Comparative analysis of Cetus using benchmarks
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
With the increased induction of multicore architecture in the computing devices, the automatic parallelization has emerged as an important tool to make the source code optimized with the multicore architectures, which in return make the computing more efficient. Cetus, a source-to-source C compiler, helps laying the ground for the research on optimizing the automatic parallelization. It facilitates the source-to-source transformations of the C programs with important parallelization passes and their enabling techniques. It also aids researchers in analyzing the original and the transformed code. ☐ Having focus on for-loops parallelization, we analyze Cetus’ performance by running it over various known benchmarks and some custom algorithms and later compare its output with one of the highest-quality source-to-source parallelizer named as Rose. The goal of our analysis is to identify algorithms/programs where Cetus fails to optimize compared to Rose or voice-versa. Some bugs and their solution in Cetus will also be identified. In the end, a future work to improve Cetus and its usage, a Cetus wish-list will be proposed which has all the identified bugs and weaknesses that can be fixed/improved in future. ☐ Tools to be utilized are Cetus compiler, Rose compiler and GCC by GNU. Ubuntu 16.04.1 LTS operating system environment will be used on the MacBook Air 13-inch, Early 2015, with a hardware specification of 2.2 GHz, Intel core i7 processor with 8GB of RAM.
Description
Keywords
Benchmarks, Cetus, Compiler, NAS, Parallel computing, Rose