Meeting Banner
Abstract #3807

NUFFT: Fast Auto-Tuned GPU-Based Library

Teresa Ou1, Frank Ong1, Martin Uecker2, Laura Waller1, and Michael Lustig1

1University of California, Berkeley, Berkeley, CA, United States, 2University of Göttingen, Göttingen, Germany

We present a fast auto-tuned library for computing non-uniform fast Fourier Transform (NUFFT) on GPU. The library includes forward and adjoint NUFFT using precomputation-free and fully-precomputed methods, as well as Toeplitz-based operation for computing forward and adjoint NUFFT in a single step. Computation of NUFFT depends heavily on gridding parameters, desired accuracy, trajectory type, amount of undersampling, and level of precomputation. The library automatically chooses optimal gridding parameters and algorithms, and it can be easily extended to include implementations from other libraries. The library allows researchers to accelerate iterative reconstructions without the difficulties of choosing optimal parameters and algorithms.

This abstract and the presentation materials are available to members only; a login is required.

Join Here