diff --git a/hipamd/src/hip_code_object.cpp b/hipamd/src/hip_code_object.cpp
index 047ce5b301..224f0549d1 100644
--- a/hipamd/src/hip_code_object.cpp
+++ b/hipamd/src/hip_code_object.cpp
@@ -1379,13 +1379,10 @@ const char* StatCO::getStatFuncName(const void* hostFunction) {
 }
 
 hipError_t StatCO::getStatFunc(hipFunction_t* hfunc, const void* hostFunction, int deviceId) {
-  amd::ScopedLock lock(sclock_);
-
   const auto it = functions_.find(hostFunction);
   if (it == functions_.end()) {
     return hipErrorInvalidSymbol;
   }
-
   return it->second->getStatFunc(hfunc, deviceId);
 }
 
diff --git a/hipamd/src/hip_global.cpp b/hipamd/src/hip_global.cpp
index 124b91090b..67f007efc0 100644
--- a/hipamd/src/hip_global.cpp
+++ b/hipamd/src/hip_global.cpp
@@ -90,7 +90,7 @@ DeviceVar::DeviceVar(std::string name,
 
 DeviceVar::~DeviceVar() {
   // device_ptr_ is being removed and its amd:Memory obj is being released/deleted during
-  // ihipFree in hip::StatCO::removeFatBinary however in DynCO path, it seems to bypass 
+  // ihipFree in hip::StatCO::removeFatBinary however in DynCO path, it seems to bypass
   // ihipFree and hence it needs to be removed+released here. In order to avoid issue with
   // StatCO, It is better to check if mem obj is found.
   if (amd::MemObjMap::FindMemObj(device_ptr_) !=  nullptr && amd_mem_obj_ != nullptr) {
@@ -155,21 +155,24 @@ bool Function::isValidDynFunc(const void* hfunc) {
 }
 
 hipError_t Function::getStatFunc(hipFunction_t* hfunc, int deviceId) {
-  guarantee(modules_ != nullptr, "Module not initialized");
-
-  if (dFunc_.size() != g_devices.size()) {
+  if (deviceId >= dFunc_.size()) {
     return hipErrorNoBinaryForGpu;
   }
-
+  if (dFunc_[deviceId] != nullptr) {
+    *hfunc = dFunc_[deviceId]->asHipFunction();
+    return hipSuccess;
+  }
+  amd::ScopedLock lock(fc_lock_);
+  // Check for the compiled kernel again, to make sure only one thread does compilation
+  if (dFunc_[deviceId] != nullptr) {
+    *hfunc = dFunc_[deviceId]->asHipFunction();
+    return hipSuccess;
+  }
   hipModule_t hmod = nullptr;
   IHIP_RETURN_ONFAIL((*modules_)->BuildProgram(deviceId));
   IHIP_RETURN_ONFAIL((*modules_)->GetModule(deviceId, &hmod));
-
-  if (dFunc_[deviceId] == nullptr) {
-    dFunc_[deviceId] = new DeviceFunc(name_, hmod);
-  }
+  dFunc_[deviceId] = new DeviceFunc(name_, hmod);
   *hfunc = dFunc_[deviceId]->asHipFunction();
-
   return hipSuccess;
 }
 
diff --git a/hipamd/src/hip_global.hpp b/hipamd/src/hip_global.hpp
index 22ba881809..65ecff6640 100644
--- a/hipamd/src/hip_global.hpp
+++ b/hipamd/src/hip_global.hpp
@@ -93,9 +93,10 @@ public:
   const std::string& name() const { return name_; }
 
 private:
-  std::vector<DeviceFunc*> dFunc_;  //DeviceFuncObj per Device
-  std::string name_;                //name of the func(not unique identifier)
-  FatBinaryInfo** modules_;      // static module where it is referenced
+  std::vector<DeviceFunc*> dFunc_;  //!< DeviceFuncObj per Device
+  std::string name_;                //!< name of the func(not unique identifier)
+  FatBinaryInfo** modules_;         //!< static module where it is referenced
+  amd::Monitor fc_lock_{true};      //!< Lock for the function create
 };
 
 class Var {