English
资源说明：CUDA-accelerated assembly of a 2D Laplacian Matrix operator based on linear 3-node triangular elements
This repository holds the files necessary to test a crude CUDA-accelerated linear 2D finite element based laplacian operator assembly routine. The main driver program is "laplace_test.m" The driver program reads a finite element mesh contained in "ex_5_9_2.msh" and loads the data into a MATLAB structure. The driver then runs two functions that are written to take the mesh structure and construct a 2D linear finite element Laplacian operator using 3-node linear triangular elements. The first function accomplishes this task using MATLAB's native M-language only. The second function accomplishes this using a compiled C/C++ program that uses CUDA on NVIDIA GPUs to do the matrix assembly which turns out to be much faster albiet in single precision arithmetic that seems to accumulate a lot of error. (sorry) The goal was to test the FEM operator assembly routine with CUDA acceleration. As written, the bulk of the time for the CUDA version is taken up writing an input file to disk describing the element geometry. Actual matrix assembly takes up a small fraction of that time. The results of this code and the experience gained writing it are behind a not-so-recent blog post which can be viewed at: http://noiceinmyscotchplease.blogspot.com/2011/12/cuda-accelerated-sparse-matrix-assembly.html
本源码包内暂不包含可直接显示的源代码文件，请下载源码包。
