Description

Make more atomic functions available in warp, especially atomicCAS and atomicExch

Context

atomicCAS and atomicExch allow to create small spin locks/mutex constructs. While not being as efficient as e. g. atomicAdd, these spinlocks can simplify concurrent operations if used carefully (there should not be too much code that gets serialized by the spin lock). One application that came up recently would be atomic min/max with keeping track of the index (argmin, argmax).