diff --git a/llvm/lib/Target/NVPTX/CMakeLists.txt b/llvm/lib/Target/NVPTX/CMakeLists.txt
--- a/llvm/lib/Target/NVPTX/CMakeLists.txt
+++ b/llvm/lib/Target/NVPTX/CMakeLists.txt
@@ -12,6 +12,7 @@
 
 set(NVPTXCodeGen_sources
   NVPTXAllocaHoisting.cpp
+  NVPTXAtomicLower.cpp
   NVPTXAsmPrinter.cpp
   NVPTXAssignValidGlobalNames.cpp
   NVPTXFrameLowering.cpp
diff --git a/llvm/lib/Target/NVPTX/NVPTXAtomicLower.h b/llvm/lib/Target/NVPTX/NVPTXAtomicLower.h
new file mode 100644
--- /dev/null
+++ b/llvm/lib/Target/NVPTX/NVPTXAtomicLower.h
@@ -0,0 +1,22 @@
+//===-- NVPTXAtomicLower.h - Lower atomics of local memory ------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+//  Lower atomics of local memory to simple load/stores
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_NVPTX_NVPTXATOMICLOWER_H
+#define LLVM_LIB_TARGET_NVPTX_NVPTXATOMICLOWER_H
+
+namespace llvm {
+class FunctionPass;
+
+extern FunctionPass *createNVPTXAtomicLowerPass();
+} // end namespace llvm
+
+#endif
diff --git a/llvm/lib/Target/NVPTX/NVPTXAtomicLower.cpp b/llvm/lib/Target/NVPTX/NVPTXAtomicLower.cpp
new file mode 100644
--- /dev/null
+++ b/llvm/lib/Target/NVPTX/NVPTXAtomicLower.cpp
@@ -0,0 +1,84 @@
+//===-- NVPTXAtomicLower.cpp - Lower atomics of local memory ----*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+//  Lower atomics of local memory to simple load/stores
+//
+//===----------------------------------------------------------------------===//
+
+#include "NVPTXAtomicLower.h"
+#include "llvm/CodeGen/StackProtector.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+
+#include "MCTargetDesc/NVPTXBaseInfo.h"
+using namespace llvm;
+
+namespace {
+// Hoisting the alloca instructions in the non-entry blocks to the entry
+// block.
+class NVPTXAtomicLower : public FunctionPass {
+public:
+  static char ID; // Pass ID
+  NVPTXAtomicLower() : FunctionPass(ID) {}
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+  }
+
+  StringRef getPassName() const override {
+    return "NVPTX lower atomics of local memory";
+  }
+
+  bool runOnFunction(Function &function) override;
+};
+} // namespace
+
+bool NVPTXAtomicLower::runOnFunction(Function &F) {
+  SmallVector<AtomicRMWInst *, 3> V;
+  for (auto &BB : F) {
+    for (auto &I : BB) {
+      if (auto N = dyn_cast<AtomicRMWInst>(&I)) {
+        if (N->getPointerAddressSpace() == ADDRESS_SPACE_LOCAL &&
+            N->getOrdering() == AtomicOrdering::Monotonic) {
+          V.push_back(N);
+        }
+      }
+    }
+  }
+  bool changed = false;
+  for (auto I : V) {
+    if (I->getOperation() == llvm::AtomicRMWInst::BinOp::FAdd) {
+      IRBuilder<> B(I);
+      changed = true;
+      auto L = B.CreateLoad(I->getType(), I->getPointerOperand());
+      L->setAtomic(I->getOrdering(), I->getSyncScopeID());
+      auto SI = B.CreateStore(B.CreateFAdd(L, I->getValOperand()),
+                              I->getPointerOperand());
+      SI->setAtomic(I->getOrdering(), I->getSyncScopeID());
+      I->eraseFromParent();
+    }
+  }
+  return changed;
+}
+
+char NVPTXAtomicLower::ID = 0;
+
+namespace llvm {
+void initializeNVPTXAtomicLowerPass(PassRegistry &);
+}
+
+INITIALIZE_PASS(
+    NVPTXAtomicLower, "nvptx-atomic-lower",
+    "Hoisting alloca instructions in non-entry blocks to the entry block",
+    false, false)
+
+FunctionPass *llvm::createNVPTXAtomicLowerPass() {
+  return new NVPTXAtomicLower();
+}
diff --git a/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp b/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
--- a/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp
@@ -13,6 +13,7 @@
 #include "NVPTXTargetMachine.h"
 #include "NVPTX.h"
 #include "NVPTXAllocaHoisting.h"
+#include "NVPTXAtomicLower.h"
 #include "NVPTXLowerAggrCopies.h"
 #include "NVPTXTargetObjectFile.h"
 #include "NVPTXTargetTransformInfo.h"
@@ -252,6 +253,7 @@
   addPass(createSROAPass());
   addPass(createNVPTXLowerAllocaPass());
   addPass(createInferAddressSpacesPass());
+  addPass(createNVPTXAtomicLowerPass());
 }
 
 void NVPTXPassConfig::addStraightLineScalarOptimizationPasses() {