[llvm] [NVPTX] Enhance `mul.wide` and `mad.wide` peepholes (PR #150477)

[Alex MacLean via llvm-commits]

================
@@ -5408,6 +5408,53 @@ static SDValue PerformREMCombine(SDNode *N,
   return SDValue();
 }
 
+// (any_extend|sign_extend|zero_extend (mul|shl) x, y) -> (mul.wide x, y)
+static SDValue
+PerformExtendMULWIDECombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
----------------
AlexMaclean wrote:

I think there is a `doMulWide` option we should check here.

https://github.com/llvm/llvm-project/pull/150477