[llvm] [NVPTX] Fold vector FMAs in the IR peephole pass (PR #224018)

[Alex MacLean via llvm-commits]

================
@@ -125,8 +125,9 @@ static bool foldFMA(Function &F) {
       if (!BI->hasAllowContract())
         continue;
 
-      // Only float and double are supported.
-      if (!BI->getType()->isFloatTy() && !BI->getType()->isDoubleTy())
+      // Float, double, and vectors thereof are supported.
+      Type *ScalarTy = BI->getType()->getScalarType();
+      if (!ScalarTy->isFloatTy() && !ScalarTy->isDoubleTy())
         continue;
----------------
AlexMaclean wrote:

At this point, what does this check even exclude? F16 and BF16? These both support FMA too so perhaps we should just remove it.

https://github.com/llvm/llvm-project/pull/224018