Remember to actually update SplitAnalysis statistics now that we have a fancy

[oota-llvm.git] / lib / Transforms / InstCombine / InstCombineCasts.cpp

diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp
index b0137c4a32ca128b0a65ed8a0dc3ea63cdc437ba..dbe5200d0009fa1371d33897089e5a5c2a252c31 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -22,19 +22,18 @@ using namespace PatternMatch;
 /// X*Scale+Offset.
 ///
 static Value *DecomposeSimpleLinearExpr(Value *Val, unsigned &Scale,
-                                        int &Offset) {
-  assert(Val->getType()->isIntegerTy(32) && "Unexpected allocation size type!");
+                                        uint64_t &Offset) {
   if (ConstantInt *CI = dyn_cast<ConstantInt>(Val)) {
     Offset = CI->getZExtValue();
     Scale  = 0;
-    return ConstantInt::get(Type::getInt32Ty(Val->getContext()), 0);
+    return ConstantInt::get(Val->getType(), 0);
   }
   
   if (BinaryOperator *I = dyn_cast<BinaryOperator>(Val)) {
     if (ConstantInt *RHS = dyn_cast<ConstantInt>(I->getOperand(1))) {
       if (I->getOpcode() == Instruction::Shl) {
         // This is a value scaled by '1 << the shift amt'.
-        Scale = 1U << RHS->getZExtValue();
+        Scale = UINT64_C(1) << RHS->getZExtValue();
         Offset = 0;
         return I->getOperand(0);
       }
@@ -100,7 +99,7 @@ Instruction *InstCombiner::PromoteCastOfAllocation(BitCastInst &CI,
   // See if we can satisfy the modulus by pulling a scale out of the array
   // size argument.
   unsigned ArraySizeScale;
-  int ArrayOffset;
+  uint64_t ArrayOffset;
   Value *NumElements = // See if the array size is a decomposable linear expr.
     DecomposeSimpleLinearExpr(AI.getOperand(0), ArraySizeScale, ArrayOffset);
  
@@ -114,13 +113,13 @@ Instruction *InstCombiner::PromoteCastOfAllocation(BitCastInst &CI,
   if (Scale == 1) {
     Amt = NumElements;
   } else {
-    Amt = ConstantInt::get(Type::getInt32Ty(CI.getContext()), Scale);
+    Amt = ConstantInt::get(AI.getArraySize()->getType(), Scale);
     // Insert before the alloca, not before the cast.
     Amt = AllocaBuilder.CreateMul(Amt, NumElements, "tmp");
   }
   
-  if (int Offset = (AllocElTySize*ArrayOffset)/CastElTySize) {
-    Value *Off = ConstantInt::get(Type::getInt32Ty(CI.getContext()),
+  if (uint64_t Offset = (AllocElTySize*ArrayOffset)/CastElTySize) {
+    Value *Off = ConstantInt::get(AI.getArraySize()->getType(),
                                   Offset, true);
     Amt = AllocaBuilder.CreateAdd(Amt, Off, "tmp");
   }
@@ -1098,6 +1097,33 @@ Instruction *InstCombiner::visitFPTrunc(FPTruncInst &CI) {
       break;  
     }
   }
+  
+  // Fold (fptrunc (sqrt (fpext x))) -> (sqrtf x)
+  // NOTE: This should be disabled by -fno-builtin-sqrt if we ever support it.
+  CallInst *Call = dyn_cast<CallInst>(CI.getOperand(0));
+  if (Call && Call->getCalledFunction() &&
+      Call->getCalledFunction()->getName() == "sqrt" &&
+      Call->getNumArgOperands() == 1) {
+    CastInst *Arg = dyn_cast<CastInst>(Call->getArgOperand(0));
+    if (Arg && Arg->getOpcode() == Instruction::FPExt &&
+        CI.getType()->isFloatTy() &&
+        Call->getType()->isDoubleTy() &&
+        Arg->getType()->isDoubleTy() &&
+        Arg->getOperand(0)->getType()->isFloatTy()) {
+      Function *Callee = Call->getCalledFunction();
+      Module *M = CI.getParent()->getParent()->getParent();
+      Constant* SqrtfFunc = M->getOrInsertFunction("sqrtf", 
+                                                   Callee->getAttributes(),
+                                                   Builder->getFloatTy(),
+                                                   Builder->getFloatTy(),
+                                                   NULL);
+      CallInst *ret = CallInst::Create(SqrtfFunc, Arg->getOperand(0),
+                                       "sqrtfcall");
+      ret->setAttributes(Callee->getAttributes());
+      return ret;
+    }
+  }
+  
   return 0;
 }