[DO NOT MERGE] Re-enable -fdetach flag

include/llvm/Transforms/Tapir.h

@@ -62,7 +62,7 @@ FunctionPass *createSpawnUnswitchPass();
 //
 // PromoteDetachToCilk
 //
-ModulePass *createLowerTapirToTargetPass(tapir::TapirTarget*);
+FunctionPass *createLowerTapirToTargetPass(tapir::TapirTarget*);
 
 } // End llvm namespace
 

lib/Transforms/IPO/PassManagerBuilder.cpp

@@ -268,6 +268,10 @@ void PassManagerBuilder::populateFunctionPassManager(
 
   if (OptLevel == 0) return;
 
+  if (tapirTarget && DisableTapirOpts) {// -fdetach
+    FPM.add(createLowerTapirToTargetPass(tapirTarget));
+  }
+
   addInitialAliasAnalysisPasses(FPM);
 
   FPM.add(createCFGSimplificationPass());
@@ -322,6 +326,11 @@ void PassManagerBuilder::addPGOInstrPasses(legacy::PassManagerBase &MPM) {
 }
 void PassManagerBuilder::addFunctionSimplificationPasses(
     legacy::PassManagerBase &MPM) {
+
+      if (tapirTarget && DisableTapirOpts) {// -fdetach
+        MPM.add(createLowerTapirToTargetPass(tapirTarget));
+      }
+
   // Start of function pass.
   // Break up aggregate allocas, using SSAUpdater.
   MPM.add(createSROAPass());
@@ -412,6 +421,14 @@ void PassManagerBuilder::addFunctionSimplificationPasses(
 // void PassManagerBuilder::prepopulateModulePassManager(
 void PassManagerBuilder::populateModulePassManager(
     legacy::PassManagerBase &MPM) {
+
+      bool TapirHasBeenLowered = (tapirTarget == nullptr);
+
+      if (tapirTarget && DisableTapirOpts) {// -fdetach
+        MPM.add(createLowerTapirToTargetPass(tapirTarget));
+        TapirHasBeenLowered = true;
+      }
+
   if (!PGOSampleUse.empty()) {
     MPM.add(createPruneEHPass());
     MPM.add(createSampleProfileLoaderPass(PGOSampleUse));
@@ -485,12 +502,6 @@ void PassManagerBuilder::populateModulePassManager(
     DisableUnrollLoops = true;
 
   bool RerunAfterTapirLowering = false;
-  bool TapirHasBeenLowered = (tapirTarget == nullptr);
-
-  if (tapirTarget && DisableTapirOpts) {// -fdetach
-    MPM.add(createLowerTapirToTargetPass(tapirTarget));
-    TapirHasBeenLowered = true;
-  }
 
   do {
     RerunAfterTapirLowering =

lib/Transforms/Tapir/OpenMPABI.cpp

@@ -12,7 +12,7 @@
 // runtime system.  This interface does the low-level dirty work of passes
 // such as LowerToCilk.
 //
-//===----------------------------------------------------------------------===//
+//===--------------------------------------------------------
 
 #include "llvm/Transforms/Tapir/OpenMPABI.h"
 #include "llvm/IR/DebugInfoMetadata.h"
@@ -202,7 +202,7 @@ CallInst *emitRuntimeCall(Value *Callee, ArrayRef<Value *> Args,
   return call;
 }
 
-Value *getThreadID(Function *F, IRBuilder<> &IRBuilder) {
+Value *getThreadID(Function *F, IRBuilder<> &IRBuilder0) {
   Value *ThreadID = nullptr;
   auto I = OpenMPThreadIDLoadMap.find(F);
   if (I != OpenMPThreadIDLoadMap.end()) {
@@ -216,9 +216,9 @@ Value *getThreadID(Function *F, IRBuilder<> &IRBuilder) {
   if (I2 != OpenMPThreadIDAllocaMap.end()) {
     DataLayout DL(F->getParent());
     auto Alloca = I2->second;
-    auto ThreadIDAddrs = IRBuilder.CreateLoad(Alloca);
+    auto ThreadIDAddrs = IRBuilder0.CreateLoad(Alloca);
     ThreadIDAddrs->setAlignment(DL.getTypeAllocSize(ThreadIDAddrs->getType()));
-    ThreadID = IRBuilder.CreateLoad(ThreadIDAddrs);
+    ThreadID = IRBuilder0.CreateLoad(ThreadIDAddrs);
     ((LoadInst *)ThreadID)
         ->setAlignment(DL.getTypeAllocSize(ThreadID->getType()));
     auto &Elem = OpenMPThreadIDLoadMap.FindAndConstruct(F);
@@ -228,7 +228,8 @@ Value *getThreadID(Function *F, IRBuilder<> &IRBuilder) {
 
   auto GTIDFn = createRuntimeFunction(
       OpenMPRuntimeFunction::OMPRTL__kmpc_global_thread_num, F->getParent());
-  ThreadID = emitRuntimeCall(GTIDFn, {DefaultOpenMPLocation}, "", IRBuilder);
+      IRBuilder<> b(F->getEntryBlock().getFirstNonPHIOrDbgOrLifetime());
+  ThreadID = emitRuntimeCall(GTIDFn, {DefaultOpenMPLocation}, "", b);
   auto &Elem = OpenMPThreadIDLoadMap.FindAndConstruct(F);
   Elem.second = ThreadID;
 

lib/Transforms/Tapir/TapirToTarget.cpp

@@ -27,11 +27,11 @@ using namespace llvm::tapir;
 
 namespace {
 
-struct LowerTapirToTarget : public ModulePass {
+struct LowerTapirToTarget : public FunctionPass {
   static char ID; // Pass identification, replacement for typeid
   TapirTarget* tapirTarget;
   explicit LowerTapirToTarget(TapirTarget* tapirTarget = nullptr)
-      : ModulePass(ID), tapirTarget(tapirTarget) {
+      : FunctionPass(ID), tapirTarget(tapirTarget) {
     assert(tapirTarget);
     initializeLowerTapirToTargetPass(*PassRegistry::getPassRegistry());
   }
@@ -40,7 +40,7 @@ struct LowerTapirToTarget : public ModulePass {
     return "Simple Lowering of Tapir to Target ABI";
   }
 
-  bool runOnModule(Module &M) override;
+  bool runOnFunction(Function &F) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<AssumptionCacheTracker>();
@@ -130,6 +130,7 @@ SmallVectorImpl<Function *>
   if (!changed) return NewHelpers;
 
   if (verifyFunction(F, &errs())) {
+    F.getParent()->dump();
     DEBUG(F.dump());
     assert(0);
   }
@@ -142,46 +143,34 @@ SmallVectorImpl<Function *>
   return NewHelpers;
 }
 
-bool LowerTapirToTarget::runOnModule(Module &M) {
-  if (skipModule(M))
+bool LowerTapirToTarget::runOnFunction(Function &F) {
+  if (skipFunction(F))
     return false;
+  bool shouldGo = false;
 
-  // Add functions that detach to the work list.
-  SmallVector<Function *, 4> WorkList;
-  for (Function &F : M)
-    for (BasicBlock &BB : F)
-      if (isa<DetachInst>(BB.getTerminator())) {
-        WorkList.push_back(&F);
-        break;
-      }
 
-  if (WorkList.empty())
+  for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) {
+    if (DetachInst* DI = dyn_cast_or_null<DetachInst>(I->getTerminator())) {
+      shouldGo = true;
+    } else if (SyncInst* SI = dyn_cast_or_null<SyncInst>(I->getTerminator())) {
+      shouldGo = true;
+    }
+  }
+  if (shouldGo == false) {
     return false;
+  }
 
-  bool Changed = false;
-  std::unique_ptr<SmallVectorImpl<Function *>> NewHelpers;
-  while (!WorkList.empty()) {
-    // Process the next function.
-    Function *F = WorkList.back();
-    WorkList.pop_back();
-    DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>(*F).getDomTree();
+    DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
     AssumptionCacheTracker &ACT = getAnalysis<AssumptionCacheTracker>();
-    NewHelpers.reset(processFunction(*F, DT, ACT.getAssumptionCache(*F)));
-    Changed |= !NewHelpers->empty();
-    // Check the generated helper functions to see if any need to be processed,
-    // that is, to see if any of them themselves detach a subtask.
-    for (Function *Helper : *NewHelpers)
-      for (BasicBlock &BB : *Helper)
-        if (isa<DetachInst>(BB.getTerminator()))
-          WorkList.push_back(Helper);
-  }
+    bool Changed = processFunction(F, DT, ACT.getAssumptionCache(F))->size() > 0;
+
   return Changed;
 }
 
 // createLowerTapirToTargetPass - Provide an entry point to create this pass.
 //
 namespace llvm {
-ModulePass *createLowerTapirToTargetPass(TapirTarget* tapirTarget) {
+FunctionPass *createLowerTapirToTargetPass(TapirTarget* tapirTarget) {
   return new LowerTapirToTarget(tapirTarget);
 }
 }