NEWS: GQD 1.1 has been released, see Downloads

We provide two different libraries for extended precision on the GPU.

GQD: This library has implemented double-double (31 decimal digits) and quad-double (62 decimal digits) precision on the GPU. Due to built-in vectors available in CUDA, this library has a very good portability and is also easy to use. Most operations are implemented using function overriding. The algorithms mainly refer to the QD library on the CPU.

GARPREC: This library supports arbitrary precision using varied number of native precision numbers. Due to the coalesced access adopted on the GPU for the efficiency, it uses a different memory layout from the traditional one. Therefore, corresponding modifications are necessary compared with the native precision. The algorithms are from the ARPREC library on the CPU.

Our test cases also need QD and ARPREC library support, which can be downloaded here: http://crd.lbl.gov/~dhbailey/mpdist/