Carl Pearson

It would be great if the format check printed what it thought the diff should be! core and kernels do this

Created here, I think https://github.com/james-see/iptcinfo3/blob/dfbfd902f64205ad42517dce288b46e06bc4b585/iptcinfo3.py#L700 It seems like it's intended to not create this file if you do something like `.save('overwrite')` or `.save(options='overwrite')` but that doesn't work for me.

The `0.1` and (in your change) `131072` basically means it downloads a 128KiB chunk ever 0.1s, or ~1.3 MB/s. @drdarshan may have asked the suitsparse people how hard they want...

I've been poking around with this: In the SpGEMM, it seems that `Kokkos::atomic_add(addr, val);` always results in `*addr = 0`, (or possibly `*addr` unchanged, e.g. `Kokkos::atomic_add` is a no-op). however,...

Reimplementing `alignPtr` as ```c++ template KOKKOS_INLINE_FUNCTION T *alignPtr(InPtr p) { std::uintptr_t ptrVal = reinterpret_cast(p); while (ptrVal % alignof(T)) { ++ptrVal; } return reinterpret_cast(ptrVal); } ``` seems to make the SpGEMM...

```c++ unsigned int f1(unsigned int i, unsigned int align) // today { return ((i + align - 1) & (~(align - 1))); } unsigned int f2(unsigned int i, unsigned int...

This looks newer than related https://github.com/kokkos/kokkos-kernels/issues/395

There are several examples that use this too that could probably use `SharedSpace` instead.

I propose working towards three related things 1. a consistent policy: * every record is conceptually {`namespace`, `component`, `fields`..., `value`} thing. * In Core we would just use the `Kokkos`...

Hi, thank you. In commit 3c4a1be7b2ff793fabf2b299dfa673e2fae8e86f, I no longer see the crash. However, what I would like to do is see how the MPI implementation handles types on the GPU,...