diff --git a/docs/CONTRIBUTORS b/docs/CONTRIBUTORS
index 0b1072cb2..9f14c65ee 100644
--- a/docs/CONTRIBUTORS
+++ b/docs/CONTRIBUTORS
@@ -210,6 +210,7 @@ William D. Jones
 Wilson Snyder
 Xi Zhang
 Yan Xu
+Yangyu Chen
 Yinan Xu
 Yoda Lee
 Yossi Nivin
diff --git a/src/V3EmitCFunc.cpp b/src/V3EmitCFunc.cpp
index dc6393bbf..b2b26c502 100644
--- a/src/V3EmitCFunc.cpp
+++ b/src/V3EmitCFunc.cpp
@@ -445,8 +445,12 @@ void EmitCFunc::emitDereference(AstNode* nodep, const string& pointer) {
         putns(nodep, pointer.substr(2, pointer.length() - 3));
         puts(".");
     } else {
-        putns(nodep, pointer);
-        puts("->");
+        if (pointer == "vlSelf" && m_usevlSelfRef) {
+            puts("vlSelfRef.");
+        } else {
+            putns(nodep, pointer);
+            puts("->");
+        }
     }
 }
 
diff --git a/src/V3EmitCFunc.h b/src/V3EmitCFunc.h
index fa2a102c7..6dea21240 100644
--- a/src/V3EmitCFunc.h
+++ b/src/V3EmitCFunc.h
@@ -148,6 +148,7 @@ class EmitCFunc VL_NOT_FINAL : public EmitCConstInit {
 protected:
     EmitCLazyDecls m_lazyDecls;  // Visitor for emitting lazy declarations
     bool m_useSelfForThis = false;  // Replace "this" with "vlSelf"
+    bool m_usevlSelfRef = false;  // Use vlSelfRef reference instead of vlSelf pointer
     const AstNodeModule* m_modp = nullptr;  // Current module being emitted
     const AstCFunc* m_cfuncp = nullptr;  // Current function being emitted
     bool m_instantiatesOwnProcess = false;
@@ -343,6 +344,21 @@ public:
             }
         }
 
+        if (m_useSelfForThis) {
+            m_usevlSelfRef = true;
+            /*
+             * Using reference to the vlSelf pointer will help the C++
+             * compiler to have dereferenceable hints, which can help to
+             * reduce the need for branch instructions in the generated
+             * code to allow the compiler to generate load store after the
+             * if condition (including short-circuit evaluation)
+             * speculatively and also reduce the data cache pollution when
+             * executing in the wrong path to make verilator-generated code
+             * run faster.
+             */
+            puts("auto &vlSelfRef = std::ref(*vlSelf).get();\n");
+        }
+
         if (nodep->initsp()) {
             putsDecoration(nodep, "// Init\n");
             iterateAndNextConstNull(nodep->initsp());
@@ -358,6 +374,8 @@ public:
             iterateAndNextConstNull(nodep->finalsp());
         }
 
+        m_usevlSelfRef = false;
+
         puts("}\n");
         if (nodep->ifdef() != "") puts("#endif  // " + nodep->ifdef() + "\n");
     }