The framework and compilation techniques for directive-based GPU cluster programming

Using directive techniques for accelerating loops on heterogeneous systems, object-oriented programs can be hundreds of folds faster with a short learning curve required from programmers. Despite much research on various loop parallelization techniques, the opportunity to accelerate a compact construct like range-based loops using directive-based compilers has not yet been investigated. The paper addresses this issue and presents directive-based compiler techniques for accelerating range-based loops using OpenCL. The experimental results based on two workloads show the speedup of 3-135 folds achieved from accelerating range-based loops using the proposed compiler. The literature review depicts an average of 12-135 folds speedup gain by using traditional loop acceleration by existing directive-based compilers. This highlights the potential of exploiting loop-level parallelism by using range-based loop acceleration.

show abstract

The framework and compilation techniques for directive-based GPU cluster programming

Cited by 2 publications

References 9 publications

Accelerating range-based loops on heterogeneous systems

Accelerating range-based loops on heterogeneous systems

Compiler Support for Accelerating C++11 Range-Based Loops on Heterogeneous Systems

Contact Info

Product

Resources

About