This commit takes advantage of the new `load.async.to.lds` intrinsic in order to add an `async` mode to `gather_to_lds`. In this mode, completion of the load needs to be managed with `asyncmark` and `wait.asyncmark` intrinsics instead of being implicitly derived by alias analysis. This commit adds the flag, a lowering for it, and updates tests. Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Multi-Level Intermediate Representation
See https://mlir.llvm.org/ for more information.