WSEAS Transactions on Computers
Print ISSN: 1109-2750, E-ISSN: 2224-2880
Volume 14, 2015
Enhancement of Phase Order Searching Using an Effective Tuning Strategy
Author:
Abstract: Modern compilers provide a large number of compiler options. Each option has two states called as enable or disable. Enabling on some options may improve or degrade the performance of the program. The objective is to increase the performance of the source program by means of adjusting the compiler options. The selection and ordering of the most efficient compiler options is required to improve the execution time, code and speed up. Some option combination may affect the execution time. The problem of getting the best combination of options is found by using modified genetic algorithm with the help of genetic operators. Finding the better ordering of best options will change the final erformance of the program. Ordering of the best combination of options obtained from selection algorithm is fed to phase order search algorithm. The existing algorithms such as combined elimination, batch elimination, branch and bound, push and pop with combined elimination algorithm, optimality random search and iterative random search algorithm are modified and the results are compared with the newly created combined push and pop with modified genetic algorithm. It is found that the combined push and pop with modified genetic algorithm shows better performance when compared to other algorithms. The phase searching algorithm shows increase in the program performance for some benchmark applications than combined push and pop with modified genetic algorithm. The experimental results show that 9% improvement in both tuning time and normalized tuning time. The speedup exhibit 11% increase over the set of benchmark applications. The combination of combined push and pop with modified genetic algorithm and phase order searching provide a 10% overall improvement in the program performance from the existing algorithms.
Search Articles
Keywords: Optimization, Combined Push and pop with modified Genetic Algorithm, Phase order searching
Pages: 411-419
WSEAS Transactions on Computers, ISSN / E-ISSN: 1109-2750 / 2224-2880, Volume 14, 2015, Art. #42