1 files changed, 71 insertions, 7 deletions

diff --git a/llvm/lib/Transforms/Utils/LoopUtils.cpp b/llvm/lib/Transforms/Utils/LoopUtils.cpp
index 7df8651ede15..7d6662c44f07 100644
--- a/llvm/lib/Transforms/Utils/LoopUtils.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUtils.cpp
@@ -466,6 +466,19 @@ llvm::collectChildrenInLoop(DomTreeNode *N, const Loop *CurLoop) {
   return Worklist;
 }
 
+bool llvm::isAlmostDeadIV(PHINode *PN, BasicBlock *LatchBlock, Value *Cond) {
+  int LatchIdx = PN->getBasicBlockIndex(LatchBlock);
+  Value *IncV = PN->getIncomingValue(LatchIdx);
+
+  for (User *U : PN->users())
+    if (U != Cond && U != IncV) return false;
+
+  for (User *U : IncV->users())
+    if (U != Cond && U != PN) return false;
+  return true;
+}
+
+
 void llvm::deleteDeadLoop(Loop *L, DominatorTree *DT, ScalarEvolution *SE,
                           LoopInfo *LI, MemorySSA *MSSA) {
   assert((!DT || L->isLCSSAForm(*DT)) && "Expected LCSSA!");
@@ -628,18 +641,17 @@ void llvm::deleteDeadLoop(Loop *L, DominatorTree *DT, ScalarEvolution *SE,
       }
 
     // After the loop has been deleted all the values defined and modified
-    // inside the loop are going to be unavailable.
-    // Since debug values in the loop have been deleted, inserting an undef
-    // dbg.value truncates the range of any dbg.value before the loop where the
-    // loop used to be. This is particularly important for constant values.
+    // inside the loop are going to be unavailable. Values computed in the
+    // loop will have been deleted, automatically causing their debug uses
+    // be be replaced with undef. Loop invariant values will still be available.
+    // Move dbg.values out the loop so that earlier location ranges are still
+    // terminated and loop invariant assignments are preserved.
     Instruction *InsertDbgValueBefore = ExitBlock->getFirstNonPHI();
     assert(InsertDbgValueBefore &&
            "There should be a non-PHI instruction in exit block, else these "
            "instructions will have no parent.");
-    for (auto *DVI : DeadDebugInst) {
-      DVI->setKillLocation();
+    for (auto *DVI : DeadDebugInst)
       DVI->moveBefore(InsertDbgValueBefore);
-    }
   }
 
   // Remove the block from the reference counting scheme, so that we can
@@ -880,6 +892,29 @@ bool llvm::hasIterationCountInvariantInParent(Loop *InnerLoop,
   return true;
 }
 
+Intrinsic::ID llvm::getMinMaxReductionIntrinsicOp(RecurKind RK) {
+  switch (RK) {
+  default:
+    llvm_unreachable("Unknown min/max recurrence kind");
+  case RecurKind::UMin:
+    return Intrinsic::umin;
+  case RecurKind::UMax:
+    return Intrinsic::umax;
+  case RecurKind::SMin:
+    return Intrinsic::smin;
+  case RecurKind::SMax:
+    return Intrinsic::smax;
+  case RecurKind::FMin:
+    return Intrinsic::minnum;
+  case RecurKind::FMax:
+    return Intrinsic::maxnum;
+  case RecurKind::FMinimum:
+    return Intrinsic::minimum;
+  case RecurKind::FMaximum:
+    return Intrinsic::maximum;
+  }
+}
+
 CmpInst::Predicate llvm::getMinMaxReductionPredicate(RecurKind RK) {
   switch (RK) {
   default:
@@ -896,6 +931,9 @@ CmpInst::Predicate llvm::getMinMaxReductionPredicate(RecurKind RK) {
     return CmpInst::FCMP_OLT;
   case RecurKind::FMax:
     return CmpInst::FCMP_OGT;
+  // We do not add FMinimum/FMaximum recurrence kind here since there is no
+  // equivalent predicate which compares signed zeroes according to the
+  // semantics of the intrinsics (llvm.minimum/maximum).
   }
 }
 
@@ -910,6 +948,14 @@ Value *llvm::createSelectCmpOp(IRBuilderBase &Builder, Value *StartVal,
 
 Value *llvm::createMinMaxOp(IRBuilderBase &Builder, RecurKind RK, Value *Left,
                             Value *Right) {
+  Type *Ty = Left->getType();
+  if (Ty->isIntOrIntVectorTy() ||
+      (RK == RecurKind::FMinimum || RK == RecurKind::FMaximum)) {
+    // TODO: Add float minnum/maxnum support when FMF nnan is set.
+    Intrinsic::ID Id = getMinMaxReductionIntrinsicOp(RK);
+    return Builder.CreateIntrinsic(Ty, Id, {Left, Right}, nullptr,
+                                   "rdx.minmax");
+  }
   CmpInst::Predicate Pred = getMinMaxReductionPredicate(RK);
   Value *Cmp = Builder.CreateCmp(Pred, Left, Right, "rdx.minmax.cmp");
   Value *Select = Builder.CreateSelect(Cmp, Left, Right, "rdx.minmax.select");
@@ -1055,6 +1101,10 @@ Value *llvm::createSimpleTargetReduction(IRBuilderBase &Builder,
     return Builder.CreateFPMaxReduce(Src);
   case RecurKind::FMin:
     return Builder.CreateFPMinReduce(Src);
+  case RecurKind::FMinimum:
+    return Builder.CreateFPMinimumReduce(Src);
+  case RecurKind::FMaximum:
+    return Builder.CreateFPMaximumReduce(Src);
   default:
     llvm_unreachable("Unhandled opcode");
   }
@@ -1123,6 +1173,20 @@ bool llvm::isKnownNonNegativeInLoop(const SCEV *S, const Loop *L,
          SE.isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SGE, S, Zero);
 }
 
+bool llvm::isKnownPositiveInLoop(const SCEV *S, const Loop *L,
+                                 ScalarEvolution &SE) {
+  const SCEV *Zero = SE.getZero(S->getType());
+  return SE.isAvailableAtLoopEntry(S, L) &&
+         SE.isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SGT, S, Zero);
+}
+
+bool llvm::isKnownNonPositiveInLoop(const SCEV *S, const Loop *L,
+                                    ScalarEvolution &SE) {
+  const SCEV *Zero = SE.getZero(S->getType());
+  return SE.isAvailableAtLoopEntry(S, L) &&
+         SE.isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SLE, S, Zero);
+}
+
 bool llvm::cannotBeMinInLoop(const SCEV *S, const Loop *L, ScalarEvolution &SE,
                              bool Signed) {
   unsigned BitWidth = cast<IntegerType>(S->getType())->getBitWidth();