prog

108

View full text Add to dashboard Cite

OpenCL is a standard for parallel programming of heterogeneous systems. The benefits of a common programming standard are clear; multiple vendors can provide support for application descriptions written according to the standard, thus reducing the program porting effort. While the standard brings the obvious benefits of platform portability, the performance portability aspects are largely left to the programmer. The situation is made worse due to multiple proprietary vendor implementations with different characteristics, and, thus, required optimization strategies. In this paper, we propose an OpenCL implementation that is both portable and performance portable. At its core is a kernel compiler that can be used to exploit the data parallelism of OpenCL programs on multiple platforms with different parallel hardware styles. The kernel compiler is modularized to perform target-independent parallel region formation separately from the target-specific parallel mapping of the regions to enable support for various styles of fine-grained parallel resources such as subword SIMD extensions, SIMD datapaths and static multi-issue. Unlike previous similar techniques that work on the source level, the parallel region formation retains the information of the data parallelism using the LLVM IR and its metadata infrastructure. This data can be exploited by the later generic compiler passes for efficient parallelization. The proposed open source implementation of OpenCL is also platform portable, enabling OpenCL on a wide range of architectures, both already commercialized and on those that are still under research. The paper describes how the portability of the implementation is achieved. Our results show that most of the benchmarked applications when compiled using pocl were faster or close to as fast as the best proprietary OpenCL implementation for the platform at hand.Comment: This article was published in 2015; it is now openly accessible via arxi

show abstract

Section: Discussionmentioning

confidence: 99%

pocl: A Performance-Portable OpenCL Implementation

Jääskeläinen

Lama²,

Schnetter

et al. 2014

Int J Parallel Prog

108

View full text Add to dashboard Cite

show abstract

“…Cammarota et al [10] select inlining vectors for program performance (completion time) optimization by applying machine learning techniques using hardware performance counters. They use 10-fold crossvalidation to select the most suitable technique similar to us.…”

Section: Machine Learning Techniquesmentioning

confidence: 99%

Using Machine Learning Techniques to Detect Parallel Patterns of Multi-threaded Applications

Deniz

Şen

2015

Int J Parallel Prog

View full text Add to dashboard Cite

Multicore hardware and software are becoming increasingly more complex. The programmability problem of multicore software has led to the use of parallel patterns. Parallel patterns reduce the effort and time required to develop multicore software by effectively capturing its thread communication and data sharing characteristics. Hence, detecting the parallel pattern used in a multi-threaded application is crucial for performance improvements and enables many architectural optimizations; however, this topic has not been widely studied. We apply machine learning techniques in a novel approach to automatically detect parallel patterns and compare these techniques in terms of accuracy and speed. We experimentally validate the detection ability of our techniques on benchmarks including PARSEC and Rodinia. Our experiments show that the k-nearest neighbor, decision trees, and naive Bayes classifier are the most accurate techniques. Overall, decision trees are the fastest technique with the lowest characterization overhead producing the best combination of detection results. We also show the usefulness of the proposed techniques on synthetic benchmark generation.

show abstract