[llvm] [AMDGPU] Add waterfall intrinsics (PR #192409)

[Sebastian Neubauer via llvm-commits]

================
@@ -2693,6 +2693,73 @@ def int_amdgcn_cs_chain:
             ],
             [IntrConvergent, IntrNoReturn, ImmArg<ArgIndex<4>>]>;
 
+// Waterfall intrinsics are used to mark a region as requiring waterfall loops to
+// activate and deactivate lanes in a loop (and sometimes transform VGPR values
+// into SGPR values).
+// Best case (uniform index in VGPR) the loop will execute once, worst case (all
----------------
Flakebi wrote:

Referring to Jay’s comment: Maybe call it `value` instead of `index`?
Probably makes sense to still mention index as a motivating use-case in AMDGPUUsage.

(The rest of the sentence would become something like: worst case all lanes have different values)

https://github.com/llvm/llvm-project/pull/192409