Index: lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -139,7 +139,6 @@
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<AMDGPUArgumentUsageInfo>();
-    AU.addRequired<AMDGPUPerfHintAnalysis>();
     AU.addRequired<LegacyDivergenceAnalysis>();
 #ifdef EXPENSIVE_CHECKS
     AU.addRequired<DominatorTreeWrapperPass>();
Index: lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
+++ lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
@@ -29,13 +29,13 @@
   // except reserved size is not correctly aligned.
   const Function &F = MF.getFunction();
 
-  if (auto *Resolver = MF.getMMI().getResolver()) {
-    if (AMDGPUPerfHintAnalysis *PHA = static_cast<AMDGPUPerfHintAnalysis*>(
-          Resolver->getAnalysisIfAvailable(&AMDGPUPerfHintAnalysisID, true))) {
-      MemoryBound = PHA->isMemoryBound(&F);
-      WaveLimiter = PHA->needsWaveLimiter(&F);
-    }
-  }
+  Attribute MemBoundAttr = F.getFnAttribute("amdgpu-memory-bound");
+  MemoryBound = MemBoundAttr.isStringAttribute() &&
+                MemBoundAttr.getValueAsString() == "true";
+
+  Attribute WaveLimitAttr = F.getFnAttribute("amdgpu-wave-limiter");
+  WaveLimiter = WaveLimitAttr.isStringAttribute() &&
+                WaveLimitAttr.getValueAsString() == "true";
 
   CallingConv::ID CC = F.getCallingConv();
   if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL)
Index: lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.h
+++ lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.h
@@ -1,4 +1,4 @@
-//===- AMDGPUPerfHintAnalysis.h - analysis of functions memory traffic ----===//
+//===- AMDGPUPerfHintAnalysis.h ---- analysis of memory traffic -*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -14,18 +14,20 @@
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_MDGPUPERFHINTANALYSIS_H
 #define LLVM_LIB_TARGET_AMDGPU_MDGPUPERFHINTANALYSIS_H
+
+#include "llvm/Analysis/CallGraphSCCPass.h"
 #include "llvm/IR/ValueMap.h"
 #include "llvm/Pass.h"
 
 namespace llvm {
 
-struct AMDGPUPerfHintAnalysis : public FunctionPass {
+struct AMDGPUPerfHintAnalysis : public CallGraphSCCPass {
   static char ID;
 
 public:
-  AMDGPUPerfHintAnalysis() : FunctionPass(ID) {}
+  AMDGPUPerfHintAnalysis() : CallGraphSCCPass(ID) {}
 
-  bool runOnFunction(Function &F) override;
+  bool runOnSCC(CallGraphSCC &SCC) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesAll();
Index: lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.cpp
+++ lib/Target/AMDGPU/AMDGPUPerfHintAnalysis.cpp
@@ -17,6 +17,7 @@
 #include "Utils/AMDGPUBaseInfo.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
@@ -71,7 +72,7 @@
                  const TargetLowering *TLI_)
       : FIM(FIM_), DL(nullptr), TLI(TLI_) {}
 
-  void runOnFunction(Function &F);
+  bool runOnFunction(Function &F);
 
 private:
   struct MemAccessInfo {
@@ -100,7 +101,7 @@
 
   const TargetLowering *TLI;
 
-  void visit(const Function &F);
+  AMDGPUPerfHintAnalysis::FuncInfo *visit(const Function &F);
   static bool isMemBound(const AMDGPUPerfHintAnalysis::FuncInfo &F);
   static bool needLimitWave(const AMDGPUPerfHintAnalysis::FuncInfo &F);
 
@@ -202,12 +203,8 @@
   return false;
 }
 
-void AMDGPUPerfHint::visit(const Function &F) {
-  auto FIP = FIM.insert(std::make_pair(&F, AMDGPUPerfHintAnalysis::FuncInfo()));
-  if (!FIP.second)
-    return;
-
-  AMDGPUPerfHintAnalysis::FuncInfo &FI = FIP.first->second;
+AMDGPUPerfHintAnalysis::FuncInfo *AMDGPUPerfHint::visit(const Function &F) {
+  AMDGPUPerfHintAnalysis::FuncInfo &FI = FIM[&F];
 
   LLVM_DEBUG(dbgs() << "[AMDGPUPerfHint] process " << F.getName() << '\n');
 
@@ -233,7 +230,6 @@
         if (&F == Callee) // Handle immediate recursion
           continue;
 
-        visit(*Callee);
         auto Loc = FIM.find(Callee);
 
         assert(Loc != FIM.end() && "No func info");
@@ -256,36 +252,39 @@
       }
     }
   }
-}
 
-void AMDGPUPerfHint::runOnFunction(Function &F) {
-  if (FIM.find(&F) != FIM.end())
-    return;
+  return &FI;
+}
 
+bool AMDGPUPerfHint::runOnFunction(Function &F) {
   const Module &M = *F.getParent();
   DL = &M.getDataLayout();
 
-  visit(F);
-  auto Loc = FIM.find(&F);
+  if (F.hasFnAttribute("amdgpu-wave-limiter") &&
+      F.hasFnAttribute("amdgpu-memory-bound"))
+    return false;
 
-  assert(Loc != FIM.end() && "No func info");
-  LLVM_DEBUG(dbgs() << F.getName() << " MemInst: " << Loc->second.MemInstCount
-                    << '\n'
-                    << " IAMInst: " << Loc->second.IAMInstCount << '\n'
-                    << " LSMInst: " << Loc->second.LSMInstCount << '\n'
-                    << " TotalInst: " << Loc->second.InstCount << '\n');
+  const AMDGPUPerfHintAnalysis::FuncInfo *Info = visit(F);
 
-  auto &FI = Loc->second;
+  LLVM_DEBUG(dbgs() << F.getName() << " MemInst: " << Info->MemInstCount
+                    << '\n'
+                    << " IAMInst: " << Info->IAMInstCount << '\n'
+                    << " LSMInst: " << Info->LSMInstCount << '\n'
+                    << " TotalInst: " << Info->InstCount << '\n');
 
-  if (isMemBound(FI)) {
+  if (isMemBound(*Info)) {
     LLVM_DEBUG(dbgs() << F.getName() << " is memory bound\n");
     NumMemBound++;
+    F.addFnAttr("amdgpu-memory-bound", "true");
   }
 
-  if (AMDGPU::isEntryFunctionCC(F.getCallingConv()) && needLimitWave(FI)) {
+  if (AMDGPU::isEntryFunctionCC(F.getCallingConv()) && needLimitWave(*Info)) {
     LLVM_DEBUG(dbgs() << F.getName() << " needs limit wave\n");
     NumLimitWave++;
+    F.addFnAttr("amdgpu-wave-limiter", "true");
   }
+
+  return true;
 }
 
 bool AMDGPUPerfHint::isMemBound(const AMDGPUPerfHintAnalysis::FuncInfo &FI) {
@@ -364,17 +363,27 @@
 }
 } // namespace
 
-bool AMDGPUPerfHintAnalysis::runOnFunction(Function &F) {
+bool AMDGPUPerfHintAnalysis::runOnSCC(CallGraphSCC &SCC) {
   auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
   if (!TPC)
     return false;
 
   const TargetMachine &TM = TPC->getTM<TargetMachine>();
-  const TargetSubtargetInfo *ST = TM.getSubtargetImpl(F);
 
-  AMDGPUPerfHint Analyzer(FIM, ST->getTargetLowering());
-  Analyzer.runOnFunction(F);
-  return false;
+  bool Changed = false;
+  for (CallGraphNode *I : SCC) {
+    Function *F = I->getFunction();
+    if (!F || F->isDeclaration())
+      continue;
+
+    const TargetSubtargetInfo *ST = TM.getSubtargetImpl(*F);
+    AMDGPUPerfHint Analyzer(FIM, ST->getTargetLowering());
+
+    if (Analyzer.runOnFunction(*F))
+      Changed = true;
+  }
+
+  return Changed;
 }
 
 bool AMDGPUPerfHintAnalysis::isMemoryBound(const Function *F) const {
Index: lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -740,6 +740,8 @@
       EnableLowerKernelArguments)
     addPass(createAMDGPULowerKernelArgumentsPass());
 
+  addPass(&AMDGPUPerfHintAnalysisID);
+
   TargetPassConfig::addCodeGenPrepare();
 
   if (EnableLoadStoreVectorizer)