Robert Rusch

Yeah, that's what I was getting at with the second part of my comment. My current solution is to return the references wrapped such that later async_copy calls can consume...

How is pinned host memory done right now? Is that what the DeviceCopy trait indicates?

Additionally, implementing the unsafe wrapper layer is done now, save for the test `async_copy_device_to_host` not actually synchronizing for some reason and failing some of the time. I think the stack...

Alright, so AsyncMemcpy requires pinned memory, but also runtime errors properly if given not page-locked memory, so we don't necessarily need to mark that in the wrapper.

The error I was mentioning only appears when multiple tests are run at the same time. EDIT: Nevermind, it appears rarely when run alone.

Alright to sum up my current thoughts on this. 1. AsyncMemcpy requires locked memory, thus we can async copy from and to devices using only DeviceFoo, UnifiedFoo, and LockedFoo. 2....

Previously, I thought `AsyncCopyDestination` might be able to take `T: DeviceCopy` references as values for `copy_from` and `copy_to`, just like `CopyDestination`. However, it appears not using page-locked memory will throw...

Let me finish up 1. It's pretty much done with a PR up right now, I just need to rebase it and clean a bit more, but have been slow...

See #20 for the PR I'm writing.

Previously, my test failures entirely vanished by switching from pageable to page-locked memory, but sure, I'll look into it. I can see it possibly resulting from some other issue that...