Index: llvm/lib/Target/AMDGPU/AMDGPU.h
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPU.h
+++ llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -232,6 +232,16 @@
   PreservedAnalyses run(Function &, FunctionAnalysisManager &);
 };
 
+class AMDGPULowerKernelArgumentsPass
+    : public PassInfoMixin<AMDGPULowerKernelArgumentsPass> {
+private:
+  TargetMachine &TM;
+
+public:
+  AMDGPULowerKernelArgumentsPass(TargetMachine &TM) : TM(TM){};
+  PreservedAnalyses run(Function &, FunctionAnalysisManager &);
+};
+
 FunctionPass *createAMDGPUAnnotateUniformValues();
 
 ModulePass *createAMDGPUPrintfRuntimeBinding();
Index: llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp
@@ -24,7 +24,7 @@
 
 namespace {
 
-class AMDGPULowerKernelArguments : public FunctionPass{
+class AMDGPULowerKernelArguments : public FunctionPass {
 public:
   static char ID;
 
@@ -55,14 +55,11 @@
   return InsPt;
 }
 
-bool AMDGPULowerKernelArguments::runOnFunction(Function &F) {
+static bool lowerKernelArguments(Function &F, const TargetMachine &TM) {
   CallingConv::ID CC = F.getCallingConv();
   if (CC != CallingConv::AMDGPU_KERNEL || F.arg_empty())
     return false;
 
-  auto &TPC = getAnalysis<TargetPassConfig>();
-
-  const TargetMachine &TM = TPC.getTM<TargetMachine>();
   const GCNSubtarget &ST = TM.getSubtarget<GCNSubtarget>(F);
   LLVMContext &Ctx = F.getParent()->getContext();
   const DataLayout &DL = F.getParent()->getDataLayout();
@@ -232,6 +229,12 @@
   return true;
 }
 
+bool AMDGPULowerKernelArguments::runOnFunction(Function &F) {
+  auto &TPC = getAnalysis<TargetPassConfig>();
+  const TargetMachine &TM = TPC.getTM<TargetMachine>();
+  return lowerKernelArguments(F, TM);
+}
+
 INITIALIZE_PASS_BEGIN(AMDGPULowerKernelArguments, DEBUG_TYPE,
                       "AMDGPU Lower Kernel Arguments", false, false)
 INITIALIZE_PASS_END(AMDGPULowerKernelArguments, DEBUG_TYPE, "AMDGPU Lower Kernel Arguments",
@@ -242,3 +245,16 @@
 FunctionPass *llvm::createAMDGPULowerKernelArgumentsPass() {
   return new AMDGPULowerKernelArguments();
 }
+
+PreservedAnalyses
+AMDGPULowerKernelArgumentsPass::run(Function &F, FunctionAnalysisManager &AM) {
+  bool Changed = lowerKernelArguments(F, TM);
+  if (Changed) {
+    // TODO: Preserves a lot more.
+    PreservedAnalyses PA;
+    PA.preserveSet<CFGAnalyses>();
+    return PA;
+  }
+
+  return PreservedAnalyses::all();
+}
Index: llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -665,6 +665,10 @@
           PM.addPass(AMDGPUCodeGenPreparePass(*this));
           return true;
         }
+        if (PassName == "amdgpu-lower-kernel-arguments") {
+          PM.addPass(AMDGPULowerKernelArgumentsPass(*this));
+          return true;
+        }
         return false;
       });
 
Index: llvm/test/CodeGen/AMDGPU/lower-kernargs.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/lower-kernargs.ll
+++ llvm/test/CodeGen/AMDGPU/lower-kernargs.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
-; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -o - -amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,HSA %s
-; RUN: opt -mtriple=amdgcn-- -S -o - -amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,MESA %s
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -o - -passes=amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,HSA %s
+; RUN: opt -mtriple=amdgcn-- -S -o - -passes=amdgpu-lower-kernel-arguments %s | FileCheck -check-prefixes=GCN,MESA %s
 
 target datalayout = "A5"