Atomics wishlist

[adayton1]

Builtin atomics

• [x] Investigate using types with guaranteed bitwidth if available (e.g. uint32_t and uint64_t)
• [x] Investigate supporting builtin atomics on smaller types (e.g. uint8_t and uint16_t)
• [x] Investigate builtin atomics where available (e.g. __atomic_fetch_add) instead of implementing in terms of atomicCAS
• [ ] Consider adding the builtin atomics to test configurations (i.e. Windows)

Openmp atomics

• [x] Add implementations for atomicMin and atomicMax using OpenMP 5.1 (https://developers.redhat.com/articles/2022/11/17/new-features-openmp-51-and-openmp-52#atomic_directive_extensions_in_openmp_5_1)
• [ ] Report bug in openmp implementation on tioga preventing atomicCAS from being implemented using OpenMP 5.1
• [ ] Consider adding the openmp atomics to test configurations (i.e. Windows)

All atomics

• [x] Remove use of volatile
• [x] Add atomicLoad and atomicStore

[adayton1]

I'm working on adding atomicLoad and atomicStore, and I noticed that atomicStore is very similar to atomicExchange, which we already have implemented. Should we add atomicStore as an alias to atomicExchange? Or should we make it different by not returning the old value? I'm leaning toward making it an alias to atomicExchange.

EDIT: I made atomicStore a different function that does not return the old value. Jason pointed out some platform intrinsics that don't return a value are more efficient.