This is an old project which developed single and double precision approximations of the inverse error function which are particularly efficient for GPU execution. The implementations in NVIDIA's math library are now based on these, and they are also part of the Apache Commons Math Library.
This is a recent project with a similar goal, to approximate the inverse of the Poisson cumulative distribution function.
This project investigates how best to implement explicit and implicit discretisations of the 1D Black-Scholes PDE. Julien Demouth and Jeremy Appleyard (NVIDIA) contributed to this work.
This project investigates how best to implement explicit and implicit discretisations of the 3D Black-Scholes PDE. Istvan Reguly has contributed to this work.
Numerical algorithms are sometimes defined as a task DAG (Directed Acyclic Graph) as a number of tasks with specified interdependencies. This project supports the dynamic execution of such algorithms on a GPU. Unlike other software such as StarPU (which is used in MAGMA) and OmpSs which maintain and update the task list on the CPU, this software does it on the GPU, which is likely to give better performance when the tasks are very small.