Should the user be able to specify the exact type of copy to dispatch to?

[ZelboK]

I noticed that for SM80 the default dispatch for copy is to LDGSTS instructions. This requires you to do 3 syncs for correctness. Is it more intuitive to have the copy be able take another type that would specify exactly which you want to dispatch to? So if you want LDGSTS, you'd have to specify so as a type.

I think I know how I would implement this PR after messing around with the code, it shouldn't be too difficult. But I don't know if I'm missing any context.

[ZelboK]

NVM, I see that this can already be done.

TiledCopy copyB = make_tiled_copy(Copy_Atom<SM80_CP_ASYNC_CACHEALWAYS<uint128_t>, TB>{}, ....

I presume that it still shouldn't be the default to dispatch to but I'll let a maintainer answer that for me.