[llvm] s390x: pattern match saturated truncation (PR #155377)

[Ulrich Weigand via llvm-commits]

================
@@ -1269,11 +1282,24 @@ multiclass IntegerMinMaxVectorOps<ValueType type, SDPatternOperator cmph,
   }
 }
 
+multiclass SignedIntegerMinMaxVectorOps<ValueType type, SDPatternOperator cmph,
+                                  Instruction min, Instruction max> {
+  defm : IntegerMinMaxVectorOps<type, cmph, min, max>;
+
+  let Predicates = [FeatureVector] in {
+    def : Pat<(type (smax VR128:$x, VR128:$y)),
----------------
uweigand wrote:

This can be done simply by using `smax` instead of `null_frag` in the VMX... patterns.   (You can still keep the `IntegerMinMaxVectorOps` as they are to match additional opportunities if necessary ... but doesn't common DAGCombiner code recognize them already?)

https://github.com/llvm/llvm-project/pull/155377