Evan Weinberg

Dslash host verification cleanup

2

This PR is a collection of cleanups to the host verification code relevant for the various dslashes. This includes: - Adding doxygen to the Wilson and DWF routines - Removing...

Specifying a specific type of host space in a DualView

14

A DualView currently uses the default `HostSpace` for its host mirror, which isn't always ideal. It would be useful if one could override the default host space, with a simple...

**Summary** This PR introduces work batching within the core kernels in the Kokkos implementation of SNAP: `ComputeUi`, `ComputeYi`, and `ComputeFusedDeidrj`. The batching of work is designed to increase instruction-level parallelism,...

The current interface doesn't support initial guesses because of a sign convention difference between MILC and QUDA. The fix is flipping the sign on an initial guess before handing it...

Update `cudaMemPrefetchAsync` to use `cudaMemPrefetchAsync_v2`

1

`cudaMemPrefetchAsync` currently accepts either a GPU device ID for moving data to devices, or `-1` to prefetch data to the host. The newer routine `cudaMemPrefetchAsync_v2` gives more granularity over migration...

There are a few known cases (especially with Arnoldi) where the Ritz values (computed in double precision) meet the convergence criteria for the eigensolver, but the actual eigenvalues from the...