摘要
With the rapid development of high-rise buildings and long-span structures in the recent years, high performance com- putation (HPC) is becoming more and more important, sometimes even crucial, for the design and construction of com- plex building structures. To satisfy the engineering requirements of HPC, a parallel FEA computing kernel, which is designed typically for the analysis of complex building structures, will be presented and illustrated in this paper. This kernel program is based on the Intel Math Kernel Library (MKL) and coded by FORTRAN 2008 syntax, which is a parallel computer language. To improve the capability and efficiency of the computing kernel program, the parallel concepts of modern FORTRAN, such as elemental procedure, do concurrent, etc., have been applied extensively in coding and the famous PARDISO solver in MKL has been called to solve the Large-sparse system of linear equations. The ultimate objective of developing the computing kernel is to make the personal computer have the ability to analysis large building structures up to ten million degree of freedoms (DOFs). Up to now, the linear static analysis and dynamic analysis have been achieved while the nonlinear analysis, including geometric and material nonlinearity, has not been finished yet. Therefore, the numerical examples in this paper will be concentrated on demonstrating the validity and efficiency of the linear analysis and modal analysis for large FE models, while ignoring the verification of the nonlinear analysis capabilities.
With the rapid development of high-rise buildings and long-span structures in the recent years, high performance com- putation (HPC) is becoming more and more important, sometimes even crucial, for the design and construction of com- plex building structures. To satisfy the engineering requirements of HPC, a parallel FEA computing kernel, which is designed typically for the analysis of complex building structures, will be presented and illustrated in this paper. This kernel program is based on the Intel Math Kernel Library (MKL) and coded by FORTRAN 2008 syntax, which is a parallel computer language. To improve the capability and efficiency of the computing kernel program, the parallel concepts of modern FORTRAN, such as elemental procedure, do concurrent, etc., have been applied extensively in coding and the famous PARDISO solver in MKL has been called to solve the Large-sparse system of linear equations. The ultimate objective of developing the computing kernel is to make the personal computer have the ability to analysis large building structures up to ten million degree of freedoms (DOFs). Up to now, the linear static analysis and dynamic analysis have been achieved while the nonlinear analysis, including geometric and material nonlinearity, has not been finished yet. Therefore, the numerical examples in this paper will be concentrated on demonstrating the validity and efficiency of the linear analysis and modal analysis for large FE models, while ignoring the verification of the nonlinear analysis capabilities.
