[llvm] [AArch64][Codegen]Transform saturating smull to sqdmulh (PR #143671)

[Nashe Mncube via llvm-commits]

================
@@ -20918,6 +20928,79 @@ static SDValue performBuildVectorCombine(SDNode *N,
   return SDValue();
 }
 
+// A special combine for the sqdmulh family of instructions.
+// smin( sra ( mul( sext v0, sext v1 ) ), SHIFT_AMOUNT ),
+// SATURATING_VAL ) can be reduced to sext(sqdmulh(...))
+static SDValue trySQDMULHCombine(SDNode *N, SelectionDAG &DAG) {
+
+  if (N->getOpcode() != ISD::TRUNCATE)
----------------
nasherm wrote:

I'm not sure if you mean starting the combine at SMIN, or calling the combine from a "performSMINCombine" call. I don't believe the latter is possible (at least in a similar way to what I have already) in our test case because the DAG is defined with the SMIN nodes being child nodes of a the `truncate` call. 

I could be understanding this wrong but it seems that the `performDAGCombine` call pattern matches on parent (root?) nodes of a DAG. Meaning we don't necessarily know if there's an SMIN child node when attempting combines without traversing the child nodes.

However, if you meant just starting the `trySQDMULHCombine` call from SMIN, then yes that's easy.

https://github.com/llvm/llvm-project/pull/143671