[llvm] [ScalarizeMaskedMemIntr] Don't use a scalar mask on GPUs (PR #104842)

[Matt Arsenault via llvm-commits]

================
@@ -813,8 +834,9 @@ static void scalarizeMaskedExpandLoad(const DataLayout &DL, CallInst *CI,
   ModifiedDT = true;
 }
 
-static void scalarizeMaskedCompressStore(const DataLayout &DL, CallInst *CI,
-                                         DomTreeUpdater *DTU,
+static void scalarizeMaskedCompressStore(const DataLayout &DL,
+                                         const TargetTransformInfo &TTI,
----------------
arsenm wrote:

Maybe just pass in the branch divergence as a top level parameter instead of passing in TTI? The same thing is repeated in all of these cases 

https://github.com/llvm/llvm-project/pull/104842