Split SimplifyCFG to run obscuring switch transforms only during last phase

[joerg]

Summary

The SimplifyCFG pass currently contains two transformations for switches that can obscure flow and inhibit other optimisations:

1. switch of constants is replaced with look-up table
2. sparse switches are replaced with transformation of the condition to make them dense

Most importantly, both transformations can benefit from DCE as a result of CVP and inlining. At the same time, they can inhibit inlining when run before the corresponding module pass as part of the early function cleanup. Neither transform looks like it is very useful early on, so split the existing simplifycfg pass into two and replace the last instance in the normal pass pipeline with the latesimplifycfg pass.

Initial testing of a version covering only point (1) of the list above fixes the regression in PR 32003. A LNT run by Mehdi Amini of the same versions shows:
MultiSource/Benchmarks/ASC_Sequoia/CrystalMk/CrystalMk 2.5% slower
SingleSource/Benchmarks/Misc/matmul_f64_4x4 34.2% faster
SingleSource/Benchmarks/Misc/fp-convert 3.0% faster
MultiSource/Benchmarks/Prolangs-C++/life/life 2.8% faster

[efriedma]

lib/Transforms/IPO/PassManagerBuilder.cpp

377

This is still interleaved with inlining; is that intentional?

[joerg]

LNT run with this version shows a consistent 1%-2% improvement in both compile and execution time for many instances. One execution time regression remains.

[joerg]

lib/Transforms/IPO/PassManagerBuilder.cpp

377

Inlining is done in line 398 or 476, this function is called from 485. Is my understanding of the pass ordering wrong?
The intention is definitely to defer the switch transforms until all inlining is done.

[efriedma]

The inliner is interleaved with the passes from addFunctionSimplificationPasses: first we simplify a leaf function, then we inline that function into another function, then we simplify the caller, etc. If you want to see this in action, try looking at the output of echo "int a(); int b() { return a(); } int c() { return b(); }" | clang -x c - -o - -S -O2 -mllvm -print-after-all |& less.

[mehdi_amini]

My favorite way is: echo "" | clang -x c - -c -mllvm -debug-pass=Structure -O2

Profile summary info
  ModulePass Manager
    [...]
    FunctionPass Manager
      [...]
      Simplify the CFG
    [...]
    Call Graph SCC Pass Manager
      Remove unused exception handling info
      Function Integration/Inlining
      Deduce function attributes
      FunctionPass Manager
        [...]
        Simplify the CFG
        [...]
        Simplify the CFG

The indentation shows the nesting, and the Call Graph SCC Pass Manager will schedule whatever is nested under it to run first on a callee before moving to the caller (SCC wise).

[joerg]

Move the full version out of the function simplification set into the late optimisation steps after all function inlining is done.

[mehdi_amini]

lib/Transforms/IPO/PassManagerBuilder.cpp

622

The comment is misleading, it seems that the only reason we run the pass here is for this purpose.

lib/Transforms/Scalar/SimplifyCFGPass.cpp

256

Why two passes instead of a parameter?

lib/Transforms/Utils/SimplifyCFG.cpp

173

Document.

5569–5574

Comment.

6030

Document the parameter in the doxygen.

[mehdi_amini]

lib/Transforms/IPO/PassManagerBuilder.cpp

592

What about this one?

[joerg]

lib/Transforms/IPO/PassManagerBuilder.cpp

592

I don't have a strong opinion on whether it should be done before or after SLP handling. I don't think the code is likely to be vectorised.

622

Well, we also want to do the usual transformations again after loop unrolling. But at this point, we are also commiting to the the lookup table conversions.

lib/Transforms/Scalar/SimplifyCFGPass.cpp

256

Two passes makes it easier to test it on the command line.

[jmolloy]

Looks good on my side, but please wait for Mehdi.

[mehdi_amini]

lib/Transforms/Scalar/SimplifyCFGPass.cpp

256

I can understand this aspect, but on the other hand that does not scale with adding more options, which is why we rarely do this. What other passes have the same approach?

[joerg]

lib/Transforms/Scalar/SimplifyCFGPass.cpp

256

I don't think we have a good example where parts of a pass are disabled. I could split them off into a separate pass, but that seems to be even worse. Normally we only have adjustable limits as configuration of passes.

[joerg]

Add comments on why the split pass is useful.

[mehdi_amini]

LGTM.

lib/Transforms/Scalar/SimplifyCFGPass.cpp

256

Normally we only have adjustable limits as configuration of passes ; I don't understand what you meant here?

What I was referring to was things like createFunctionInliningPass(Threshold);, createLoopRotatePass(SizeLevel == 2 ? 0 : -1), createSimpleLoopUnrollPass(OptLevel), and createLoopVectorizePass(true, LoopVectorize).

include/llvm-c/Transforms/Scalar.h

 void LLVMAddBitTrackingDCEPass(LLVMPassManagerRef PM);
 
 /** See llvm::createAlignmentFromAssumptionsPass function. */
 void LLVMAddAlignmentFromAssumptionsPass(LLVMPassManagerRef PM);
 
 /** See llvm::createCFGSimplificationPass function. */
 void LLVMAddCFGSimplificationPass(LLVMPassManagerRef PM);
 
+/** See llvm::createLateCFGSimplificationPass function. */
+void LLVMAddLateCFGSimplificationPass(LLVMPassManagerRef PM);
+
 /** See llvm::createDeadStoreEliminationPass function. */
 void LLVMAddDeadStoreEliminationPass(LLVMPassManagerRef PM);
 
 /** See llvm::createScalarizerPass function. */
 void LLVMAddScalarizerPass(LLVMPassManagerRef PM);
 
 /** See llvm::createMergedLoadStoreMotionPass function. */
 void LLVMAddMergedLoadStoreMotionPass(LLVMPassManagerRef PM);

include/llvm/InitializePasses.h

 void initializeBranchFolderPassPass(PassRegistry&);
 void initializeBranchProbabilityInfoWrapperPassPass(PassRegistry&);
 void initializeBranchRelaxationPass(PassRegistry&);
 void initializeBreakCriticalEdgesPass(PassRegistry&);
 void initializeCFGOnlyViewerLegacyPassPass(PassRegistry&);
 void initializeCFGPrinterLegacyPassPass(PassRegistry&);
 void initializeCFGOnlyPrinterLegacyPassPass(PassRegistry&);
 void initializeCFGSimplifyPassPass(PassRegistry&);
+void initializeLateCFGSimplifyPassPass(PassRegistry&);
 void initializeCFGViewerLegacyPassPass(PassRegistry&);
 void initializeCFLAndersAAWrapperPassPass(PassRegistry&);
 void initializeCFLSteensAAWrapperPassPass(PassRegistry&);
 void initializeCallGraphDOTPrinterPass(PassRegistry&);
 void initializeCallGraphPrinterLegacyPassPass(PassRegistry&);
 void initializeCallGraphViewerPass(PassRegistry&);
 void initializeCallGraphWrapperPassPass(PassRegistry &);
 void initializeCodeGenPreparePass(PassRegistry&);

include/llvm/LinkAllPasses.h

ForcePassLinking() {
       (void) llvm::createSCEVAAWrapperPass();
       (void) llvm::createTypeBasedAAWrapperPass();
       (void) llvm::createScopedNoAliasAAWrapperPass();
       (void) llvm::createBoundsCheckingPass();
       (void) llvm::createBreakCriticalEdgesPass();
       (void) llvm::createCallGraphDOTPrinterPass();
       (void) llvm::createCallGraphViewerPass();
       (void) llvm::createCFGSimplificationPass();
+      (void) llvm::createLateCFGSimplificationPass();
       (void) llvm::createCFLAndersAAWrapperPass();
       (void) llvm::createCFLSteensAAWrapperPass();
       (void) llvm::createStructurizeCFGPass();
       (void) llvm::createLibCallsShrinkWrapPass();
       (void) llvm::createConstantMergePass();
       (void) llvm::createConstantPropagationPass();
       (void) llvm::createCostModelAnalysisPass();
       (void) llvm::createDeadArgEliminationPass();

include/llvm/Transforms/Scalar.h

 // CFGSimplification - Merge basic blocks, eliminate unreachable blocks,
 // simplify terminator instructions, etc...
 //
 FunctionPass *createCFGSimplificationPass(
     int Threshold = -1, std::function<bool(const Function &)> Ftor = nullptr);
 
 //===----------------------------------------------------------------------===//
 //
+// LateCFGSimplification - Like CFGSimplification, but may also
+// convert switches to lookup tables.
+//
+FunctionPass *createLateCFGSimplificationPass(
+    int Threshold = -1, std::function<bool(const Function &)> Ftor = nullptr);
+
+//===----------------------------------------------------------------------===//
+//
 // FlattenCFG - flatten CFG, reduce number of conditional branches by using
 // parallel-and and parallel-or mode, etc...
 //
 FunctionPass *createFlattenCFGPass();
 
 //===----------------------------------------------------------------------===//
 //
 // CFG Structurization - Remove irreducible control flow

include/llvm/Transforms/Scalar/SimplifyCFG.h

 
 /// \brief A pass to simplify and canonicalize the CFG of a function.
 ///
 /// This pass iteratively simplifies the entire CFG of a function, removing
 /// unnecessary control flows and bringing it into the canonical form expected
 /// by the rest of the mid-level optimizer.
 class SimplifyCFGPass : public PassInfoMixin<SimplifyCFGPass> {
   int BonusInstThreshold;
+  bool LateSimplifyCFG;
 
 public:
-  /// \brief Construct a pass with the default thresholds.
+  /// \brief Construct a pass with the default thresholds
+  /// and switch optimizations.
   SimplifyCFGPass();
 
-  /// \brief Construct a pass with a specific bonus threshold.
-  SimplifyCFGPass(int BonusInstThreshold);
+  /// \brief Construct a pass with a specific bonus threshold
+  /// and optional switch optimizations.
+  SimplifyCFGPass(int BonusInstThreshold, bool LateSimplifyCFG);
 
   /// \brief Run the pass over the function.
   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
 };
 
 }
 
 #endif

include/llvm/Transforms/Utils/Local.h

 /// example, it adjusts branches to branches to eliminate the extra hop, it
 /// eliminates unreachable basic blocks, and does other "peephole" optimization
 /// of the CFG.  It returns true if a modification was made, possibly deleting
 /// the basic block that was pointed to. LoopHeaders is an optional input
 /// parameter, providing the set of loop header that SimplifyCFG should not
 /// eliminate.
 bool SimplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI,
                  unsigned BonusInstThreshold, AssumptionCache *AC = nullptr,
-                 SmallPtrSetImpl<BasicBlock *> *LoopHeaders = nullptr);
+                 SmallPtrSetImpl<BasicBlock *> *LoopHeaders = nullptr,
+                 bool LateSimplifyCFG = false);
 
 /// This function is used to flatten a CFG. For example, it uses parallel-and
 /// and parallel-or mode to collapse if-conditions and merge if-regions with
 /// identical statements.
 bool FlattenCFG(BasicBlock *BB, AliasAnalysis *AA = nullptr);
 
 /// If this basic block is ONLY a setcc and a branch, and if a predecessor
 /// branches to us and one of our successors, fold the setcc into the

lib/LTO/LTOCodeGenerator.cpp

void LTOCodeGenerator::initializeLTOPasses() {
   initializeReversePostOrderFunctionAttrsLegacyPassPass(R);
   initializeGlobalsAAWrapperPassPass(R);
   initializeLegacyLICMPassPass(R);
   initializeMergedLoadStoreMotionLegacyPassPass(R);
   initializeGVNLegacyPassPass(R);
   initializeMemCpyOptLegacyPassPass(R);
   initializeDCELegacyPassPass(R);
   initializeCFGSimplifyPassPass(R);
+  initializeLateCFGSimplifyPassPass(R);
 }
 
 void LTOCodeGenerator::setAsmUndefinedRefs(LTOModule *Mod) {
   const std::vector<StringRef> &undefs = Mod->getAsmUndefinedRefs();
   for (int i = 0, e = undefs.size(); i != e; ++i)
     AsmUndefinedRefs[undefs[i]] = 1;
 }
 

lib/Transforms/IPO/PassManagerBuilder.cpp

if (BBVectorize) {
         MPM.add(createLoopUnrollPass(OptLevel));
     }
   }
 
   if (LoadCombine)
     MPM.add(createLoadCombinePass());
 
   MPM.add(createAggressiveDCEPass());         // Delete dead instructions
   MPM.add(createCFGSimplificationPass()); // Merge & remove BBs
   // Clean up after everything.
   addInstructionCombiningPass(MPM);
   addExtensionsToPM(EP_Peephole, MPM);
 }
 
 void PassManagerBuilder::populateModulePassManager(
     legacy::PassManagerBase &MPM) {
   if (!PGOSampleUse.empty()) {
if (OptLevel > 1 && ExtraVectorizerPasses) {
     // common computations, hoist loop-invariant aspects out of any outer loop,
     // and unswitch the runtime checks if possible. Once hoisted, we may have
     // dead (or speculatable) control flows or more combining opportunities.
     MPM.add(createEarlyCSEPass());
     MPM.add(createCorrelatedValuePropagationPass());
     addInstructionCombiningPass(MPM);
     MPM.add(createLICMPass());
     MPM.add(createLoopUnswitchPass(SizeLevel || OptLevel < 3));
     MPM.add(createCFGSimplificationPass());
     addInstructionCombiningPass(MPM);
   }
 
   if (RunSLPAfterLoopVectorization) {
     if (SLPVectorize) {
       MPM.add(createSLPVectorizerPass());   // Vectorize parallel scalar chains.
       if (OptLevel > 1 && ExtraVectorizerPasses) {
         MPM.add(createEarlyCSEPass());
if (BBVectorize) {
 
       // BBVectorize may have significantly shortened a loop body; unroll again.
       if (!DisableUnrollLoops)
         MPM.add(createLoopUnrollPass(OptLevel));
     }
   }
 
   addExtensionsToPM(EP_Peephole, MPM);
-  MPM.add(createCFGSimplificationPass());
+  MPM.add(createLateCFGSimplificationPass()); // Switches to lookup tables
   addInstructionCombiningPass(MPM);
 
   if (!DisableUnrollLoops) {
     MPM.add(createLoopUnrollPass(OptLevel));    // Unroll small loops
 
     // LoopUnroll may generate some redundency to cleanup.
     addInstructionCombiningPass(MPM);
 

lib/Transforms/Scalar/Scalar.cpp

void llvm::initializeScalarOpts(PassRegistry &Registry) {
   initializePartiallyInlineLibCallsLegacyPassPass(Registry);
   initializeReassociateLegacyPassPass(Registry);
   initializeRegToMemPass(Registry);
   initializeRewriteStatepointsForGCPass(Registry);
   initializeSCCPLegacyPassPass(Registry);
   initializeIPSCCPLegacyPassPass(Registry);
   initializeSROALegacyPassPass(Registry);
   initializeCFGSimplifyPassPass(Registry);
+  initializeLateCFGSimplifyPassPass(Registry);
   initializeStructurizeCFGPass(Registry);
   initializeSinkingLegacyPassPass(Registry);
   initializeTailCallElimPass(Registry);
   initializeSeparateConstOffsetFromGEPPass(Registry);
   initializeSpeculativeExecutionLegacyPassPass(Registry);
   initializeStraightLineStrengthReducePass(Registry);
   initializeLoadCombinePass(Registry);
   initializePlaceBackedgeSafepointsImplPass(Registry);

 void LLVMAddAlignmentFromAssumptionsPass(LLVMPassManagerRef PM) {
   unwrap(PM)->add(createAlignmentFromAssumptionsPass());
 }
 
 void LLVMAddCFGSimplificationPass(LLVMPassManagerRef PM) {
   unwrap(PM)->add(createCFGSimplificationPass());
 }
 
+void LLVMAddLateCFGSimplificationPass(LLVMPassManagerRef PM) {
+  unwrap(PM)->add(createLateCFGSimplificationPass());
+}
+
 void LLVMAddDeadStoreEliminationPass(LLVMPassManagerRef PM) {
   unwrap(PM)->add(createDeadStoreEliminationPass());
 }
 
 void LLVMAddScalarizerPass(LLVMPassManagerRef PM) {
   unwrap(PM)->add(createScalarizerPass());
 }
 

lib/Transforms/Scalar/SimplifyCFGPass.cpp

static bool mergeEmptyReturnBlocks(Function &F) {
 
   return Changed;
 }
 
 /// Call SimplifyCFG on all the blocks in the function,
 /// iterating until no more changes are made.
 static bool iterativelySimplifyCFG(Function &F, const TargetTransformInfo &TTI,
                                    AssumptionCache *AC,
-                                   unsigned BonusInstThreshold) {
+                                   unsigned BonusInstThreshold,
+                                   bool LateSimplifyCFG) {
   bool Changed = false;
   bool LocalChange = true;
 
   SmallVector<std::pair<const BasicBlock *, const BasicBlock *>, 32> Edges;
   FindFunctionBackedges(F, Edges);
   SmallPtrSet<BasicBlock *, 16> LoopHeaders;
   for (unsigned i = 0, e = Edges.size(); i != e; ++i)
     LoopHeaders.insert(const_cast<BasicBlock *>(Edges[i].second));
 
   while (LocalChange) {
     LocalChange = false;
 
     // Loop over all of the basic blocks and remove them if they are unneeded.
     for (Function::iterator BBIt = F.begin(); BBIt != F.end(); ) {
-      if (SimplifyCFG(&*BBIt++, TTI, BonusInstThreshold, AC, &LoopHeaders)) {
+      if (SimplifyCFG(&*BBIt++, TTI, BonusInstThreshold, AC, &LoopHeaders, LateSimplifyCFG)) {
         LocalChange = true;
         ++NumSimpl;
       }
     }
     Changed |= LocalChange;
   }
   return Changed;
 }
 
 static bool simplifyFunctionCFG(Function &F, const TargetTransformInfo &TTI,
-                                AssumptionCache *AC, int BonusInstThreshold) {
+                                AssumptionCache *AC, int BonusInstThreshold,
+                                bool LateSimplifyCFG) {
   bool EverChanged = removeUnreachableBlocks(F);
   EverChanged |= mergeEmptyReturnBlocks(F);
-  EverChanged |= iterativelySimplifyCFG(F, TTI, AC, BonusInstThreshold);
+  EverChanged |= iterativelySimplifyCFG(F, TTI, AC, BonusInstThreshold,
+                                        LateSimplifyCFG);
 
   // If neither pass changed anything, we're done.
   if (!EverChanged) return false;
 
   // iterativelySimplifyCFG can (rarely) make some loops dead.  If this happens,
   // removeUnreachableBlocks is needed to nuke them, which means we should
   // iterate between the two optimizations.  We structure the code like this to
   // avoid rerunning iterativelySimplifyCFG if the second pass of
   // removeUnreachableBlocks doesn't do anything.
   if (!removeUnreachableBlocks(F))
     return true;
 
   do {
-    EverChanged = iterativelySimplifyCFG(F, TTI, AC, BonusInstThreshold);
+    EverChanged = iterativelySimplifyCFG(F, TTI, AC, BonusInstThreshold,
+                                         LateSimplifyCFG);
     EverChanged |= removeUnreachableBlocks(F);
   } while (EverChanged);
 
   return true;
 }
 
 SimplifyCFGPass::SimplifyCFGPass()
-    : BonusInstThreshold(UserBonusInstThreshold) {}
+    : BonusInstThreshold(UserBonusInstThreshold),
+      LateSimplifyCFG(true) {}
 
-SimplifyCFGPass::SimplifyCFGPass(int BonusInstThreshold)
-    : BonusInstThreshold(BonusInstThreshold) {}
+SimplifyCFGPass::SimplifyCFGPass(int BonusInstThreshold, bool LateSimplifyCFG)
+    : BonusInstThreshold(BonusInstThreshold),
+      LateSimplifyCFG(LateSimplifyCFG) {}
 
 PreservedAnalyses SimplifyCFGPass::run(Function &F,
                                        FunctionAnalysisManager &AM) {
   auto &TTI = AM.getResult<TargetIRAnalysis>(F);
   auto &AC = AM.getResult<AssumptionAnalysis>(F);
 
-  if (!simplifyFunctionCFG(F, TTI, &AC, BonusInstThreshold))
+  if (!simplifyFunctionCFG(F, TTI, &AC, BonusInstThreshold, LateSimplifyCFG))
     return PreservedAnalyses::all();
   PreservedAnalyses PA;
   PA.preserve<GlobalsAA>();
   return PA;
 }
 
 namespace {
-struct CFGSimplifyPass : public FunctionPass {
-  static char ID; // Pass identification, replacement for typeid
+struct BaseCFGSimplifyPass : public FunctionPass {
   unsigned BonusInstThreshold;
   std::function<bool(const Function &)> PredicateFtor;
+  bool LateSimplifyCFG;
 
-  CFGSimplifyPass(int T = -1,
-                  std::function<bool(const Function &)> Ftor = nullptr)
-      : FunctionPass(ID), PredicateFtor(std::move(Ftor)) {
+  BaseCFGSimplifyPass(int T, bool LateSimplifyCFG,
+                      std::function<bool(const Function &)> Ftor,
+                      char &ID)
+      : FunctionPass(ID), PredicateFtor(std::move(Ftor)),
+        LateSimplifyCFG(LateSimplifyCFG) {
     BonusInstThreshold = (T == -1) ? UserBonusInstThreshold : unsigned(T);
-    initializeCFGSimplifyPassPass(*PassRegistry::getPassRegistry());
   }
   bool runOnFunction(Function &F) override {
     if (skipFunction(F) || (PredicateFtor && !PredicateFtor(F)))
       return false;
 
     AssumptionCache *AC =
         &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
     const TargetTransformInfo &TTI =
         getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
-    return simplifyFunctionCFG(F, TTI, AC, BonusInstThreshold);
+    return simplifyFunctionCFG(F, TTI, AC, BonusInstThreshold, LateSimplifyCFG);
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<AssumptionCacheTracker>();
     AU.addRequired<TargetTransformInfoWrapperPass>();
     AU.addPreserved<GlobalsAAWrapperPass>();
   }
 };
+
+struct CFGSimplifyPass : public BaseCFGSimplifyPass {
+  static char ID; // Pass identification, replacement for typeid
+
+  CFGSimplifyPass(int T = -1,
+                  std::function<bool(const Function &)> Ftor = nullptr)
+                  : BaseCFGSimplifyPass(T, false, Ftor, ID) {
+    initializeCFGSimplifyPassPass(*PassRegistry::getPassRegistry());
+  }
+};
+
+struct LateCFGSimplifyPass : public BaseCFGSimplifyPass {
+  static char ID; // Pass identification, replacement for typeid
+
+  LateCFGSimplifyPass(int T = -1,
+                      std::function<bool(const Function &)> Ftor = nullptr)
+                      : BaseCFGSimplifyPass(T, true, Ftor, ID) {
+    initializeLateCFGSimplifyPassPass(*PassRegistry::getPassRegistry());
+  }
+};
 }
 
 char CFGSimplifyPass::ID = 0;
 INITIALIZE_PASS_BEGIN(CFGSimplifyPass, "simplifycfg", "Simplify the CFG", false,
                       false)
 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
 INITIALIZE_PASS_END(CFGSimplifyPass, "simplifycfg", "Simplify the CFG", false,
                     false)
 
+char LateCFGSimplifyPass::ID = 0;
+INITIALIZE_PASS_BEGIN(LateCFGSimplifyPass, "latesimplifycfg",
+                      "Simplify the CFG more aggressively", false, false)
+INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
+INITIALIZE_PASS_END(LateCFGSimplifyPass, "latesimplifycfg",
+                    "Simplify the CFG more aggressively", false, false)
+
 // Public interface to the CFGSimplification pass
 FunctionPass *
 llvm::createCFGSimplificationPass(int Threshold,
-                                  std::function<bool(const Function &)> Ftor) {
   return new CFGSimplifyPass(Threshold, std::move(Ftor));
 }
+
+// Public interface to the LateCFGSimplification pass
+FunctionPass *
+llvm::createLateCFGSimplificationPass(int Threshold, 
+                                  std::function<bool(const Function &)> Ftor) {
+  return new LateCFGSimplifyPass(Threshold, std::move(Ftor));
+}

lib/Transforms/Utils/SimplifyCFG.cpp

 };
 
 class SimplifyCFGOpt {
   const TargetTransformInfo &TTI;
   const DataLayout &DL;
   unsigned BonusInstThreshold;
   AssumptionCache *AC;
   SmallPtrSetImpl<BasicBlock *> *LoopHeaders;
+  // See comments in SimplifyCFGOpt::SimplifySwitch.
+  bool LateSimplifyCFG;
   Value *isValueEqualityComparison(TerminatorInst *TI);
   BasicBlock *GetValueEqualityComparisonCases(
       TerminatorInst *TI, std::vector<ValueEqualityComparisonCase> &Cases);
   bool SimplifyEqualityComparisonWithOnlyPredecessor(TerminatorInst *TI,
                                                      BasicBlock *Pred,
                                                      IRBuilder<> &Builder);
   bool FoldValueComparisonIntoPredecessors(TerminatorInst *TI,
                                            IRBuilder<> &Builder);
 
   bool SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder);
   bool SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder);
   bool SimplifySingleResume(ResumeInst *RI);
   bool SimplifyCommonResume(ResumeInst *RI);
   bool SimplifyCleanupReturn(CleanupReturnInst *RI);
   bool SimplifyUnreachable(UnreachableInst *UI);
   bool SimplifySwitch(SwitchInst *SI, IRBuilder<> &Builder);
   bool SimplifyIndirectBr(IndirectBrInst *IBI);
   bool SimplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder);
   bool SimplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder);
 
 public:
   SimplifyCFGOpt(const TargetTransformInfo &TTI, const DataLayout &DL,
                  unsigned BonusInstThreshold, AssumptionCache *AC,
-                 SmallPtrSetImpl<BasicBlock *> *LoopHeaders)
+                 SmallPtrSetImpl<BasicBlock *> *LoopHeaders,
+                 bool LateSimplifyCFG)
       : TTI(TTI), DL(DL), BonusInstThreshold(BonusInstThreshold), AC(AC),
-        LoopHeaders(LoopHeaders) {}
+        LoopHeaders(LoopHeaders), LateSimplifyCFG(LateSimplifyCFG) {}
 
   bool run(BasicBlock *BB);
 };
 
 } // end anonymous namespace
 
 /// Return true if it is safe to merge these two
 /// terminator instructions together.
if (EliminateDeadSwitchCases(SI, AC, DL))
     return SimplifyCFG(BB, TTI, BonusInstThreshold, AC) | true;
 
   if (SwitchToSelect(SI, Builder, AC, DL, TTI))
     return SimplifyCFG(BB, TTI, BonusInstThreshold, AC) | true;
 
   if (ForwardSwitchConditionToPHI(SI))
     return SimplifyCFG(BB, TTI, BonusInstThreshold, AC) | true;
 
-  if (SwitchToLookupTable(SI, Builder, DL, TTI))
+  // The conversion from switch to lookup tables results in difficult
+  // to analyze code and makes pruning branches much harder.
+  // This is a problem of the switch expression itself can still be
+  // restricted as a result of inlining or CVP. There only apply this
+  // transformation during late steps of the optimisation chain.
+  if (LateSimplifyCFG && SwitchToLookupTable(SI, Builder, DL, TTI))
     return SimplifyCFG(BB, TTI, BonusInstThreshold, AC) | true;
 
   if (ReduceSwitchRange(SI, Builder, DL, TTI))
     return SimplifyCFG(BB, TTI, BonusInstThreshold, AC) | true;
 
   return false;
 }
 
bool SimplifyCFGOpt::run(BasicBlock *BB) {
 
   return Changed;
 }
 
 /// This function is used to do simplification of a CFG.
 /// For example, it adjusts branches to branches to eliminate the extra hop,
 /// eliminates unreachable basic blocks, and does other "peephole" optimization
 /// of the CFG.  It returns true if a modification was made.
 ///
 bool llvm::SimplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI,
                        unsigned BonusInstThreshold, AssumptionCache *AC,
-                       SmallPtrSetImpl<BasicBlock *> *LoopHeaders) {
+                       SmallPtrSetImpl<BasicBlock *> *LoopHeaders,
+                       bool LateSimplifyCFG) {
   return SimplifyCFGOpt(TTI, BB->getModule()->getDataLayout(),
-                        BonusInstThreshold, AC, LoopHeaders)
+                        BonusInstThreshold, AC, LoopHeaders, LateSimplifyCFG)
       .run(BB);
 }

test/Transforms/SimplifyCFG/ARM/switch-to-lookup-table.ll

-; RUN: opt -S -simplifycfg -mtriple=arm -relocation-model=static    < %s | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE
-; RUN: opt -S -simplifycfg -mtriple=arm -relocation-model=pic       < %s | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE
-; RUN: opt -S -simplifycfg -mtriple=arm -relocation-model=ropi      < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
-; RUN: opt -S -simplifycfg -mtriple=arm -relocation-model=rwpi      < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
-; RUN: opt -S -simplifycfg -mtriple=arm -relocation-model=ropi-rwpi < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
+; RUN: opt -S -latesimplifycfg -mtriple=arm -relocation-model=static    < %s | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE
+; RUN: opt -S -latesimplifycfg -mtriple=arm -relocation-model=pic       < %s | FileCheck %s --check-prefix=CHECK --check-prefix=ENABLE
+; RUN: opt -S -latesimplifycfg -mtriple=arm -relocation-model=ropi      < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
+; RUN: opt -S -latesimplifycfg -mtriple=arm -relocation-model=rwpi      < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
+; RUN: opt -S -latesimplifycfg -mtriple=arm -relocation-model=ropi-rwpi < %s | FileCheck %s --check-prefix=CHECK --check-prefix=DISABLE
 
 ; CHECK:       @{{.*}} = private unnamed_addr constant [3 x i32] [i32 1234, i32 5678, i32 15532]
 ; ENABLE:      @{{.*}} = private unnamed_addr constant [3 x i32*] [i32* @c1, i32* @c2, i32* @c3]
 ; DISABLE-NOT: @{{.*}} = private unnamed_addr constant [3 x i32*] [i32* @c1, i32* @c2, i32* @c3]
 ; ENABLE:      @{{.*}} = private unnamed_addr constant [3 x i32*] [i32* @g1, i32* @g2, i32* @g3]
 ; DISABLE-NOT: @{{.*}} = private unnamed_addr constant [3 x i32*] [i32* @g1, i32* @g2, i32* @g3]
 ; ENABLE:      @{{.*}} = private unnamed_addr constant [3 x i32 (i32, i32)*] [i32 (i32, i32)* @f1, i32 (i32, i32)* @f2, i32 (i32, i32)* @f3]
 ; DISABLE-NOT: @{{.*}} = private unnamed_addr constant [3 x i32 (i32, i32)*] [i32 (i32, i32)* @f1, i32 (i32, i32)* @f2, i32 (i32, i32)* @f3]

test/Transforms/SimplifyCFG/CoveredLookupTable.ll

-; RUN: opt -simplifycfg -S %s | FileCheck %s
+; RUN: opt -latesimplifycfg -S %s | FileCheck %s
 ; rdar://15268442
 
 target datalayout = "e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin12.0.0"
 
 ; CHECK-LABEL: define i3 @coveredswitch_test(
 ; CHECK: entry:
 ; CHECK-NEXT: sub i3 %input, -4

test/Transforms/SimplifyCFG/X86/switch-covered-bug.ll

-; RUN: opt -S -simplifycfg < %s -mtriple=x86_64-apple-darwin12.0.0 | FileCheck %s
+; RUN: opt -S -latesimplifycfg < %s -mtriple=x86_64-apple-darwin12.0.0 | FileCheck %s
 ; rdar://17887153
 target datalayout = "e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin12.0.0"
 
 ; When we have a covered lookup table, make sure we don't delete PHINodes that
 ; are cached in PHIs.
 ; CHECK-LABEL: @test
 ; CHECK: entry:

test/Transforms/SimplifyCFG/X86/switch-table-bug.ll

-; RUN: opt -S -simplifycfg < %s -mtriple=x86_64-apple-darwin12.0.0 | FileCheck %s
+; RUN: opt -S -latesimplifycfg < %s -mtriple=x86_64-apple-darwin12.0.0 | FileCheck %s
 ; rdar://17735071
 target datalayout = "e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin12.0.0"
 
 ; When tableindex can't fit into i2, we should extend the type to i3.
 ; CHECK-LABEL: @_TFO6reduce1E5toRawfS0_FT_Si
 ; CHECK: entry:
 ; CHECK-NEXT: sub i2 %0, -2

test/Transforms/SimplifyCFG/X86/switch_to_lookup_table.ll

-; RUN: opt < %s -simplifycfg -S -mtriple=x86_64-unknown-linux-gnu | FileCheck %s
+; RUN: opt < %s -latesimplifycfg -S -mtriple=x86_64-unknown-linux-gnu | FileCheck %s
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 ; The table for @f
 ; CHECK: @switch.table = private unnamed_addr constant [7 x i32] [i32 55, i32 123, i32 0, i32 -1, i32 27, i32 62, i32 1]
 
 ; The float table for @h
sw.epilog:
   br i1 %cmp, label %if.then, label %if.end
 if.then: br label %return
 if.end: br label %return
 return:
   %retval.0 = phi i32 [ %r.0, %if.then ], [ 100, %if.end ]
   ret i32 %retval.0
 ; CHECK-LABEL: @reuse_cmp2(
 ; CHECK: entry:
-; CHECK-NEXT: %switch.tableidx = sub i32 %x, 0
-; CHECK-NEXT: [[C:%.+]] = icmp ult i32 %switch.tableidx, 4
+; CHECK-NEXT: %switch = icmp ult i32 %x, 4
+; CHECK-NEXT: %x. = select i1 %switch, i32 %x, i32 4
+; CHECK-NEXT: [[C:%.+]] = icmp ne i32 %x., 4
 ; CHECK:      [[R:%.+]] = select i1 [[C]], i32 {{.*}}, i32 100
 ; CHECK-NEXT: ret i32 [[R]]
 }
 
 ; Cannot reuse the table range compare, because the default value is the same
 ; as one of the case values.
 define i32 @no_reuse_cmp(i32 %x) {
 entry:

test/Transforms/SimplifyCFG/rangereduce.ll

-; RUN: opt < %s -simplifycfg -S | FileCheck %s
+; RUN: opt < %s -latesimplifycfg -S | FileCheck %s
 
 target datalayout = "e-n32"
 
 ; CHECK-LABEL: @test1
 ; CHECK: %[[SUB:.*]] = sub i32 %a, 97
 ; CHECK: %[[LSHR:.*]] = lshr i32 %[[SUB]], 2
 ; CHECK: %[[SHL:.*]] = shl i32 %[[SUB]], 30
 ; CHECK: %[[OR:.*]] = or i32 %[[LSHR]], %[[SHL]]