Wesley Maxey

> Can't we just mark those tests as unsupported for nvcc/nvrtc instead of deleting them? > > In a _very_ far away future I could imagine those tests passing one...

Are there any DVS virtuals or verification that the new configurations work? The additions look good to me, but I don't have the ability to test them.

I agree with the above, but this is just for satisfying an overload for now. We should figure out what the correct thing to do would be for this. I...

Hi @benchislett! Good question! For using `std::numbers`: It should be as simple as using a host compiler with the header and an NVCC that supports C++20 and concepts (this does...

What are the trade-offs? Wasn't there a point to using sequential consistency for device atomics? Something was missing I believe that warranted the usage... I'm not opposed to the change,...

I updated the issue to encompass all the errors that remain and provide which compilers are known to be working.

Hi Leo, I've been pondering this myself. One problem is that it's a device only type and being a heterogeneous library it would probably need to live outside `cuda::std::` as...

> do you mean the support of half complex will likely live in cuda::device:: or just cuda:: Where it would end up is probably open for discussion, but I don't...

@gonzalobg Is this good to get in?

Any thoughts on wrapping the removed lines with this: ``` NV_IF_TARGET( NV_IS_HOST, .... checks .... ) ``` This way we can keep the slow path intact on CPU? There could...