[DA][TTI][AMDGPU] Add option to select GPUDA with TTI

Summary: Enable the new diveregence analysis by default for AMDGPU.

Reviewers: rampitec, nhaehnle, arsenm

Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73049

include/llvm/Analysis/LegacyDivergenceAnalysis.h

@@ -54,7 +54,8 @@ public:
 
 private:
   // Whether analysis should be performed by GPUDivergenceAnalysis.
-  bool shouldUseGPUDivergenceAnalysis(const Function &F) const;
+  bool shouldUseGPUDivergenceAnalysis(const Function &F,
+                                      const TargetTransformInfo &TTI) const;
 
   // (optional) handle to new DivergenceAnalysis
   std::unique_ptr<GPUDivergenceAnalysis> gpuDA;

include/llvm/Analysis/TargetTransformInfo.h

@@ -342,6 +342,10 @@ public:
   /// branches.
   bool hasBranchDivergence() const;
 
+  /// Return true if the target prefers to use GPU divergence analysis to
+  /// replace the legacy version.
+  bool useGPUDivergenceAnalysis() const;
+
   /// Returns whether V is a source of divergence.
   ///
   /// This function provides the target-dependent information for
@@ -1198,6 +1202,7 @@ public:
   virtual int
   getUserCost(const User *U, ArrayRef<const Value *> Operands) = 0;
   virtual bool hasBranchDivergence() = 0;
+  virtual bool useGPUDivergenceAnalysis() = 0;
   virtual bool isSourceOfDivergence(const Value *V) = 0;
   virtual bool isAlwaysUniform(const Value *V) = 0;
   virtual unsigned getFlatAddressSpace() = 0;
@@ -1452,6 +1457,7 @@ public:
     return Impl.getUserCost(U, Operands);
   }
   bool hasBranchDivergence() override { return Impl.hasBranchDivergence(); }
+  bool useGPUDivergenceAnalysis() override { return Impl.useGPUDivergenceAnalysis(); }
   bool isSourceOfDivergence(const Value *V) override {
     return Impl.isSourceOfDivergence(V);
   }

include/llvm/Analysis/TargetTransformInfoImpl.h

@@ -152,6 +152,8 @@ public:
 
   bool hasBranchDivergence() { return false; }
 
+  bool useGPUDivergenceAnalysis() { return false; }
+
   bool isSourceOfDivergence(const Value *V) { return false; }
 
   bool isAlwaysUniform(const Value *V) { return false; }

include/llvm/CodeGen/BasicTTIImpl.h

@@ -207,6 +207,8 @@ public:
 
   bool hasBranchDivergence() { return false; }
 
+  bool useGPUDivergenceAnalysis() { return false; }
+
   bool isSourceOfDivergence(const Value *V) { return false; }
 
   bool isAlwaysUniform(const Value *V) { return false; }

lib/Analysis/LegacyDivergenceAnalysis.cpp

@@ -301,14 +301,13 @@ FunctionPass *llvm::createLegacyDivergenceAnalysisPass() {
 void LegacyDivergenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.addRequired<DominatorTreeWrapperPass>();
   AU.addRequired<PostDominatorTreeWrapperPass>();
-  if (UseGPUDA)
-    AU.addRequired<LoopInfoWrapperPass>();
+  AU.addRequired<LoopInfoWrapperPass>();
   AU.setPreservesAll();
 }
 
 bool LegacyDivergenceAnalysis::shouldUseGPUDivergenceAnalysis(
-    const Function &F) const {
-  if (!UseGPUDA)
+    const Function &F, const TargetTransformInfo &TTI) const {
+  if (!(UseGPUDA || TTI.useGPUDivergenceAnalysis()))
     return false;
 
   // GPUDivergenceAnalysis requires a reducible CFG.
@@ -337,7 +336,7 @@ bool LegacyDivergenceAnalysis::runOnFunction(Function &F) {
   auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
   auto &PDT = getAnalysis<PostDominatorTreeWrapperPass>().getPostDomTree();
 
-  if (shouldUseGPUDivergenceAnalysis(F)) {
+  if (shouldUseGPUDivergenceAnalysis(F, TTI)) {
     // run the new GPU divergence analysis
     auto &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
     gpuDA = std::make_unique<GPUDivergenceAnalysis>(F, DT, PDT, LI, TTI);

lib/Analysis/TargetTransformInfo.cpp

@@ -212,6 +212,10 @@ bool TargetTransformInfo::hasBranchDivergence() const {
   return TTIImpl->hasBranchDivergence();
 }
 
+bool TargetTransformInfo::useGPUDivergenceAnalysis() const {
+  return TTIImpl->useGPUDivergenceAnalysis();
+}
+
 bool TargetTransformInfo::isSourceOfDivergence(const Value *V) const {
   return TTIImpl->isSourceOfDivergence(V);
 }

lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

@@ -69,6 +69,11 @@ static cl::opt<unsigned> UnrollThresholdIf(
   cl::desc("Unroll threshold increment for AMDGPU for each if statement inside loop"),
   cl::init(150), cl::Hidden);
 
+static cl::opt<bool> UseLegacyDA(
+  "amdgpu-use-legacy-divergence-analysis",
+  cl::desc("Enable legacy divergence analysis for AMDGPU"),
+  cl::init(false), cl::Hidden);
+
 static bool dependsOnLocalPhi(const Loop *L, const Value *Cond,
                               unsigned Depth = 0) {
   const Instruction *I = dyn_cast<Instruction>(Cond);
@@ -601,6 +606,11 @@ static bool isArgPassedInSGPR(const Argument *A) {
   }
 }
 
+/// \returns true if the new GPU divergence analysis is enabled.
+bool GCNTTIImpl::useGPUDivergenceAnalysis() const {
+  return !UseLegacyDA;
+}
+
 /// \returns true if the result of the value could potentially be
 /// different across workitems in a wavefront.
 bool GCNTTIImpl::isSourceOfDivergence(const Value *V) const {

lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h

@@ -136,6 +136,7 @@ public:
       HasFP32Denormals(ST->hasFP32Denormals(F)) { }
 
   bool hasBranchDivergence() { return true; }
+  bool useGPUDivergenceAnalysis() const;
 
   void getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                TTI::UnrollingPreferences &UP);