Merge pull request #39 from schweitzpgi/release_60

merge LLVM.org changes, fix to Fortran debug onto Release 60

Author: sscalpone

CMakeLists.txt

@@ -24,7 +24,7 @@ if(NOT DEFINED LLVM_VERSION_MINOR)
   set(LLVM_VERSION_MINOR 0)
 endif()
 if(NOT DEFINED LLVM_VERSION_PATCH)
-  set(LLVM_VERSION_PATCH 0)
+  set(LLVM_VERSION_PATCH 1)
 endif()
 if(NOT DEFINED LLVM_VERSION_SUFFIX)
   set(LLVM_VERSION_SUFFIX "")

include/llvm/CodeGen/MachineBasicBlock.h

@@ -449,6 +449,13 @@ class MachineBasicBlock
   /// Replace successor OLD with NEW and update probability info.
   void replaceSuccessor(MachineBasicBlock *Old, MachineBasicBlock *New);
 
+  /// Copy a successor (and any probability info) from original block to this
+  /// block's. Uses an iterator into the original blocks successors.
+  ///
+  /// This is useful when doing a partial clone of successors. Afterward, the
+  /// probabilities may need to be normalized.
+  void copySuccessor(MachineBasicBlock *Orig, succ_iterator I);
+
   /// Transfers all the successors from MBB to this machine basic block (i.e.,
   /// copies all the successors FromMBB and remove all the successors from
   /// FromMBB).

include/llvm/CodeGen/TargetInstrInfo.h

@@ -421,7 +421,8 @@ class TargetInstrInfo : public MCInstrInfo {
   /// Build the equivalent inputs of a REG_SEQUENCE for the given \p MI
   /// and \p DefIdx.
   /// \p [out] InputRegs of the equivalent REG_SEQUENCE. Each element of
-  /// the list is modeled as <Reg:SubReg, SubIdx>.
+  /// the list is modeled as <Reg:SubReg, SubIdx>. Operands with the undef
+  /// flag are not added to this list.
   /// E.g., REG_SEQUENCE %1:sub1, sub0, %2, sub1 would produce
   /// two elements:
   /// - %1:sub1, sub0
@@ -446,7 +447,8 @@ class TargetInstrInfo : public MCInstrInfo {
   /// - %1:sub1, sub0
   ///
   /// \returns true if it is possible to build such an input sequence
-  /// with the pair \p MI, \p DefIdx. False otherwise.
+  /// with the pair \p MI, \p DefIdx and the operand has no undef flag set.
+  /// False otherwise.
   ///
   /// \pre MI.isExtractSubreg() or MI.isExtractSubregLike().
   ///
@@ -465,7 +467,8 @@ class TargetInstrInfo : public MCInstrInfo {
   /// - InsertedReg: %1:sub1, sub3
   ///
   /// \returns true if it is possible to build such an input sequence
-  /// with the pair \p MI, \p DefIdx. False otherwise.
+  /// with the pair \p MI, \p DefIdx and the operand has no undef flag set.
+  /// False otherwise.
   ///
   /// \pre MI.isInsertSubreg() or MI.isInsertSubregLike().
   ///

include/llvm/IR/IntrinsicsPowerPC.td

@@ -36,8 +36,12 @@ let TargetPrefix = "ppc" in {  // All intrinsics start with "llvm.ppc.".
 
   // Intrinsics used to generate ctr-based loops. These should only be
   // generated by the PowerPC backend!
+  // The branch intrinsic is marked as NoDuplicate because loop rotation will
+  // attempt to duplicate it forming loops where a block reachable from one
+  // instance of it can contain another.
   def int_ppc_mtctr : Intrinsic<[], [llvm_anyint_ty], []>;
-  def int_ppc_is_decremented_ctr_nonzero : Intrinsic<[llvm_i1_ty], [], []>;
+  def int_ppc_is_decremented_ctr_nonzero :
+    Intrinsic<[llvm_i1_ty], [], [IntrNoDuplicate]>;
 
   // Intrinsics for [double]word extended forms of divide instructions
   def int_ppc_divwe : GCCBuiltin<"__builtin_divwe">,

lib/Analysis/GlobalsModRef.cpp

@@ -502,6 +502,8 @@ void GlobalsAAResult::AnalyzeCallGraph(CallGraph &CG, Module &M) {
     }
 
     FunctionInfo &FI = FunctionInfos[F];
+    Handles.emplace_front(*this, F);
+    Handles.front().I = Handles.begin();
     bool KnowNothing = false;
 
     // Collect the mod/ref properties due to called functions.  We only compute

lib/Analysis/MemorySSA.cpp

@@ -153,9 +153,14 @@ class MemoryLocOrCall {
     if (IsCall != Other.IsCall)
       return false;
 
-    if (IsCall)
-      return CS.getCalledValue() == Other.CS.getCalledValue();
-    return Loc == Other.Loc;
+    if (!IsCall)
+      return Loc == Other.Loc;
+
+    if (CS.getCalledValue() != Other.CS.getCalledValue())
+      return false;
+
+    return CS.arg_size() == Other.CS.arg_size() &&
+           std::equal(CS.arg_begin(), CS.arg_end(), Other.CS.arg_begin());
   }
 
 private:
@@ -179,12 +184,18 @@ template <> struct DenseMapInfo<MemoryLocOrCall> {
   }
 
   static unsigned getHashValue(const MemoryLocOrCall &MLOC) {
-    if (MLOC.IsCall)
-      return hash_combine(MLOC.IsCall,
-                          DenseMapInfo<const Value *>::getHashValue(
-                              MLOC.getCS().getCalledValue()));
-    return hash_combine(
-        MLOC.IsCall, DenseMapInfo<MemoryLocation>::getHashValue(MLOC.getLoc()));
+    if (!MLOC.IsCall)
+      return hash_combine(
+          MLOC.IsCall,
+          DenseMapInfo<MemoryLocation>::getHashValue(MLOC.getLoc()));
+
+    hash_code hash =
+        hash_combine(MLOC.IsCall, DenseMapInfo<const Value *>::getHashValue(
+                                      MLOC.getCS().getCalledValue()));
+
+    for (const Value *Arg : MLOC.getCS().args())
+      hash = hash_combine(hash, DenseMapInfo<const Value *>::getHashValue(Arg));
+    return hash;
   }
 
   static bool isEqual(const MemoryLocOrCall &LHS, const MemoryLocOrCall &RHS) {

lib/CodeGen/AsmPrinter/DwarfUnit.cpp

@@ -1362,12 +1362,12 @@ void DwarfUnit::constructFortranSubrangeDIE(DIE &Buffer,
 
   if (!SR->getLowerBound()) {
     int64_t BVC = SR->getCLowerBound();
-    addSInt(DW_Subrange, dwarf::DW_AT_lower_bound, None, BVC);
+    addSInt(DW_Subrange, dwarf::DW_AT_lower_bound, dwarf::DW_FORM_sdata, BVC);
   }
 
   if ((!SR->getUpperBound()) && (!SR->noUpperBound())) {
     int64_t BVC = SR->getCUpperBound();
-    addSInt(DW_Subrange, dwarf::DW_AT_upper_bound, None, BVC);
+    addSInt(DW_Subrange, dwarf::DW_AT_upper_bound, dwarf::DW_FORM_sdata, BVC);
   }
 }
 

lib/CodeGen/IfConversion.cpp

@@ -1714,20 +1714,25 @@ bool IfConverter::IfConvertDiamondCommon(
   }
 
   // Remove the duplicated instructions at the beginnings of both paths.
-  // Skip dbg_value instructions
+  // Skip dbg_value instructions.
   MachineBasicBlock::iterator DI1 = MBB1.getFirstNonDebugInstr();
   MachineBasicBlock::iterator DI2 = MBB2.getFirstNonDebugInstr();
   BBI1->NonPredSize -= NumDups1;
   BBI2->NonPredSize -= NumDups1;
 
   // Skip past the dups on each side separately since there may be
-  // differing dbg_value entries.
+  // differing dbg_value entries. NumDups1 can include a "return"
+  // instruction, if it's not marked as "branch".
   for (unsigned i = 0; i < NumDups1; ++DI1) {
+    if (DI1 == MBB1.end())
+      break;
     if (!DI1->isDebugValue())
       ++i;
   }
   while (NumDups1 != 0) {
     ++DI2;
+    if (DI2 == MBB2.end())
+      break;
     if (!DI2->isDebugValue())
       --NumDups1;
   }
@@ -1738,11 +1743,16 @@ bool IfConverter::IfConvertDiamondCommon(
       Redefs.stepForward(MI, Dummy);
     }
   }
+
   BBI.BB->splice(BBI.BB->end(), &MBB1, MBB1.begin(), DI1);
   MBB2.erase(MBB2.begin(), DI2);
 
-  // The branches have been checked to match, so it is safe to remove the branch
-  // in BB1 and rely on the copy in BB2
+  // The branches have been checked to match, so it is safe to remove the
+  // branch in BB1 and rely on the copy in BB2. The complication is that
+  // the blocks may end with a return instruction, which may or may not
+  // be marked as "branch". If it's not, then it could be included in
+  // "dups1", leaving the blocks potentially empty after moving the common
+  // duplicates.
 #ifndef NDEBUG
   // Unanalyzable branches must match exactly. Check that now.
   if (!BBI1->IsBrAnalyzable)
@@ -1768,11 +1778,14 @@ bool IfConverter::IfConvertDiamondCommon(
   if (RemoveBranch)
     BBI2->NonPredSize -= TII->removeBranch(*BBI2->BB);
   else {
-    do {
-      assert(DI2 != MBB2.begin());
-      DI2--;
-    } while (DI2->isBranch() || DI2->isDebugValue());
-    DI2++;
+    // Make DI2 point to the end of the range where the common "tail"
+    // instructions could be found.
+    while (DI2 != MBB2.begin()) {
+      MachineBasicBlock::iterator Prev = std::prev(DI2);
+      if (!Prev->isBranch() && !Prev->isDebugValue())
+        break;
+      DI2 = Prev;
+    }
   }
   while (NumDups2 != 0) {
     // NumDups2 only counted non-dbg_value instructions, so this won't
@@ -1833,11 +1846,15 @@ bool IfConverter::IfConvertDiamondCommon(
   // a non-predicated in BBI2, then we don't want to predicate the one from
   // BBI2. The reason is that if we merged these blocks, we would end up with
   // two predicated terminators in the same block.
+  // Also, if the branches in MBB1 and MBB2 were non-analyzable, then don't
+  // predicate them either. They were checked to be identical, and so the
+  // same branch would happen regardless of which path was taken.
   if (!MBB2.empty() && (DI2 == MBB2.end())) {
     MachineBasicBlock::iterator BBI1T = MBB1.getFirstTerminator();
     MachineBasicBlock::iterator BBI2T = MBB2.getFirstTerminator();
-    if (BBI1T != MBB1.end() && TII->isPredicated(*BBI1T) &&
-        BBI2T != MBB2.end() && !TII->isPredicated(*BBI2T))
+    bool BB1Predicated = BBI1T != MBB1.end() && TII->isPredicated(*BBI1T);
+    bool BB2NonPredicated = BBI2T != MBB2.end() && !TII->isPredicated(*BBI2T);
+    if (BB2NonPredicated && (BB1Predicated || !BBI2->IsBrAnalyzable))
       --DI2;
   }
 

lib/CodeGen/LiveDebugVariables.cpp

@@ -514,6 +514,39 @@ bool LDVImpl::handleDebugValue(MachineInstr &MI, SlotIndex Idx) {
     return false;
   }
 
+  // Detect invalid DBG_VALUE instructions, with a debug-use of a virtual
+  // register that hasn't been defined yet. If we do not remove those here, then
+  // the re-insertion of the DBG_VALUE instruction after register allocation
+  // will be incorrect.
+  // TODO: If earlier passes are corrected to generate sane debug information
+  // (and if the machine verifier is improved to catch this), then these checks
+  // could be removed or replaced by asserts.
+  bool Discard = false;
+  if (MI.getOperand(0).isReg() &&
+      TargetRegisterInfo::isVirtualRegister(MI.getOperand(0).getReg())) {
+    const unsigned Reg = MI.getOperand(0).getReg();
+    if (!LIS->hasInterval(Reg)) {
+      // The DBG_VALUE is described by a virtual register that does not have a
+      // live interval. Discard the DBG_VALUE.
+      Discard = true;
+      DEBUG(dbgs() << "Discarding debug info (no LIS interval): "
+            << Idx << " " << MI);
+    } else {
+      // The DBG_VALUE is only valid if either Reg is live out from Idx, or Reg
+      // is defined dead at Idx (where Idx is the slot index for the instruction
+      // preceeding the DBG_VALUE).
+      const LiveInterval &LI = LIS->getInterval(Reg);
+      LiveQueryResult LRQ = LI.Query(Idx);
+      if (!LRQ.valueOutOrDead()) {
+        // We have found a DBG_VALUE with the value in a virtual register that
+        // is not live. Discard the DBG_VALUE.
+        Discard = true;
+        DEBUG(dbgs() << "Discarding debug info (reg not live): "
+              << Idx << " " << MI);
+      }
+    }
+  }
+
   // Get or create the UserValue for (variable,offset) here.
   bool IsIndirect = MI.getOperand(1).isImm();
   if (IsIndirect)
@@ -522,7 +555,13 @@ bool LDVImpl::handleDebugValue(MachineInstr &MI, SlotIndex Idx) {
   const DIExpression *Expr = MI.getDebugExpression();
   UserValue *UV =
       getUserValue(Var, Expr, MI.getDebugLoc());
-  UV->addDef(Idx, MI.getOperand(0), IsIndirect);
+  if (!Discard)
+    UV->addDef(Idx, MI.getOperand(0), IsIndirect);
+  else {
+    MachineOperand MO = MachineOperand::CreateReg(0U, false);
+    MO.setIsDebug();
+    UV->addDef(Idx, MO, false);
+  }
   return true;
 }
 

lib/CodeGen/MachineBasicBlock.cpp

@@ -646,6 +646,14 @@ void MachineBasicBlock::replaceSuccessor(MachineBasicBlock *Old,
   removeSuccessor(OldI);
 }
 
+void MachineBasicBlock::copySuccessor(MachineBasicBlock *Orig,
+                                      succ_iterator I) {
+  if (Orig->Probs.empty())
+    addSuccessor(*I, Orig->getSuccProbability(I));
+  else
+    addSuccessorWithoutProb(*I);
+}
+
 void MachineBasicBlock::addPredecessor(MachineBasicBlock *Pred) {
   Predecessors.push_back(Pred);
 }

lib/CodeGen/MachineBlockPlacement.cpp

@@ -513,6 +513,11 @@ class MachineBlockPlacement : public MachineFunctionPass {
 
   bool runOnMachineFunction(MachineFunction &F) override;
 
+  bool allowTailDupPlacement() const {
+    assert(F);
+    return TailDupPlacement && !F->getTarget().requiresStructuredCFG();
+  }
+
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<MachineBranchProbabilityInfo>();
     AU.addRequired<MachineBlockFrequencyInfo>();
@@ -1018,7 +1023,7 @@ MachineBlockPlacement::getBestTrellisSuccessor(
     MachineBasicBlock *Succ1 = BestA.Dest;
     MachineBasicBlock *Succ2 = BestB.Dest;
     // Check to see if tail-duplication would be profitable.
-    if (TailDupPlacement && shouldTailDuplicate(Succ2) &&
+    if (allowTailDupPlacement() && shouldTailDuplicate(Succ2) &&
         canTailDuplicateUnplacedPreds(BB, Succ2, Chain, BlockFilter) &&
         isProfitableToTailDup(BB, Succ2, MBPI->getEdgeProbability(BB, Succ1),
                               Chain, BlockFilter)) {
@@ -1044,7 +1049,7 @@ MachineBlockPlacement::getBestTrellisSuccessor(
   return Result;
 }
 
-/// When the option TailDupPlacement is on, this method checks if the
+/// When the option allowTailDupPlacement() is on, this method checks if the
 /// fallthrough candidate block \p Succ (of block \p BB) can be tail-duplicated
 /// into all of its unplaced, unfiltered predecessors, that are not BB.
 bool MachineBlockPlacement::canTailDuplicateUnplacedPreds(
@@ -1493,7 +1498,7 @@ MachineBlockPlacement::selectBestSuccessor(
     if (hasBetterLayoutPredecessor(BB, Succ, SuccChain, SuccProb, RealSuccProb,
                                    Chain, BlockFilter)) {
       // If tail duplication would make Succ profitable, place it.
-      if (TailDupPlacement && shouldTailDuplicate(Succ))
+      if (allowTailDupPlacement() && shouldTailDuplicate(Succ))
         DupCandidates.push_back(std::make_tuple(SuccProb, Succ));
       continue;
     }
@@ -1702,7 +1707,7 @@ void MachineBlockPlacement::buildChain(
     auto Result = selectBestSuccessor(BB, Chain, BlockFilter);
     MachineBasicBlock* BestSucc = Result.BB;
     bool ShouldTailDup = Result.ShouldTailDup;
-    if (TailDupPlacement)
+    if (allowTailDupPlacement())
       ShouldTailDup |= (BestSucc && shouldTailDuplicate(BestSucc));
 
     // If an immediate successor isn't available, look for the best viable
@@ -1724,7 +1729,7 @@ void MachineBlockPlacement::buildChain(
 
     // Placement may have changed tail duplication opportunities.
     // Check for that now.
-    if (TailDupPlacement && BestSucc && ShouldTailDup) {
+    if (allowTailDupPlacement() && BestSucc && ShouldTailDup) {
       // If the chosen successor was duplicated into all its predecessors,
       // don't bother laying it out, just go round the loop again with BB as
       // the chain end.
@@ -2758,7 +2763,7 @@ bool MachineBlockPlacement::runOnMachineFunction(MachineFunction &MF) {
       TailDupSize = TailDupPlacementAggressiveThreshold;
   }
 
-  if (TailDupPlacement) {
+  if (allowTailDupPlacement()) {
     MPDT = &getAnalysis<MachinePostDominatorTree>();
     if (MF.getFunction().optForSize())
       TailDupSize = 1;