HYDA: A HYbrid Dependence Analysis for the adaptive optimisation of OpenCL kernels

Authors:
Christos Margiolas;Michael F. P. O'Boyle
Affiliations:
School of Informatics, University of Edinburgh;School of Informatics, University of Edinburgh
Venue:
Proceedings of International Workshop on Adaptive Self-tuning Computing Systems
Year:
2014

Citing 5
Cited 0

Practical dependence testing

PLDI '91 Proceedings of the ACM SIGPLAN 1991 conference on Programming language design and implementation
The Omega test: a fast and practical integer programming algorithm for dependence analysis

Proceedings of the 1991 ACM/IEEE conference on Supercomputing
A practical algorithm for exact array dependence analysis

Communications of the ACM
LLVM: A Compilation Framework for Lifelong Program Analysis & Transformation

Proceedings of the international symposium on Code generation and optimization: feedback-directed and runtime optimization
Code Generation in the Polyhedral Model Is Easier Than You Think

Proceedings of the 13th International Conference on Parallel Architectures and Compilation Techniques

Quantified Score

Hi-index	0.00

Visualization

Abstract

We present our work on a hybrid compiler analysis for the detection of memory access dependences on OpenCL kernels. Our analysis relies on (i) information extracted statically from the code and on (ii) the dynamic tracing of kernel memory accesses. Our analysis extracts memory access patterns that we formalise. We target aggressive data layout transformations, automatic memory coalescing and transparent partitioning of non data-parallel kernels. Our work integrates with the standard OpenCL runtime and operates transparently for both OpenCL environment and application. The analysis results are used for the Just In Time (JIT) optimisation of kernel codes.