Avoid unnecessary reordering for nodes with several consumers before …

…output

src/plugins/intel_cpu/src/graph.cpp

@@ -227,6 +227,15 @@ void Graph::Replicate(const std::shared_ptr<const ov::Model> &model) {
         const auto parentEdge = outputNode->getParentEdgeAt(0);
         const auto parent = parentEdge->getParent();
         parent->setOriginalOutputPrecisionAtPort(parentEdge->getInputNum(), precToSet);
+        // If the parent has consumers except Output, precToSet is propagated to consumer's inputs
+        // to avoid precision mismatch (which leads to reaorder insertion and unnecessary performance overheads)
+        if (parent->getChildEdges().size() > 1) {
+            for (size_t i = 0; i < parent->getChildEdges().size(); ++i) {
+                const auto childEdge = parent->getChildEdgeAt(i);
+                const auto child = childEdge->getChild();
+                child->setOriginalInputPrecisionAtPort(childEdge->getOutputNum(), precToSet);
+            }
+        }
     }
 }