GPU Implementation for Wilson-θ Method

The GPU implementation for Wilson-θ method computing the dynamic response of structures is presented. We use a four-channel, two-dimensional texture to represent a matrix and a vector, and present a new texture reduction algorithm not requiring the texture size be the power of 2. A GPU based algorithm for solving the inverse of a matrix is also presented, where the aforementioned reduction algorithm is used to search the full pivot and the pivot´s row and column indices. GPU is used to accelerate the computations for the inverse of the effective stiffness matrix and the acceleration, velocity and displacement of the structure at every time step. The computations are compared against those on the CPU, proving the proposed GPU based algorithms have good accuracy and fast running time.