Split the Add, Sub, and Mul instruction opcodes into separate

integer and floating-point opcodes, introducing
FAdd, FSub, and FMul.

For now, the AsmParser, BitcodeReader, and IRBuilder all preserve
backwards compatability, and the Core LLVM APIs preserve backwards
compatibility for IR producers. Most front-ends won't need to change
immediately.

This implements the first step of the plan outlined here:
http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72897 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 182 insertions, 64 deletions

diff --git a/lib/Transforms/Scalar/GVN.cpp b/lib/Transforms/Scalar/GVN.cpp
index 733dfa9..673d38b 100644
--- a/lib/Transforms/Scalar/GVN.cpp
+++ b/lib/Transforms/Scalar/GVN.cpp
@@ -59,7 +59,8 @@ cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));
 /// two values.
 namespace {
   struct VISIBILITY_HIDDEN Expression {
-    enum ExpressionOpcode { ADD, SUB, MUL, UDIV, SDIV, FDIV, UREM, SREM, 
+    enum ExpressionOpcode { ADD, FADD, SUB, FSUB, MUL, FMUL,
+                            UDIV, SDIV, FDIV, UREM, SREM,
                             FREM, SHL, LSHR, ASHR, AND, OR, XOR, ICMPEQ, 
                             ICMPNE, ICMPUGT, ICMPUGE, ICMPULT, ICMPULE, 
                             ICMPSGT, ICMPSGE, ICMPSLT, ICMPSLE, FCMPOEQ, 
@@ -200,8 +201,11 @@ Expression::ExpressionOpcode ValueTable::getOpcode(BinaryOperator* BO) {
   default: // THIS SHOULD NEVER HAPPEN
     assert(0 && "Binary operator with unknown opcode?");
   case Instruction::Add:  return Expression::ADD;
+  case Instruction::FAdd: return Expression::FADD;
   case Instruction::Sub:  return Expression::SUB;
+  case Instruction::FSub: return Expression::FSUB;
   case Instruction::Mul:  return Expression::MUL;
+  case Instruction::FMul: return Expression::FMUL;
   case Instruction::UDiv: return Expression::UDIV;
   case Instruction::SDiv: return Expression::SDIV;
   case Instruction::FDiv: return Expression::FDIV;
diff --git a/lib/Transforms/Scalar/GVNPRE.cpp b/lib/Transforms/Scalar/GVNPRE.cpp
index e3b0937..0f3153f 100644
--- a/lib/Transforms/Scalar/GVNPRE.cpp
+++ b/lib/Transforms/Scalar/GVNPRE.cpp
@@ -55,7 +55,8 @@ namespace {
 /// two values.
 
 struct Expression {
-  enum ExpressionOpcode { ADD, SUB, MUL, UDIV, SDIV, FDIV, UREM, SREM, 
+  enum ExpressionOpcode { ADD, FADD, SUB, FSUB, MUL, FMUL,
+                          UDIV, SDIV, FDIV, UREM, SREM,
                           FREM, SHL, LSHR, ASHR, AND, OR, XOR, ICMPEQ, 
                           ICMPNE, ICMPUGT, ICMPUGE, ICMPULT, ICMPULE, 
                           ICMPSGT, ICMPSGE, ICMPSLT, ICMPSLE, FCMPOEQ, 
@@ -202,10 +203,16 @@ Expression::ExpressionOpcode
   switch(BO->getOpcode()) {
     case Instruction::Add:
       return Expression::ADD;
+    case Instruction::FAdd:
+      return Expression::FADD;
     case Instruction::Sub:
       return Expression::SUB;
+    case Instruction::FSub:
+      return Expression::FSUB;
     case Instruction::Mul:
       return Expression::MUL;
+    case Instruction::FMul:
+      return Expression::FMUL;
     case Instruction::UDiv:
       return Expression::UDIV;
     case Instruction::SDiv:
diff --git a/lib/Transforms/Scalar/IndVarSimplify.cpp b/lib/Transforms/Scalar/IndVarSimplify.cpp
index af61eae..83503fd 100644
--- a/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -754,7 +754,7 @@ void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PH) {
   BinaryOperator *Incr =
     dyn_cast<BinaryOperator>(PH->getIncomingValue(BackEdge));
   if (!Incr) return;
-  if (Incr->getOpcode() != Instruction::Add) return;
+  if (Incr->getOpcode() != Instruction::FAdd) return;
   ConstantFP *IncrValue = NULL;
   unsigned IncrVIndex = 1;
   if (Incr->getOperand(1) == PH)
diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp
index 8bb16f3..97bd34c 100644
--- a/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -167,8 +167,11 @@ namespace {
     //   otherwise    - Change was made, replace I with returned instruction
     //
     Instruction *visitAdd(BinaryOperator &I);
+    Instruction *visitFAdd(BinaryOperator &I);
     Instruction *visitSub(BinaryOperator &I);
+    Instruction *visitFSub(BinaryOperator &I);
     Instruction *visitMul(BinaryOperator &I);
+    Instruction *visitFMul(BinaryOperator &I);
     Instruction *visitURem(BinaryOperator &I);
     Instruction *visitSRem(BinaryOperator &I);
     Instruction *visitFRem(BinaryOperator &I);
@@ -403,7 +406,8 @@ X("instcombine", "Combine redundant instructions");
 //   0 -> undef, 1 -> Const, 2 -> Other, 3 -> Arg, 3 -> Unary, 4 -> OtherInst
 static unsigned getComplexity(Value *V) {
   if (isa<Instruction>(V)) {
-    if (BinaryOperator::isNeg(V) || BinaryOperator::isNot(V))
+    if (BinaryOperator::isNeg(V) || BinaryOperator::isFNeg(V) ||
+        BinaryOperator::isNot(V))
       return 3;
     return 4;
   }
@@ -576,6 +580,25 @@ static inline Value *dyn_castNegVal(Value *V) {
   return 0;
 }
 
+// dyn_castFNegVal - Given a 'fsub' instruction, return the RHS of the
+// instruction if the LHS is a constant negative zero (which is the 'negate'
+// form).
+//
+static inline Value *dyn_castFNegVal(Value *V) {
+  if (BinaryOperator::isFNeg(V))
+    return BinaryOperator::getFNegArgument(V);
+
+  // Constants can be considered to be negated values if they can be folded.
+  if (ConstantFP *C = dyn_cast<ConstantFP>(V))
+    return ConstantExpr::getFNeg(C);
+
+  if (ConstantVector *C = dyn_cast<ConstantVector>(V))
+    if (C->getType()->getElementType()->isFloatingPoint())
+      return ConstantExpr::getFNeg(C);
+
+  return 0;
+}
+
 static inline Value *dyn_castNotVal(Value *V) {
   if (BinaryOperator::isNot(V))
     return BinaryOperator::getNotArgument(V);
@@ -1733,12 +1756,12 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
           default: assert(0 && "Case stmts out of sync!");
           case Intrinsic::x86_sse_sub_ss:
           case Intrinsic::x86_sse2_sub_sd:
-            TmpV = InsertNewInstBefore(BinaryOperator::CreateSub(LHS, RHS,
+            TmpV = InsertNewInstBefore(BinaryOperator::CreateFSub(LHS, RHS,
                                                         II->getName()), *II);
             break;
           case Intrinsic::x86_sse_mul_ss:
           case Intrinsic::x86_sse2_mul_sd:
-            TmpV = InsertNewInstBefore(BinaryOperator::CreateMul(LHS, RHS,
+            TmpV = InsertNewInstBefore(BinaryOperator::CreateFMul(LHS, RHS,
                                                          II->getName()), *II);
             break;
           }
@@ -2052,14 +2075,8 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
       return ReplaceInstUsesWith(I, RHS);
 
     // X + 0 --> X
-    if (!I.getType()->isFPOrFPVector()) { // NOTE: -0 + +0 = +0.
-      if (RHSC->isNullValue())
-        return ReplaceInstUsesWith(I, LHS);
-    } else if (ConstantFP *CFP = dyn_cast<ConstantFP>(RHSC)) {
-      if (CFP->isExactlyValue(ConstantFP::getNegativeZero
-                              (I.getType())->getValueAPF()))
-        return ReplaceInstUsesWith(I, LHS);
-    }
+    if (RHSC->isNullValue())
+      return ReplaceInstUsesWith(I, LHS);
 
     if (ConstantInt *CI = dyn_cast<ConstantInt>(RHSC)) {
       // X + (signbit) --> X ^ signbit
@@ -2317,11 +2334,6 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
         return SelectInst::Create(SI->getCondition(), A, N);
     }
   }
-  
-  // Check for X+0.0.  Simplify it to X if we know X is not -0.0.
-  if (ConstantFP *CFP = dyn_cast<ConstantFP>(RHS))
-    if (CFP->getValueAPF().isPosZero() && CannotBeNegativeZero(LHS))
-      return ReplaceInstUsesWith(I, LHS);
 
   // Check for (add (sext x), y), see if we can merge this into an
   // integer add followed by a sext.
@@ -2359,7 +2371,42 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
       }
     }
   }
-  
+
+  return Changed ? &I : 0;
+}
+
+Instruction *InstCombiner::visitFAdd(BinaryOperator &I) {
+  bool Changed = SimplifyCommutative(I);
+  Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
+
+  if (Constant *RHSC = dyn_cast<Constant>(RHS)) {
+    // X + 0 --> X
+    if (ConstantFP *CFP = dyn_cast<ConstantFP>(RHSC)) {
+      if (CFP->isExactlyValue(ConstantFP::getNegativeZero
+                              (I.getType())->getValueAPF()))
+        return ReplaceInstUsesWith(I, LHS);
+    }
+
+    if (isa<PHINode>(LHS))
+      if (Instruction *NV = FoldOpIntoPhi(I))
+        return NV;
+  }
+
+  // -A + B  -->  B - A
+  // -A + -B  -->  -(A + B)
+  if (Value *LHSV = dyn_castFNegVal(LHS))
+    return BinaryOperator::CreateFSub(RHS, LHSV);
+
+  // A + -B  -->  A - B
+  if (!isa<Constant>(RHS))
+    if (Value *V = dyn_castFNegVal(RHS))
+      return BinaryOperator::CreateFSub(LHS, V);
+
+  // Check for X+0.0.  Simplify it to X if we know X is not -0.0.
+  if (ConstantFP *CFP = dyn_cast<ConstantFP>(RHS))
+    if (CFP->getValueAPF().isPosZero() && CannotBeNegativeZero(LHS))
+      return ReplaceInstUsesWith(I, LHS);
+
   // Check for (add double (sitofp x), y), see if we can merge this into an
   // integer add followed by a promotion.
   if (SIToFPInst *LHSConv = dyn_cast<SIToFPInst>(LHS)) {
@@ -2407,8 +2454,7 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
 Instruction *InstCombiner::visitSub(BinaryOperator &I) {
   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
 
-  if (Op0 == Op1 &&                        // sub X, X  -> 0
-      !I.getType()->isFPOrFPVector())
+  if (Op0 == Op1)                        // sub X, X  -> 0
     return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
 
   // If this is a 'B = x-(-A)', change to B = x+A...
@@ -2469,8 +2515,7 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
     return BinaryOperator::CreateXor(Op0, Op1);
 
   if (BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1)) {
-    if (Op1I->getOpcode() == Instruction::Add &&
-        !Op0->getType()->isFPOrFPVector()) {
+    if (Op1I->getOpcode() == Instruction::Add) {
       if (Op1I->getOperand(0) == Op0)              // X-(X+Y) == -Y
         return BinaryOperator::CreateNeg(Op1I->getOperand(1), I.getName());
       else if (Op1I->getOperand(1) == Op0)         // X-(Y+X) == -Y
@@ -2487,8 +2532,7 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
       // Replace (x - (y - z)) with (x + (z - y)) if the (y - z) subexpression
       // is not used by anyone else...
       //
-      if (Op1I->getOpcode() == Instruction::Sub &&
-          !Op1I->getType()->isFPOrFPVector()) {
+      if (Op1I->getOpcode() == Instruction::Sub) {
         // Swap the two operands of the subexpr...
         Value *IIOp0 = Op1I->getOperand(0), *IIOp1 = Op1I->getOperand(1);
         Op1I->setOperand(0, IIOp1);
@@ -2526,18 +2570,17 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
     }
   }
 
-  if (!Op0->getType()->isFPOrFPVector())
-    if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
-      if (Op0I->getOpcode() == Instruction::Add) {
-        if (Op0I->getOperand(0) == Op1)             // (Y+X)-Y == X
-          return ReplaceInstUsesWith(I, Op0I->getOperand(1));
-        else if (Op0I->getOperand(1) == Op1)        // (X+Y)-Y == X
-          return ReplaceInstUsesWith(I, Op0I->getOperand(0));
-      } else if (Op0I->getOpcode() == Instruction::Sub) {
-        if (Op0I->getOperand(0) == Op1)             // (X-Y)-X == -Y
-          return BinaryOperator::CreateNeg(Op0I->getOperand(1), I.getName());
-      }
+  if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
+    if (Op0I->getOpcode() == Instruction::Add) {
+      if (Op0I->getOperand(0) == Op1)             // (Y+X)-Y == X
+        return ReplaceInstUsesWith(I, Op0I->getOperand(1));
+      else if (Op0I->getOperand(1) == Op1)        // (X+Y)-Y == X
+        return ReplaceInstUsesWith(I, Op0I->getOperand(0));
+    } else if (Op0I->getOpcode() == Instruction::Sub) {
+      if (Op0I->getOperand(0) == Op1)             // (X-Y)-X == -Y
+        return BinaryOperator::CreateNeg(Op0I->getOperand(1), I.getName());
     }
+  }
 
   ConstantInt *C1;
   if (Value *X = dyn_castFoldableMul(Op0, C1)) {
@@ -2551,6 +2594,40 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
   return 0;
 }
 
+Instruction *InstCombiner::visitFSub(BinaryOperator &I) {
+  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
+
+  // If this is a 'B = x-(-A)', change to B = x+A...
+  if (Value *V = dyn_castFNegVal(Op1))
+    return BinaryOperator::CreateFAdd(Op0, V);
+
+  if (BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1)) {
+    if (Op1I->getOpcode() == Instruction::FAdd) {
+      if (Op1I->getOperand(0) == Op0)              // X-(X+Y) == -Y
+        return BinaryOperator::CreateFNeg(Op1I->getOperand(1), I.getName());
+      else if (Op1I->getOperand(1) == Op0)         // X-(Y+X) == -Y
+        return BinaryOperator::CreateFNeg(Op1I->getOperand(0), I.getName());
+    }
+
+    if (Op1I->hasOneUse()) {
+      // Replace (x - (y - z)) with (x + (z - y)) if the (y - z) subexpression
+      // is not used by anyone else...
+      //
+      if (Op1I->getOpcode() == Instruction::FSub) {
+        // Swap the two operands of the subexpr...
+        Value *IIOp0 = Op1I->getOperand(0), *IIOp1 = Op1I->getOperand(1);
+        Op1I->setOperand(0, IIOp1);
+        Op1I->setOperand(1, IIOp0);
+
+        // Create the new top level fadd instruction...
+        return BinaryOperator::CreateFAdd(Op0, Op1);
+      }
+    }
+  }
+
+  return 0;
+}
+
 /// isSignBitCheck - Given an exploded icmp instruction, return true if the
 /// comparison only checks the sign bit.  If it only checks the sign bit, set
 /// TrueIfSigned if the result of the comparison is true when the input value is
@@ -2613,13 +2690,6 @@ Instruction *InstCombiner::visitMul(BinaryOperator &I) {
         return BinaryOperator::CreateShl(Op0,
                  ConstantInt::get(Op0->getType(), Val.logBase2()));
       }
-    } else if (ConstantFP *Op1F = dyn_cast<ConstantFP>(Op1)) {
-      // TODO: If Op1 is zero and Op0 is finite, return zero.
-
-      // "In IEEE floating point, x*1 is not equivalent to x for nans.  However,
-      // ANSI says we can drop signals, so we can do this anyway." (from GCC)
-      if (Op1F->isExactlyValue(1.0))
-        return ReplaceInstUsesWith(I, Op0);  // Eliminate 'mul double %X, 1.0'
     } else if (isa<VectorType>(Op1->getType())) {
       // TODO: If Op1 is all zeros and Op0 is all finite, return all zeros.
 
@@ -2629,9 +2699,6 @@ Instruction *InstCombiner::visitMul(BinaryOperator &I) {
 
         // As above, vector X*splat(1.0) -> X in all defined cases.
         if (Constant *Splat = Op1V->getSplatValue()) {
-          if (ConstantFP *F = dyn_cast<ConstantFP>(Splat))
-            if (F->isExactlyValue(1.0))
-              return ReplaceInstUsesWith(I, Op0);
           if (ConstantInt *CI = dyn_cast<ConstantInt>(Splat))
             if (CI->equalsInt(1))
               return ReplaceInstUsesWith(I, Op0);
@@ -2755,6 +2822,45 @@ Instruction *InstCombiner::visitMul(BinaryOperator &I) {
   return Changed ? &I : 0;
 }
 
+Instruction *InstCombiner::visitFMul(BinaryOperator &I) {
+  bool Changed = SimplifyCommutative(I);
+  Value *Op0 = I.getOperand(0);
+
+  // Simplify mul instructions with a constant RHS...
+  if (Constant *Op1 = dyn_cast<Constant>(I.getOperand(1))) {
+    if (ConstantFP *Op1F = dyn_cast<ConstantFP>(Op1)) {
+      // "In IEEE floating point, x*1 is not equivalent to x for nans.  However,
+      // ANSI says we can drop signals, so we can do this anyway." (from GCC)
+      if (Op1F->isExactlyValue(1.0))
+        return ReplaceInstUsesWith(I, Op0);  // Eliminate 'mul double %X, 1.0'
+    } else if (isa<VectorType>(Op1->getType())) {
+      if (ConstantVector *Op1V = dyn_cast<ConstantVector>(Op1)) {
+        // As above, vector X*splat(1.0) -> X in all defined cases.
+        if (Constant *Splat = Op1V->getSplatValue()) {
+          if (ConstantFP *F = dyn_cast<ConstantFP>(Splat))
+            if (F->isExactlyValue(1.0))
+              return ReplaceInstUsesWith(I, Op0);
+        }
+      }
+    }
+
+    // Try to fold constant mul into select arguments.
+    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
+      if (Instruction *R = FoldOpIntoSelect(I, SI, this))
+        return R;
+
+    if (isa<PHINode>(Op0))
+      if (Instruction *NV = FoldOpIntoPhi(I))
+        return NV;
+  }
+
+  if (Value *Op0v = dyn_castFNegVal(Op0))     // -X * -Y = X*Y
+    if (Value *Op1v = dyn_castFNegVal(I.getOperand(1)))
+      return BinaryOperator::CreateFMul(Op0v, Op1v);
+
+  return Changed ? &I : 0;
+}
+
 /// SimplifyDivRemOfSelect - Try to fold a divide or remainder of a select
 /// instruction.
 bool InstCombiner::SimplifyDivRemOfSelect(BinaryOperator &I) {
@@ -8562,17 +8668,17 @@ Instruction *InstCombiner::visitFPTrunc(FPTruncInst &CI) {
   if (Instruction *I = commonCastTransforms(CI))
     return I;
   
-  // If we have fptrunc(add (fpextend x), (fpextend y)), where x and y are
+  // If we have fptrunc(fadd (fpextend x), (fpextend y)), where x and y are
   // smaller than the destination type, we can eliminate the truncate by doing
-  // the add as the smaller type.  This applies to add/sub/mul/div as well as
+  // the add as the smaller type.  This applies to fadd/fsub/fmul/fdiv as well as
   // many builtins (sqrt, etc).
   BinaryOperator *OpI = dyn_cast<BinaryOperator>(CI.getOperand(0));
   if (OpI && OpI->hasOneUse()) {
     switch (OpI->getOpcode()) {
     default: break;
-    case Instruction::Add:
-    case Instruction::Sub:
-    case Instruction::Mul:
+    case Instruction::FAdd:
+    case Instruction::FSub:
+    case Instruction::FMul:
     case Instruction::FDiv:
     case Instruction::FRem:
       const Type *SrcTy = OpI->getType();
@@ -9322,11 +9428,15 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
 
         // Turn select C, (X+Y), (X-Y) --> (X+(select C, Y, (-Y))).  This is
         // even legal for FP.
-        if (TI->getOpcode() == Instruction::Sub &&
-            FI->getOpcode() == Instruction::Add) {
+        if ((TI->getOpcode() == Instruction::Sub &&
+             FI->getOpcode() == Instruction::Add) ||
+            (TI->getOpcode() == Instruction::FSub &&
+             FI->getOpcode() == Instruction::FAdd)) {
           AddOp = FI; SubOp = TI;
-        } else if (FI->getOpcode() == Instruction::Sub &&
-                   TI->getOpcode() == Instruction::Add) {
+        } else if ((FI->getOpcode() == Instruction::Sub &&
+                    TI->getOpcode() == Instruction::Add) ||
+                   (FI->getOpcode() == Instruction::FSub &&
+                    TI->getOpcode() == Instruction::FAdd)) {
           AddOp = TI; SubOp = FI;
         }
 
diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index 92270b5..f05afec 100644
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -2268,7 +2268,8 @@ void LoopStrengthReduce::OptimizeShadowIV(Loop *L) {
       /* create new increment. '++d' in above example. */
       ConstantFP *CFP = ConstantFP::get(DestTy, C->getZExtValue());
       BinaryOperator *NewIncr = 
-        BinaryOperator::Create(Incr->getOpcode(),
+        BinaryOperator::Create(Incr->getOpcode() == Instruction::Add ?
+                                 Instruction::FAdd : Instruction::FSub,
                                NewPH, CFP, "IV.S.next.", Incr);
 
       NewPH->addIncoming(NewInit, PH->getIncomingBlock(Entry));
diff --git a/lib/Transforms/Scalar/SimplifyLibCalls.cpp b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
index 4b00640..59989c9 100644
--- a/lib/Transforms/Scalar/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
@@ -1009,7 +1009,7 @@ struct VISIBILITY_HIDDEN PowOpt : public LibCallOptimization {
     if (Op2C->isExactlyValue(1.0))  // pow(x, 1.0) -> x
       return Op1;
     if (Op2C->isExactlyValue(2.0))  // pow(x, 2.0) -> x*x
-      return B.CreateMul(Op1, Op1, "pow2");
+      return B.CreateFMul(Op1, Op1, "pow2");
     if (Op2C->isExactlyValue(-1.0)) // pow(x, -1.0) -> 1.0/x
       return B.CreateFDiv(ConstantFP::get(CI->getType(), 1.0), Op1, "powrecip");
     return 0;
diff --git a/lib/Transforms/Utils/SimplifyCFG.cpp b/lib/Transforms/Utils/SimplifyCFG.cpp
index 2cde765..bcc6b81 100644
--- a/lib/Transforms/Utils/SimplifyCFG.cpp
+++ b/lib/Transforms/Utils/SimplifyCFG.cpp
@@ -419,9 +419,6 @@ static bool DominatesMergePoint(Value *V, BasicBlock *BB,
       case Instruction::LShr:
       case Instruction::AShr:
       case Instruction::ICmp:
-      case Instruction::FCmp:
-        if (I->getOperand(0)->getType()->isFPOrFPVector())
-          return false;  // FP arithmetic might trap.
         break;   // These are all cheap and non-trapping instructions.
       }
 
@@ -1012,9 +1009,8 @@ static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *BB1) {
   default: return false;  // Not safe / profitable to hoist.
   case Instruction::Add:
   case Instruction::Sub:
-    // FP arithmetic might trap. Not worth doing for vector ops.
-    if (HInst->getType()->isFloatingPoint() 
-        || isa<VectorType>(HInst->getType()))
+    // Not worth doing for vector ops.
+    if (isa<VectorType>(HInst->getType()))
       return false;
     break;
   case Instruction::And: