ScholarLens Back to search
2011•Unpublished venueRequires access

SIMD re-convergence at thread frontiers

Gregory Diamos, Benjamin Ashbaugh, Subramaniam Maiyuran, Andrew Kerr, Haicheng Wu, Sudhakar Yalamanchili

Open publisher page 82 citations

Abstract

Hardware and compiler techniques for mapping data-parallel programs with divergent control flow to SIMD architectures have recently enabled the emergence of new GPGPU programming models such as CUDA, OpenCL, and DirectX Compute. The impact of branch divergence can be quite different depending upon whether the program's control flow is structured or unstructured. In this paper, we show that unstructured control flow occurs frequently in applications and can lead to significant code expansion when executed using existing approaches for handling branch divergence.

About this research paper

What this paper is about

Hardware and compiler techniques for mapping data-parallel programs with divergent control flow to SIMD architectures have recently enabled the emergence of new GPGPU programming models such as CUDA, OpenCL, and DirectX Compute. The impact of branch divergence can be quite different depending upon whether the program's control flow is structured or unstructured. In this paper, we show that unstructured control flow occurs frequently in applications and can lead to significant code expansion when executed using existing approaches for handling branch divergence.

Why it matters

OpenAlex reports 82 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Hardware and compiler techniques for mapping data-parallel programs with divergent control flow to SIMD architectures have recently enabled the emergence of new GPGPU programming models such as CUDA, OpenCL, and DirectX Compute. The impact of branch divergence can be quite different depending upon whether the program's control flow is structured or unstructured. In this paper, we show that unstructured control flow occurs frequently in applications and can lead to significant code expansion when executed using existing approaches for handling branch divergence.

Key concepts: Computer science, SIMD, Parallel computing, CUDA, Control flow, Compiler, Thread (computing), General-purpose computing on graphics processing units

Related papers

Related paper 1Related paper 2Related paper 3
Back to paper searchBrowse research topicsOriginal source
SIMD re-convergence at thread frontiers — Research Paper | ScholarLens