Change behavior of calling bitcasted alias functions.

It will now only convert the arguments / return value and call
the underlying function if the types are able to be bitcasted.
This avoids using fp<->int conversions that would occur before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187444 91177308-0d34-0410-b5e6-96231b3b80d8

10 files changed, 430 insertions, 83 deletions

diff --git a/include/llvm/IR/InstrTypes.h b/include/llvm/IR/InstrTypes.h
index f50c28a5fc..e12bb03a40 100644
--- a/include/llvm/IR/InstrTypes.h
+++ b/include/llvm/IR/InstrTypes.h
@@ -531,6 +531,12 @@ public:
     Type *DestTy ///< The Type to which the value should be cast.
   );
 
+  /// @brief Check whether a bitcast between these types is valid
+  static bool isBitCastable(
+    Type *SrcTy, ///< The Type from which the value should be cast.
+    Type *DestTy ///< The Type to which the value should be cast.
+  );
+
   /// Returns the opcode necessary to cast Val into Ty using usual casting
   /// rules.
   /// @brief Infer the opcode for cast operand and type
@@ -698,7 +704,7 @@ public:
   /// @brief Create a CmpInst
   static CmpInst *Create(OtherOps Op, unsigned short predicate, Value *S1,
                          Value *S2, const Twine &Name, BasicBlock *InsertAtEnd);
-  
+
   /// @brief Get the opcode casted to the right type
   OtherOps getOpcode() const {
     return static_cast<OtherOps>(Instruction::getOpcode());
@@ -715,15 +721,15 @@ public:
   static bool isFPPredicate(Predicate P) {
     return P >= FIRST_FCMP_PREDICATE && P <= LAST_FCMP_PREDICATE;
   }
-  
+
   static bool isIntPredicate(Predicate P) {
     return P >= FIRST_ICMP_PREDICATE && P <= LAST_ICMP_PREDICATE;
   }
-  
+
   bool isFPPredicate() const { return isFPPredicate(getPredicate()); }
   bool isIntPredicate() const { return isIntPredicate(getPredicate()); }
-  
-  
+
+
   /// For example, EQ -> NE, UGT -> ULE, SLT -> SGE,
   ///              OEQ -> UNE, UGT -> OLE, OLT -> UGE, etc.
   /// @returns the inverse predicate for the instruction's current predicate.
@@ -821,7 +827,7 @@ public:
   static inline bool classof(const Value *V) {
     return isa<Instruction>(V) && classof(cast<Instruction>(V));
   }
-  
+
   /// @brief Create a result type for fcmp/icmp
   static Type* makeCmpResultType(Type* opnd_type) {
     if (VectorType* vt = dyn_cast<VectorType>(opnd_type)) {
diff --git a/lib/IR/Instructions.cpp b/lib/IR/Instructions.cpp
index 5878f77dc1..665fe66ccc 100644
--- a/lib/IR/Instructions.cpp
+++ b/lib/IR/Instructions.cpp
@@ -2517,8 +2517,48 @@ bool CastInst::isCastable(Type *SrcTy, Type *DestTy) {
   }
 }
 
-// Provide a way to get a "cast" where the cast opcode is inferred from the 
-// types and size of the operand. This, basically, is a parallel of the 
+bool CastInst::isBitCastable(Type *SrcTy, Type *DestTy) {
+  if (!SrcTy->isFirstClassType() || !DestTy->isFirstClassType())
+    return false;
+
+  if (SrcTy == DestTy)
+    return true;
+
+  if (VectorType *SrcVecTy = dyn_cast<VectorType>(SrcTy)) {
+    if (VectorType *DestVecTy = dyn_cast<VectorType>(DestTy)) {
+      if (SrcVecTy->getNumElements() == DestVecTy->getNumElements()) {
+        // An element by element cast. Valid if casting the elements is valid.
+        SrcTy = SrcVecTy->getElementType();
+        DestTy = DestVecTy->getElementType();
+      }
+    }
+  }
+
+  if (PointerType *DestPtrTy = dyn_cast<PointerType>(DestTy)) {
+    if (PointerType *SrcPtrTy = dyn_cast<PointerType>(SrcTy)) {
+      return SrcPtrTy->getAddressSpace() == DestPtrTy->getAddressSpace();
+    }
+  }
+
+  unsigned SrcBits = SrcTy->getPrimitiveSizeInBits();   // 0 for ptr
+  unsigned DestBits = DestTy->getPrimitiveSizeInBits(); // 0 for ptr
+
+  // Could still have vectors of pointers if the number of elements doesn't
+  // match
+  if (SrcBits == 0 || DestBits == 0)
+    return false;
+
+  if (SrcBits != DestBits)
+    return false;
+
+  if (DestTy->isX86_MMXTy() || SrcTy->isX86_MMXTy())
+    return false;
+
+  return true;
+}
+
+// Provide a way to get a "cast" where the cast opcode is inferred from the
+// types and size of the operand. This, basically, is a parallel of the
 // logic in the castIsValid function below.  This axiom should hold:
 //   castIsValid( getCastOpcode(Val, Ty), Val, Ty)
 // should not assert in castIsValid. In other words, this produces a "correct"
@@ -2535,6 +2575,7 @@ CastInst::getCastOpcode(
   if (SrcTy == DestTy)
     return BitCast;
 
+  // FIXME: Check address space sizes here
   if (VectorType *SrcVecTy = dyn_cast<VectorType>(SrcTy))
     if (VectorType *DestVecTy = dyn_cast<VectorType>(DestTy))
       if (SrcVecTy->getNumElements() == DestVecTy->getNumElements()) {
@@ -2601,6 +2642,7 @@ CastInst::getCastOpcode(
     return BitCast;
   } else if (DestTy->isPointerTy()) {
     if (SrcTy->isPointerTy()) {
+      // TODO: Address space pointer sizes may not match
       return BitCast;                               // ptr -> ptr
     } else if (SrcTy->isIntegerTy()) {
       return IntToPtr;                              // int -> ptr
diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 815db335ea..9f74fd6b8d 100644
--- a/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -1010,7 +1010,8 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
 
     if (!Caller->use_empty() &&
         // void -> non-void is handled specially
-        !NewRetTy->isVoidTy() && !CastInst::isCastable(NewRetTy, OldRetTy))
+        !NewRetTy->isVoidTy() &&
+        !CastInst::isBitCastable(NewRetTy, OldRetTy))
       return false;   // Cannot transform this return value.
 
     if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
@@ -1044,8 +1045,9 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
     Type *ParamTy = FT->getParamType(i);
     Type *ActTy = (*AI)->getType();
 
-    if (!CastInst::isCastable(ActTy, ParamTy))
+    if (!CastInst::isBitCastable(ActTy, ParamTy)) {
       return false;   // Cannot transform this parameter value.
+    }
 
     if (AttrBuilder(CallerPAL.getParamAttributes(i + 1), i + 1).
           hasAttributes(AttributeFuncs::
@@ -1074,7 +1076,8 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
       ParamTy == TD->getIntPtrType(Caller->getContext())) &&
               (ActTy->isPointerTy() ||
               ActTy == TD->getIntPtrType(Caller->getContext()))));
-    if (Callee->isDeclaration() && !isConvertible) return false;
+    if (Callee->isDeclaration() && !isConvertible)
+      return false;
   }
 
   if (Callee->isDeclaration()) {
@@ -1141,12 +1144,11 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
   AI = CS.arg_begin();
   for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
     Type *ParamTy = FT->getParamType(i);
+
     if ((*AI)->getType() == ParamTy) {
       Args.push_back(*AI);
     } else {
-      Instruction::CastOps opcode = CastInst::getCastOpcode(*AI,
-          false, ParamTy, false);
-      Args.push_back(Builder->CreateCast(opcode, *AI, ParamTy));
+      Args.push_back(Builder->CreateBitCast(*AI, ParamTy));
     }
 
     // Add any parameter attributes.
@@ -1217,9 +1219,7 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
   Value *NV = NC;
   if (OldRetTy != NV->getType() && !Caller->use_empty()) {
     if (!NV->getType()->isVoidTy()) {
-      Instruction::CastOps opcode =
-        CastInst::getCastOpcode(NC, false, OldRetTy, false);
-      NV = NC = CastInst::Create(opcode, NC, OldRetTy);
+      NV = NC = CastInst::Create(CastInst::BitCast, NC, OldRetTy);
       NC->setDebugLoc(Caller->getDebugLoc());
 
       // If this is an invoke instruction, we should insert it after the first
diff --git a/test/Transforms/InstCombine/2008-01-06-BitCastAttributes.ll b/test/Transforms/InstCombine/2008-01-06-BitCastAttributes.ll
index 6588a918ba..22c078250e 100644
--- a/test/Transforms/InstCombine/2008-01-06-BitCastAttributes.ll
+++ b/test/Transforms/InstCombine/2008-01-06-BitCastAttributes.ll
@@ -4,22 +4,27 @@
 ; CHECK-NOT: bitcast
 
 define void @a() {
-	ret void
+  ret void
 }
 
 define signext i32 @b(i32* inreg  %x)   {
-	ret i32 0
+  ret i32 0
 }
 
 define void @c(...) {
-	ret void
+  ret void
 }
 
 define void @g(i32* %y) {
-	call void bitcast (void ()* @a to void (i32*)*)( i32* noalias  %y )
-	call <2 x i32> bitcast (i32 (i32*)* @b to <2 x i32> (i32*)*)( i32* inreg  null )		; <<2 x i32>>:1 [#uses=0]
+; CHECK-LABEL: @g(
+; CHECK: call i64 bitcast (i32 (i32*)* @b to i64 (i32)*)(i32 0)
 	%x = call i64 bitcast (i32 (i32*)* @b to i64 (i32)*)( i32 0 )		; <i64> [#uses=0]
-	call void bitcast (void (...)* @c to void (i32)*)( i32 0 )
-	call void bitcast (void (...)* @c to void (i32)*)( i32 zeroext  0 )
-	ret void
+
+; The rest should not have bitcasts remaining
+; CHECK-NOT: bitcast
+  call void bitcast (void ()* @a to void (i32*)*)( i32* noalias  %y )
+  call <2 x i32> bitcast (i32 (i32*)* @b to <2 x i32> (i32*)*)( i32* inreg  null )		; <<2 x i32>>:1 [#uses=0]
+  call void bitcast (void (...)* @c to void (i32)*)( i32 0 )
+  call void bitcast (void (...)* @c to void (i32)*)( i32 zeroext  0 )
+  ret void
 }
diff --git a/test/Transforms/InstCombine/2008-01-06-VoidCast.ll b/test/Transforms/InstCombine/2008-01-06-VoidCast.ll
index ebd5c4229d..5dcaa38edc 100644
--- a/test/Transforms/InstCombine/2008-01-06-VoidCast.ll
+++ b/test/Transforms/InstCombine/2008-01-06-VoidCast.ll
@@ -1,10 +1,12 @@
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
 define void @f(i16 %y) {
-	ret void
+  ret void
 }
-; CHECK-NOT: bitcast
+
 define i32 @g(i32 %y) {
-	%x = call i32 bitcast (void (i16)* @f to i32 (i32)*)( i32 %y )		; <i32> [#uses=1]
-	ret i32 %x
+; CHECK-LABEL: @g(
+; CHECK: call i32 bitcast
+  %x = call i32 bitcast (void (i16)* @f to i32 (i32)*)( i32 %y )		; <i32> [#uses=1]
+  ret i32 %x
 }
diff --git a/test/Transforms/InstCombine/apint-call-cast-target.ll b/test/Transforms/InstCombine/apint-call-cast-target.ll
index 7d7c8fa503..4e98f9b2b3 100644
--- a/test/Transforms/InstCombine/apint-call-cast-target.ll
+++ b/test/Transforms/InstCombine/apint-call-cast-target.ll
@@ -3,16 +3,17 @@
 target datalayout = "e-p:32:32"
 target triple = "i686-pc-linux-gnu"
 
-; CHECK-NOT: bitcast
-; CHECK: call
-; CHECK-NOT: bitcast
 define i32 @main() {
+; CHECK-LABEL: @main(
+; CHECK: call i32 bitcast
 entry:
 	%tmp = call i32 bitcast (i7* (i999*)* @ctime to i32 (i99*)*)( i99* null )
 	ret i32 %tmp
 }
 
 define i7* @ctime(i999*) {
+; CHECK-LABEL: @ctime(
+; CHECK: call i7* bitcast
 entry:
 	%tmp = call i7* bitcast (i32 ()* @main to i7* ()*)( )
 	ret i7* %tmp
diff --git a/test/Transforms/InstCombine/bitcast-alias-function.ll b/test/Transforms/InstCombine/bitcast-alias-function.ll
new file mode 100644
index 0000000000..a6b56f94ff
--- /dev/null
+++ b/test/Transforms/InstCombine/bitcast-alias-function.ll
@@ -0,0 +1,229 @@
+; RUN: opt -S -instcombine -o - %s | FileCheck %s
+target datalayout = "e-p:32:32:32-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v64:64:64-v128:128:128-a0:0:64"
+
+
+
+; Cases that should be bitcast
+
+; Test cast between scalars with same bit sizes
+@alias_i32_to_f32 = alias bitcast (i32 (i32)* @func_i32 to float (float)*)
+
+; Test cast between vectors with same number of elements and bit sizes
+@alias_v2i32_to_v2f32 = alias bitcast (<2 x i32> (<2 x i32>)* @func_v2i32 to <2 x float> (<2 x float>)*)
+
+; Test cast from vector to scalar with same number of bits
+@alias_v2f32_to_i64 = alias bitcast (i64 (i64)* @func_i64 to <2 x float> (<2 x float>)*)
+
+; Test cast from scalar to vector with same number of bits
+@alias_i64_to_v2f32 = alias bitcast (<2 x float> (<2 x float>)* @func_v2f32 to i64 (i64)*)
+
+; Test cast between vectors of pointers
+@alias_v2i32p_to_v2i64p = alias bitcast (<2 x i32*> (<2 x i32*>)* @func_v2i32p to <2 x i64*> (<2 x i64*>)*)
+
+
+; Cases that should be invalid and unchanged
+
+; Test cast between scalars with different bit sizes
+@alias_i64_to_f32 = alias bitcast (i64 (i64)* @func_i64 to float (float)*)
+
+; Test cast between vectors with different bit sizes but the
+; same number of elements
+@alias_v2i64_to_v2f32 = alias bitcast (<2 x i64> (<2 x i64>)* @func_v2i64 to <2 x float> (<2 x float>)*)
+
+; Test cast between vectors with same number of bits and different
+; numbers of elements
+@alias_v2i32_to_v4f32 = alias bitcast (<2 x i32> (<2 x i32>)* @func_v2i32 to <4 x float> (<4 x float>)*)
+
+; Test cast between scalar and vector with different number of bits
+@alias_i64_to_v4f32 = alias bitcast (<4 x float> (<4 x float>)* @func_v4f32 to i64 (i64)*)
+
+; Test cast between vector and scalar with different number of bits
+@alias_v4f32_to_i64 = alias bitcast (i64 (i64)* @func_i64 to <4 x float> (<4 x float>)*)
+
+; Test cast from scalar to vector of pointers with same number of bits
+; We don't know the pointer size at this point, so this can't be done
+@alias_i64_to_v2i32p = alias bitcast (<2 x i32*> (<2 x i32*>)* @func_v2i32p to i64 (i64)*)
+
+; Test cast between vector of pointers and scalar with different number of bits
+@alias_v4i32p_to_i64 = alias bitcast (i64 (i64)* @func_i64 to <4 x i32*> (<4 x i32*>)*)
+
+
+
+define internal <2 x i32> @func_v2i32(<2 x i32> %v) noinline nounwind {
+entry:
+  ret <2 x i32> %v
+}
+
+define internal <2 x float> @func_v2f32(<2 x float> %v) noinline nounwind {
+entry:
+  ret <2 x float> %v
+}
+
+define internal <4 x float> @func_v4f32(<4 x float> %v) noinline nounwind {
+entry:
+  ret <4 x float> %v
+}
+
+define internal i32 @func_i32(i32 %v) noinline nounwind {
+entry:
+  ret i32 %v
+}
+
+define internal i64 @func_i64(i64 %v) noinline nounwind {
+entry:
+  ret i64 %v
+}
+
+define internal <2 x i64> @func_v2i64(<2 x i64> %v) noinline nounwind {
+entry:
+  ret <2 x i64> %v
+}
+
+define internal <2 x i32*> @func_v2i32p(<2 x i32*> %v) noinline nounwind {
+entry:
+  ret <2 x i32*> %v
+}
+
+; Valid cases, only bitcast for argument / return type and call underlying function
+
+; Sizes match, should only bitcast
+define void @bitcast_alias_scalar(float* noalias %source, float* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_scalar
+; CHECK: bitcast float %tmp to i32
+; CHECK-NOT: fptoui
+; CHECK-NOT: uitofp
+; CHECK: bitcast i32 %call to float
+  %tmp = load float* %source, align 8
+  %call = call float @alias_i32_to_f32(float %tmp) nounwind
+  store float %call, float* %dest, align 8
+  ret void
+}
+
+; Sizes match, should only bitcast
+define void @bitcast_alias_vector(<2 x float>* noalias %source, <2 x float>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector
+; CHECK: bitcast <2 x float> %tmp to <2 x i32>
+; CHECK-NOT: fptoui
+; CHECK-NOT: uitofp
+; CHECK: bitcast <2 x i32> %call to <2 x float>
+  %tmp = load <2 x float>* %source, align 8
+  %call = call <2 x float> @alias_v2i32_to_v2f32(<2 x float> %tmp) nounwind
+  store <2 x float> %call, <2 x float>* %dest, align 8
+  ret void
+}
+
+; Sizes match, should only bitcast
+define void @bitcast_alias_vector_scalar_same_size(<2 x float>* noalias %source, <2 x float>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector_scalar_same_size
+; CHECK: bitcast <2 x float> %tmp to i64
+; CHECK: %call = call i64 @func_i64
+; CHECK: bitcast i64 %call to <2 x float>
+  %tmp = load <2 x float>* %source, align 8
+  %call = call <2 x float> @alias_v2f32_to_i64(<2 x float> %tmp) nounwind
+  store <2 x float> %call, <2 x float>* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_scalar_vector_same_size(i64* noalias %source, i64* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_scalar_vector_same_size
+; CHECK: bitcast i64 %tmp to <2 x float>
+; CHECK: call <2 x float> @func_v2f32
+; CHECK: bitcast <2 x float> %call to i64
+  %tmp = load i64* %source, align 8
+  %call = call i64 @alias_i64_to_v2f32(i64 %tmp) nounwind
+  store i64 %call, i64* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_vector_ptrs_same_size(<2 x i64*>* noalias %source, <2 x i64*>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector_ptrs_same_size
+; CHECK: bitcast <2 x i64*> %tmp to <2 x i32*>
+; CHECK: call <2 x i32*> @func_v2i32p
+; CHECK: bitcast <2 x i32*> %call to <2 x i64*>
+  %tmp = load <2 x i64*>* %source, align 8
+  %call = call <2 x i64*> @alias_v2i32p_to_v2i64p(<2 x i64*> %tmp) nounwind
+  store <2 x i64*> %call, <2 x i64*>* %dest, align 8
+  ret void
+}
+
+; Invalid cases:
+
+define void @bitcast_alias_mismatch_scalar_size(float* noalias %source, float* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_mismatch_scalar_size
+; CHECK-NOT: fptoui
+; CHECK: @alias_i64_to_f32
+; CHECK-NOT: uitofp
+  %tmp = load float* %source, align 8
+  %call = call float @alias_i64_to_f32(float %tmp) nounwind
+  store float %call, float* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_mismatch_vector_element_and_bit_size(<2 x float>* noalias %source, <2 x float>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_mismatch_vector_element_and_bit_size
+; CHECK-NOT: fptoui <2 x float> %tmp to <2 x i64>
+; CHECK: @alias_v2i64_to_v2f32
+; CHECK-NOT: uitofp <2 x i64> %call to <2 x float>
+  %tmp = load <2 x float>* %source, align 8
+  %call = call <2 x float> @alias_v2i64_to_v2f32(<2 x float> %tmp) nounwind
+  store <2 x float> %call, <2 x float>* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_vector_mismatched_number_elements(<4 x float>* noalias %source, <4 x float>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector_mismatched_number_elements
+; CHECK:  %call = call <4 x float> @alias_v2i32_to_v4f32
+  %tmp = load <4 x float>* %source, align 8
+  %call = call <4 x float> @alias_v2i32_to_v4f32(<4 x float> %tmp) nounwind
+  store <4 x float> %call, <4 x float>* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_vector_scalar_mismatched_bit_size(<4 x float>* noalias %source, <4 x float>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector_scalar_mismatched_bit_size
+; CHECK:  %call = call <4 x float> @alias_v4f32_to_i64
+  %tmp = load <4 x float>* %source, align 8
+  %call = call <4 x float> @alias_v4f32_to_i64(<4 x float> %tmp) nounwind
+  store <4 x float> %call, <4 x float>* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_vector_ptrs_scalar_mismatched_bit_size(<4 x i32*>* noalias %source, <4 x i32*>* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_vector_ptrs_scalar_mismatched_bit_size
+; CHECK: @alias_v4i32p_to_i64
+  %tmp = load <4 x i32*>* %source, align 8
+  %call = call <4 x i32*> @alias_v4i32p_to_i64(<4 x i32*> %tmp) nounwind
+  store <4 x i32*> %call, <4 x i32*>* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_scalar_vector_ptrs_same_size(i64* noalias %source, i64* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_scalar_vector_ptrs_same_size
+; CHECK: @alias_i64_to_v2i32p
+  %tmp = load i64* %source, align 8
+  %call = call i64 @alias_i64_to_v2i32p(i64 %tmp) nounwind
+  store i64 %call, i64* %dest, align 8
+  ret void
+}
+
+define void @bitcast_alias_scalar_vector_mismatched_bit_size(i64* noalias %source, i64* noalias %dest) nounwind {
+entry:
+; CHECK-LABEL: @bitcast_alias_scalar_vector_mismatched_bit_size
+; CHECK: call i64 @alias_i64_to_v4f32
+  %tmp = load i64* %source, align 8
+  %call = call i64 @alias_i64_to_v4f32(i64 %tmp) nounwind
+  store i64 %call, i64* %dest, align 8
+  ret void
+}
+
diff --git a/test/Transforms/InstCombine/call-cast-target.ll b/test/Transforms/InstCombine/call-cast-target.ll
index 19074212d3..315c51683f 100644
--- a/test/Transforms/InstCombine/call-cast-target.ll
+++ b/test/Transforms/InstCombine/call-cast-target.ll
@@ -3,14 +3,12 @@
 target datalayout = "e-p:32:32"
 target triple = "i686-pc-linux-gnu"
 
-; CHECK-NOT: bitcast
-; CHECK: call
-; CHECK-NOT: bitcast
-
 define i32 @main() {
+; CHECK-LABEL: @main
+; CHECK: call i32 bitcast
 entry:
-        %tmp = call i32 bitcast (i8* (i32*)* @ctime to i32 (i32*)*)( i32* null )          ; <i32> [#uses=1]
-        ret i32 %tmp
+  %tmp = call i32 bitcast (i8* (i32*)* @ctime to i32 (i32*)*)( i32* null )          ; <i32> [#uses=1]
+  ret i32 %tmp
 }
 
 declare i8* @ctime(i32*)
diff --git a/test/Transforms/InstCombine/call.ll b/test/Transforms/InstCombine/call.ll
index 974a871820..55833fbd1c 100644
--- a/test/Transforms/InstCombine/call.ll
+++ b/test/Transforms/InstCombine/call.ll
@@ -7,92 +7,94 @@ target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:1
 declare void @test1a(i8*)
 
 define void @test1(i32* %A) {
-        call void bitcast (void (i8*)* @test1a to void (i32*)*)( i32* %A )
-        ret void
+; CHECK-LABEL: @test1(
 ; CHECK: %1 = bitcast i32* %A to i8*
 ; CHECK: call void @test1a(i8* %1)
 ; CHECK: ret void
+  call void bitcast (void (i8*)* @test1a to void (i32*)*)( i32* %A )
+  ret void
 }
 
-; More complex case, translate argument because of resolution.  This is safe 
+; More complex case, translate argument because of resolution.  This is safe
 ; because we have the body of the function
 define void @test2a(i8 %A) {
-        ret void
+; CHECK-LABEL: @test2a(
 ; CHECK: ret void
+  ret void
 }
 
 define i32 @test2(i32 %A) {
-        call void bitcast (void (i8)* @test2a to void (i32)*)( i32 %A )
-        ret i32 %A
-; CHECK: %1 = trunc i32 %A to i8
-; CHECK: call void @test2a(i8 %1)
+; CHECK-LABEL: @test2(
+; CHECK: call void bitcast
 ; CHECK: ret i32 %A
+  call void bitcast (void (i8)* @test2a to void (i32)*)( i32 %A )
+  ret i32 %A
 }
 
 
-; Resolving this should insert a cast from sbyte to int, following the C 
+; Resolving this should insert a cast from sbyte to int, following the C
 ; promotion rules.
 define void @test3a(i8, ...) {unreachable }
 
 define void @test3(i8 %A, i8 %B) {
-        call void bitcast (void (i8, ...)* @test3a to void (i8, i8)*)( i8 %A, i8 %B 
-)
-        ret void
+; CHECK-LABEL: @test3(
 ; CHECK: %1 = zext i8 %B to i32
 ; CHECK: call void (i8, ...)* @test3a(i8 %A, i32 %1)
 ; CHECK: ret void
+  call void bitcast (void (i8, ...)* @test3a to void (i8, i8)*)( i8 %A, i8 %B)
+  ret void
 }
 
-
 ; test conversion of return value...
 define i8 @test4a() {
-        ret i8 0
+; CHECK-LABEL: @test4a(
 ; CHECK: ret i8 0
+  ret i8 0
 }
 
 define i32 @test4() {
-        %X = call i32 bitcast (i8 ()* @test4a to i32 ()*)( )            ; <i32> [#uses=1]
-        ret i32 %X
-; CHECK: %X = call i8 @test4a()
-; CHECK: %1 = zext i8 %X to i32
-; CHECK: ret i32 %1
+; CHECK-LABEL: @test4(
+; CHECK: call i32 bitcast
+  %X = call i32 bitcast (i8 ()* @test4a to i32 ()*)( )            ; <i32> [#uses=1]
+  ret i32 %X
 }
 
-
-; test conversion of return value... no value conversion occurs so we can do 
+; test conversion of return value... no value conversion occurs so we can do
 ; this with just a prototype...
 declare i32 @test5a()
 
 define i32 @test5() {
-        %X = call i32 @test5a( )                ; <i32> [#uses=1]
-        ret i32 %X
+; CHECK-LABEL: @test5(
 ; CHECK: %X = call i32 @test5a()
 ; CHECK: ret i32 %X
+  %X = call i32 @test5a( )                ; <i32> [#uses=1]
+  ret i32 %X
 }
 
-
 ; test addition of new arguments...
 declare i32 @test6a(i32)
 
 define i32 @test6() {
-        %X = call i32 bitcast (i32 (i32)* @test6a to i32 ()*)( )
-        ret i32 %X
+; CHECK-LABEL: @test6(
 ; CHECK: %X = call i32 @test6a(i32 0)
 ; CHECK: ret i32 %X
+  %X = call i32 bitcast (i32 (i32)* @test6a to i32 ()*)( )
+  ret i32 %X
 }
 
-
 ; test removal of arguments, only can happen with a function body
 define void @test7a() {
-        ret void
+; CHECK-LABEL: @test7a(
 ; CHECK: ret void
+  ret void
 }
 
 define void @test7() {
-        call void bitcast (void ()* @test7a to void (i32)*)( i32 5 )
-        ret void
+; CHECK-LABEL: @test7(
 ; CHECK: call void @test7a()
 ; CHECK: ret void
+  call void bitcast (void ()* @test7a to void (i32)*)( i32 5 )
+  ret void
 }
 
 
@@ -100,6 +102,11 @@ define void @test7() {
 declare void @test8a()
 
 define i8* @test8() {
+; CHECK-LABEL: @test8(
+; CHECK-NEXT: invoke void @test8a()
+; Don't turn this into "unreachable": the callee and caller don't agree in
+; calling conv, but the implementation of test8a may actually end up using the
+; right calling conv.
   invoke void @test8a()
           to label %invoke.cont unwind label %try.handler
 
@@ -114,19 +121,13 @@ try.handler:                                      ; preds = %entry
 
 declare i32 @__gxx_personality_v0(...)
 
-; Don't turn this into "unreachable": the callee and caller don't agree in
-; calling conv, but the implementation of test8a may actually end up using the
-; right calling conv.
-; CHECK: @test8() {
-; CHECK-NEXT: invoke void @test8a()
-
-
 
-; Don't turn this into a direct call, because test9x is just a prototype and 
+; Don't turn this into a direct call, because test9x is just a prototype and
 ; doing so will make it varargs.
 ; rdar://9038601
 declare i8* @test9x(i8*, i8*, ...) noredzone
 define i8* @test9(i8* %arg, i8* %tmp3) nounwind ssp noredzone {
+; CHECK-LABEL: @test9
 entry:
   %call = call i8* bitcast (i8* (i8*, i8*, ...)* @test9x to i8* (i8*, i8*)*)(i8* %arg, i8* %tmp3) noredzone
   ret i8* %call
diff --git a/unittests/IR/InstructionsTest.cpp b/unittests/IR/InstructionsTest.cpp
index 3aa30af6b9..34d662dcb3 100644
--- a/unittests/IR/InstructionsTest.cpp
+++ b/unittests/IR/InstructionsTest.cpp
@@ -116,11 +116,35 @@ TEST(InstructionsTest, BranchInst) {
 TEST(InstructionsTest, CastInst) {
   LLVMContext &C(getGlobalContext());
 
-  Type* Int8Ty = Type::getInt8Ty(C);
-  Type* Int64Ty = Type::getInt64Ty(C);
-  Type* V8x8Ty = VectorType::get(Int8Ty, 8);
-  Type* V8x64Ty = VectorType::get(Int64Ty, 8);
-  Type* X86MMXTy = Type::getX86_MMXTy(C);
+  Type *Int8Ty = Type::getInt8Ty(C);
+  Type *Int16Ty = Type::getInt16Ty(C);
+  Type *Int32Ty = Type::getInt32Ty(C);
+  Type *Int64Ty = Type::getInt64Ty(C);
+  Type *V8x8Ty = VectorType::get(Int8Ty, 8);
+  Type *V8x64Ty = VectorType::get(Int64Ty, 8);
+  Type *X86MMXTy = Type::getX86_MMXTy(C);
+
+  Type *HalfTy = Type::getHalfTy(C);
+  Type *FloatTy = Type::getFloatTy(C);
+  Type *DoubleTy = Type::getDoubleTy(C);
+
+  Type *V2Int32Ty = VectorType::get(Int32Ty, 2);
+  Type *V2Int64Ty = VectorType::get(Int64Ty, 2);
+  Type *V4Int16Ty = VectorType::get(Int16Ty, 4);
+
+  Type *Int32PtrTy = PointerType::get(Int32Ty, 0);
+  Type *Int64PtrTy = PointerType::get(Int64Ty, 0);
+
+  Type *Int32PtrAS1Ty = PointerType::get(Int32Ty, 1);
+  Type *Int64PtrAS1Ty = PointerType::get(Int64Ty, 1);
+
+  Type *V2Int32PtrAS1Ty = VectorType::get(Int32PtrAS1Ty, 2);
+  Type *V2Int64PtrAS1Ty = VectorType::get(Int64PtrAS1Ty, 2);
+  Type *V4Int32PtrAS1Ty = VectorType::get(Int32PtrAS1Ty, 4);
+  Type *V4Int64PtrAS1Ty = VectorType::get(Int64PtrAS1Ty, 4);
+
+  Type *V2Int64PtrTy = VectorType::get(Int64PtrTy, 2);
+  Type *V2Int32PtrTy = VectorType::get(Int32PtrTy, 2);
 
   const Constant* c8 = Constant::getNullValue(V8x8Ty);
   const Constant* c64 = Constant::getNullValue(V8x64Ty);
@@ -132,9 +156,48 @@ TEST(InstructionsTest, CastInst) {
   EXPECT_TRUE(CastInst::isCastable(V8x8Ty, V8x64Ty));
   EXPECT_EQ(CastInst::Trunc, CastInst::getCastOpcode(c64, true, V8x8Ty, true));
   EXPECT_EQ(CastInst::SExt, CastInst::getCastOpcode(c8, true, V8x64Ty, true));
-}
-
 
+  EXPECT_FALSE(CastInst::isBitCastable(V8x8Ty, X86MMXTy));
+  EXPECT_FALSE(CastInst::isBitCastable(X86MMXTy, V8x8Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int64Ty, X86MMXTy));
+  EXPECT_FALSE(CastInst::isBitCastable(V8x64Ty, V8x8Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(V8x8Ty, V8x64Ty));
+
+  // Check address space casts are rejected since we don't know the sizes here
+  EXPECT_FALSE(CastInst::isBitCastable(Int32PtrTy, Int32PtrAS1Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int32PtrAS1Ty, Int32PtrTy));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrTy, V2Int32PtrAS1Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrAS1Ty, V2Int32PtrTy));
+  EXPECT_TRUE(CastInst::isBitCastable(V2Int32PtrAS1Ty, V2Int64PtrAS1Ty));
+
+  // Test mismatched number of elements for pointers
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrAS1Ty, V4Int64PtrAS1Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(V4Int64PtrAS1Ty, V2Int32PtrAS1Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrAS1Ty, V4Int32PtrAS1Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int32PtrTy, V2Int32PtrTy));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrTy, Int32PtrTy));
+
+  EXPECT_TRUE(CastInst::isBitCastable(Int32PtrTy, Int64PtrTy));
+  EXPECT_FALSE(CastInst::isBitCastable(DoubleTy, FloatTy));
+  EXPECT_FALSE(CastInst::isBitCastable(FloatTy, DoubleTy));
+  EXPECT_TRUE(CastInst::isBitCastable(FloatTy, FloatTy));
+  EXPECT_TRUE(CastInst::isBitCastable(FloatTy, FloatTy));
+  EXPECT_TRUE(CastInst::isBitCastable(FloatTy, Int32Ty));
+  EXPECT_TRUE(CastInst::isBitCastable(Int16Ty, HalfTy));
+  EXPECT_TRUE(CastInst::isBitCastable(Int32Ty, FloatTy));
+  EXPECT_TRUE(CastInst::isBitCastable(V2Int32Ty, Int64Ty));
+
+  EXPECT_TRUE(CastInst::isBitCastable(V2Int32Ty, V4Int16Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int32Ty, Int64Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int64Ty, Int32Ty));
+
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32PtrTy, Int64Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(Int64Ty, V2Int32PtrTy));
+  EXPECT_TRUE(CastInst::isBitCastable(V2Int64PtrTy, V2Int32PtrTy));
+  EXPECT_TRUE(CastInst::isBitCastable(V2Int32PtrTy, V2Int64PtrTy));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int32Ty, V2Int64Ty));
+  EXPECT_FALSE(CastInst::isBitCastable(V2Int64Ty, V2Int32Ty));
+}
 
 TEST(InstructionsTest, VectorGep) {
   LLVMContext &C(getGlobalContext());