Hariprasad Kannan

Error while compiling gpu_spmv

1

There are some deprecated functions being called (cusparseScsrmv, cusparseDcsrmv) in gpu_spmv.cu. I am using CUDA 11.2. Suppose these calls are modified according to latest functions available in cuSPARSE, will the...

Incorrect call to kokkos_malloc while initializing the Threads backend

3

This bug has come to light while running the CI checks for PR [7676](https://github.com/kokkos/kokkos/pull/7676). In this PR, we are checking if memory management functions (`kokkos_malloc`, `kokkos_realloc`, `kokkos_free`) are called properly....

We have been experimenting with the GMRES solver with ILU(0) preconditioner. We would like to know how a non-symmetric matrix is re-ordered before computing the ILU(0) preconditioner? Could someone point...

MDRangePolicy: Nested loops w/o tiles; Host backends (begin with Serial)

8

Issue [8652](https://github.com/kokkos/kokkos/issues/8652) reported a performance issue concerning MDRangePolicy in Serial backend. [Two approaches](https://github.com/kokkos/kokkos/issues/8652#issuecomment-3595876656) were tried to iterate through the elements of a View via MDRangePolicy. The approach based on a...

Extend the work done in #8164 to allow loop unrolling (along grid stride) semantics in Views with rank 2 for Cuda backend. Update: This work will be resumed after more...

#8164 added loop unrolling capability in RangePolicy for loops for Cuda backend. The loop unrolling happens along the grid strides. This is important for ensuring coalesced memory accesses. There are...