This paper deals with the solution of an inverse time fractional diffusion equation described by a Caputo fractional derivative. Numerical simulations, involving large domains, give rise to a huge practical problem. Hence, by starting from an accurate meshless localized collocation method using radial basis functions (RBFs), here we propose a fast algorithm which exploits the GPU-CUDA capabilities. More in detail, we first developed a C code which uses the well-known numerical library LAPACK to perform basic linear algebra operations in order to implement an efficient sequential algorithm. Then we propose a GPU software based on ad hoc parallel CUDA-kernels and efficient usage of parallel numerical libraries available for GPUs. Performance analysis will show the reliability and the efficiency of the proposed parallel implementation.
A GPU-CUDA framework for solving a two-dimensional inverse anomalous diffusion problem
Galletti A.;Marcellino L.;
2020-01-01
Abstract
This paper deals with the solution of an inverse time fractional diffusion equation described by a Caputo fractional derivative. Numerical simulations, involving large domains, give rise to a huge practical problem. Hence, by starting from an accurate meshless localized collocation method using radial basis functions (RBFs), here we propose a fast algorithm which exploits the GPU-CUDA capabilities. More in detail, we first developed a C code which uses the well-known numerical library LAPACK to perform basic linear algebra operations in order to implement an efficient sequential algorithm. Then we propose a GPU software based on ad hoc parallel CUDA-kernels and efficient usage of parallel numerical libraries available for GPUs. Performance analysis will show the reliability and the efficiency of the proposed parallel implementation.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.