Theory of linear and integer programming
Theory of linear and integer programming
Scanning polyhedra with DO loops
PPOPP '91 Proceedings of the third ACM SIGPLAN symposium on Principles and practice of parallel programming
A Knowledge-Based Approach to the Analysis of Loops
IEEE Transactions on Software Engineering
Structure of Computers and Computations
Structure of Computers and Computations
Deriving Annotations for Tight Calculation of Execution Time
Euro-Par '97 Proceedings of the Third International Euro-Par Conference on Parallel Processing
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
Runtime predictability of loops
WWC '01 Proceedings of the Workload Characterization, 2001. WWC-4. 2001 IEEE International Workshop
Proceedings of the 13th ACM SIGPLAN Symposium on Principles and practice of parallel programming
Proceedings of the 7th annual IEEE/ACM International Symposium on Code Generation and Optimization
A view of the parallel computing landscape
Communications of the ACM - A View of Parallel Computing
Rodinia: A benchmark suite for heterogeneous computing
IISWC '09 Proceedings of the 2009 IEEE International Symposium on Workload Characterization (IISWC)
IEEE Micro
IEEE Spectrum
Loop transformations: convexity, pruning and optimization
Proceedings of the 38th annual ACM SIGPLAN-SIGACT symposium on Principles of programming languages
| Hi-index | 0.00 |
In applications that rely on data intensive computation, one can gain significant performance if the source code is suitably transformed for parallel hardware. A common approach is to identify the loops inside the program that consume a significant amount of time, that we call hotspots. One of the impending business need here is to quickly identify such loops for further transformation. However, the exact identification of such hotspots requires an elaborate runtime analysis. When we deal with a third party business application, only a partial version of the source code is available, with limited test inputs, which hinders a correct runtime analysis. Therefore, we resort to static analysis of source code to get a conservative loop iteration count. In this paper we describe our approach to analyze a source code to find hotspots. Our approach is based on estimating the iteration count of a loop using the polytope model for volume computation. This is then combined with the cyclomatic complexity measurement of the loop body. Both these metrics together provides an approximate idea of hotspots in a program and serves as a code transformation clue to programmers. We have run our tool on Rodinia benchmark applications and found encouraging results.