Summary: LLVM IR should support these for all cases except for compare-exchange. Currently the code goes through an integer indirection for these cases. This PR changes the behavior to use atomics directly to the target memory type.
91 KiB
91 KiB