In this work, benchmark analyses of DEM program are performed using several types of compilers and processors. The effectiveness of program optimization as well as compiler and processor performance on the calculation time is discussed. As a result, the calculation time is reduced by about 10-18% by using the cache-optimized program when PGI, PathScale, NAG, Compaq or Fujitsu compilers are used. The inline expansion is effective for Intel (ver. 8.1 and 9.1), Fujitsu, PGI or PathScale, and especially works well under the dual-core processor. Since the reduction in calculation time strongly depends on the nature of compilers and processors, proper optimization of program for a given compiler and processor is necessary for a large-scale computation.
抄録全体を表示