diff --git a/examples/BrainF/BrainF.cpp b/examples/BrainF/BrainF.cpp
index c64b87f..5cf2b88 100644
--- a/examples/BrainF/BrainF.cpp
+++ b/examples/BrainF/BrainF.cpp
@@ -25,7 +25,6 @@
 
 #include "BrainF.h"
 #include "llvm/Constants.h"
-#include "llvm/Instructions.h"
 #include "llvm/Intrinsics.h"
 #include "llvm/ADT/STLExtras.h"
 #include <iostream>
@@ -79,11 +78,7 @@
 
   //%arr = malloc i8, i32 %d
   ConstantInt *val_mem = ConstantInt::get(C, APInt(32, memtotal));
-  BasicBlock* BB = builder->GetInsertBlock();
-  const Type* IntPtrTy = IntegerType::getInt32Ty(C);
-  ptr_arr = CallInst::CreateMalloc(BB, IntPtrTy, IntegerType::getInt8Ty(C),
-                                   val_mem, NULL, "arr");
-  BB->getInstList().push_back(cast<Instruction>(ptr_arr));
+  ptr_arr = builder->CreateMalloc(IntegerType::getInt8Ty(C), val_mem, "arr");
 
   //call void @llvm.memset.i32(i8 *%arr, i8 0, i32 %d, i32 1)
   {
diff --git a/include/llvm/Instructions.h b/include/llvm/Instructions.h
index f2832c8..fbee2af 100644
--- a/include/llvm/Instructions.h
+++ b/include/llvm/Instructions.h
@@ -1044,7 +1044,7 @@
                              const Twine &Name = "");
   static Value *CreateMalloc(BasicBlock *InsertAtEnd, const Type *IntPtrTy,
                              const Type *AllocTy, Value *ArraySize = 0,
-                             Function* MallocF = 0, const Twine &Name = "");
+                             const Twine &Name = "");
 
   ~CallInst();
 
@@ -1149,11 +1149,6 @@
   const Value *getCalledValue() const { return Op<0>(); }
         Value *getCalledValue()       { return Op<0>(); }
 
-  /// setCalledFunction - Set the function called
-  void setCalledFunction(Value* Fn) {
-    Op<0>() = Fn;
-  }
-
   // Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const CallInst *) { return true; }
   static inline bool classof(const Instruction *I) {
diff --git a/lib/AsmParser/LLLexer.cpp b/lib/AsmParser/LLLexer.cpp
index 05b6f7f..315048c 100644
--- a/lib/AsmParser/LLLexer.cpp
+++ b/lib/AsmParser/LLLexer.cpp
@@ -602,9 +602,6 @@
     // Scan CurPtr ahead, seeing if there is just whitespace before the newline.
     if (JustWhitespaceNewLine(CurPtr))
       return lltok::kw_zeroext;
-  } else if (Len == 6 && !memcmp(StartChar, "malloc", 6)) {
-    // Autoupgrade malloc instruction
-    return lltok::kw_malloc;
   }
 
   // Keywords for instructions.
@@ -644,6 +641,7 @@
   INSTKEYWORD(unwind,      Unwind);
   INSTKEYWORD(unreachable, Unreachable);
 
+  INSTKEYWORD(malloc,      Malloc);
   INSTKEYWORD(alloca,      Alloca);
   INSTKEYWORD(free,        Free);
   INSTKEYWORD(load,        Load);
diff --git a/lib/AsmParser/LLParser.cpp b/lib/AsmParser/LLParser.cpp
index d90588d..0ecf847 100644
--- a/lib/AsmParser/LLParser.cpp
+++ b/lib/AsmParser/LLParser.cpp
@@ -69,27 +69,6 @@
 /// ValidateEndOfModule - Do final validity and sanity checks at the end of the
 /// module.
 bool LLParser::ValidateEndOfModule() {
-  // Update auto-upgraded malloc calls from "autoupgrade_malloc" to "malloc".
-  if (MallocF) {
-    MallocF->setName("malloc");
-    // If setName() does not set the name to "malloc", then there is already a 
-    // declaration of "malloc".  In that case, iterate over all calls to MallocF
-    // and get them to call the declared "malloc" instead.
-    if (MallocF->getName() != "malloc") {
-      Function* realMallocF = M->getFunction("malloc");
-      for (User::use_iterator UI = MallocF->use_begin(), UE= MallocF->use_end();
-           UI != UE; ) {
-        User* user = *UI;
-        UI++;
-        if (CallInst *Call = dyn_cast<CallInst>(user))
-          Call->setCalledFunction(realMallocF);
-      }
-      if (!realMallocF->doesNotAlias(0)) realMallocF->setDoesNotAlias(0);
-      MallocF->eraseFromParent();
-      MallocF = NULL;
-    }
-  }
-
   if (!ForwardRefTypes.empty())
     return Error(ForwardRefTypes.begin()->second.second,
                  "use of undefined type named '" +
@@ -2797,8 +2776,8 @@
   case lltok::kw_call:           return ParseCall(Inst, PFS, false);
   case lltok::kw_tail:           return ParseCall(Inst, PFS, true);
   // Memory.
-  case lltok::kw_alloca:         return ParseAlloc(Inst, PFS);
-  case lltok::kw_malloc:         return ParseAlloc(Inst, PFS, BB, false);
+  case lltok::kw_alloca:
+  case lltok::kw_malloc:         return ParseAlloc(Inst, PFS, KeywordVal);
   case lltok::kw_free:           return ParseFree(Inst, PFS);
   case lltok::kw_load:           return ParseLoad(Inst, PFS, false);
   case lltok::kw_store:          return ParseStore(Inst, PFS, false);
@@ -3307,7 +3286,7 @@
 }
 
 /// ParsePHI
-///   ::= 'phi' Type '[' Value ',' Value ']' (',' '[' Value ',' Value√ü ']')*
+///   ::= 'phi' Type '[' Value ',' Value ']' (',' '[' Value ',' Valueß ']')*
 bool LLParser::ParsePHI(Instruction *&Inst, PerFunctionState &PFS) {
   PATypeHolder Ty(Type::getVoidTy(Context));
   Value *Op0, *Op1;
@@ -3452,7 +3431,7 @@
 ///   ::= 'malloc' Type (',' TypeAndValue)? (',' OptionalInfo)?
 ///   ::= 'alloca' Type (',' TypeAndValue)? (',' OptionalInfo)?
 bool LLParser::ParseAlloc(Instruction *&Inst, PerFunctionState &PFS,
-                          BasicBlock* BB, bool isAlloca) {
+                          unsigned Opc) {
   PATypeHolder Ty(Type::getVoidTy(Context));
   Value *Size = 0;
   LocTy SizeLoc;
@@ -3472,21 +3451,10 @@
   if (Size && Size->getType() != Type::getInt32Ty(Context))
     return Error(SizeLoc, "element count must be i32");
 
-  if (isAlloca)
+  if (Opc == Instruction::Malloc)
+    Inst = new MallocInst(Ty, Size, Alignment);
+  else
     Inst = new AllocaInst(Ty, Size, Alignment);
-  else {
-    // Autoupgrade old malloc instruction to malloc call.
-    const Type* IntPtrTy = Type::getInt32Ty(Context);
-    const Type* Int8PtrTy = PointerType::getUnqual(Type::getInt8Ty(Context));
-    if (!MallocF)
-      // Prototype malloc as "void *autoupgrade_malloc(int32)".
-      MallocF = cast<Function>(M->getOrInsertFunction("autoupgrade_malloc",
-                               Int8PtrTy, IntPtrTy, NULL));
-      // "autoupgrade_malloc" updated to "malloc" in ValidateEndOfModule().
-
-    Inst = cast<Instruction>(CallInst::CreateMalloc(BB, IntPtrTy, Ty,
-                                                    Size, MallocF));
-  }
   return false;
 }
 
diff --git a/lib/AsmParser/LLParser.h b/lib/AsmParser/LLParser.h
index 43e9bd8..3420fcf 100644
--- a/lib/AsmParser/LLParser.h
+++ b/lib/AsmParser/LLParser.h
@@ -75,11 +75,9 @@
     std::map<std::string, std::pair<GlobalValue*, LocTy> > ForwardRefVals;
     std::map<unsigned, std::pair<GlobalValue*, LocTy> > ForwardRefValIDs;
     std::vector<GlobalValue*> NumberedVals;
-    Function* MallocF;
   public:
     LLParser(MemoryBuffer *F, SourceMgr &SM, SMDiagnostic &Err, Module *m) : 
-      Context(m->getContext()), Lex(F, SM, Err, m->getContext()),
-      M(m), MallocF(NULL) {}
+      Context(m->getContext()), Lex(F, SM, Err, m->getContext()), M(m) {}
     bool Run();
 
     LLVMContext& getContext() { return Context; }
@@ -278,8 +276,7 @@
     bool ParseShuffleVector(Instruction *&I, PerFunctionState &PFS);
     bool ParsePHI(Instruction *&I, PerFunctionState &PFS);
     bool ParseCall(Instruction *&I, PerFunctionState &PFS, bool isTail);
-    bool ParseAlloc(Instruction *&I, PerFunctionState &PFS,
-                    BasicBlock *BB = 0, bool isAlloca = true);
+    bool ParseAlloc(Instruction *&I, PerFunctionState &PFS, unsigned Opc);
     bool ParseFree(Instruction *&I, PerFunctionState &PFS);
     bool ParseLoad(Instruction *&I, PerFunctionState &PFS, bool isVolatile);
     bool ParseStore(Instruction *&I, PerFunctionState &PFS, bool isVolatile);
diff --git a/lib/Bitcode/Reader/BitcodeReader.cpp b/lib/Bitcode/Reader/BitcodeReader.cpp
index be0ec4b..f3ab806 100644
--- a/lib/Bitcode/Reader/BitcodeReader.cpp
+++ b/lib/Bitcode/Reader/BitcodeReader.cpp
@@ -2046,21 +2046,14 @@
     }
 
     case bitc::FUNC_CODE_INST_MALLOC: { // MALLOC: [instty, op, align]
-      // Autoupgrade malloc instruction to malloc call.
       if (Record.size() < 3)
         return Error("Invalid MALLOC record");
       const PointerType *Ty =
         dyn_cast_or_null<PointerType>(getTypeByID(Record[0]));
       Value *Size = getFnValueByID(Record[1], Type::getInt32Ty(Context));
+      unsigned Align = Record[2];
       if (!Ty || !Size) return Error("Invalid MALLOC record");
-      if (!CurBB) return Error("Invalid malloc instruction with no BB");
-      const Type* Int32Ty = IntegerType::getInt32Ty(CurBB->getContext());
-      if (Size->getType() != Int32Ty)
-        Size = CastInst::CreateIntegerCast(Size, Int32Ty, false /*ZExt*/,
-                                           "", CurBB);
-      Value* Malloc = CallInst::CreateMalloc(CurBB, Int32Ty,
-                                             Ty->getElementType(), Size, NULL);
-      I = cast<Instruction>(Malloc);
+      I = new MallocInst(Ty->getElementType(), Size, (1 << Align) >> 1);
       InstructionList.push_back(I);
       break;
     }
diff --git a/lib/VMCore/Core.cpp b/lib/VMCore/Core.cpp
index 248127d..1dbf5c4 100644
--- a/lib/VMCore/Core.cpp
+++ b/lib/VMCore/Core.cpp
@@ -1636,16 +1636,12 @@
 
 LLVMValueRef LLVMBuildMalloc(LLVMBuilderRef B, LLVMTypeRef Ty,
                              const char *Name) {
-  const Type* IntPtrT = Type::getInt32Ty(unwrap(B)->GetInsertBlock()->getContext());
-  return wrap(CallInst::CreateMalloc(unwrap(B)->GetInsertBlock(), IntPtrT,
-                                     unwrap(Ty), 0, 0, Twine(Name)));
+  return wrap(unwrap(B)->CreateMalloc(unwrap(Ty), 0, Name));
 }
 
 LLVMValueRef LLVMBuildArrayMalloc(LLVMBuilderRef B, LLVMTypeRef Ty,
                                   LLVMValueRef Val, const char *Name) {
-  const Type* IntPtrT = Type::getInt32Ty(unwrap(B)->GetInsertBlock()->getContext());
-  return wrap(CallInst::CreateMalloc(unwrap(B)->GetInsertBlock(), IntPtrT,
-                                     unwrap(Ty), unwrap(Val), 0, Twine(Name)));
+  return wrap(unwrap(B)->CreateMalloc(unwrap(Ty), unwrap(Val), Name));
 }
 
 LLVMValueRef LLVMBuildAlloca(LLVMBuilderRef B, LLVMTypeRef Ty,
diff --git a/lib/VMCore/Instructions.cpp b/lib/VMCore/Instructions.cpp
index 611bf16..b7acce7 100644
--- a/lib/VMCore/Instructions.cpp
+++ b/lib/VMCore/Instructions.cpp
@@ -462,8 +462,7 @@
 
 static Value *createMalloc(Instruction *InsertBefore, BasicBlock *InsertAtEnd,
                            const Type *IntPtrTy, const Type *AllocTy,
-                           Value *ArraySize, Function* MallocF, 
-                           const Twine &NameStr) {
+                           Value *ArraySize, const Twine &NameStr) {
   assert(((!InsertBefore && InsertAtEnd) || (InsertBefore && !InsertAtEnd)) &&
          "createMalloc needs either InsertBefore or InsertAtEnd");
 
@@ -500,11 +499,10 @@
   BasicBlock* BB = InsertBefore ? InsertBefore->getParent() : InsertAtEnd;
   Module* M = BB->getParent()->getParent();
   const Type *BPTy = PointerType::getUnqual(Type::getInt8Ty(BB->getContext()));
-  if (!MallocF)
-    // prototype malloc as "void *malloc(size_t)"
-    MallocF = cast<Function>(M->getOrInsertFunction("malloc", BPTy,
-                                                    IntPtrTy, NULL));
-  if (!MallocF->doesNotAlias(0)) MallocF->setDoesNotAlias(0);
+  // prototype malloc as "void *malloc(size_t)"
+  Constant *MallocF = M->getOrInsertFunction("malloc", BPTy, IntPtrTy, NULL);
+  if (!cast<Function>(MallocF)->doesNotAlias(0))
+    cast<Function>(MallocF)->setDoesNotAlias(0);
   const PointerType *AllocPtrType = PointerType::getUnqual(AllocTy);
   CallInst *MCall = NULL;
   Value    *MCast = NULL;
@@ -533,8 +531,7 @@
 Value *CallInst::CreateMalloc(Instruction *InsertBefore, const Type *IntPtrTy,
                               const Type *AllocTy, Value *ArraySize,
                               const Twine &Name) {
-  return createMalloc(InsertBefore, NULL, IntPtrTy, AllocTy, 
-                      ArraySize, NULL, Name);
+  return createMalloc(InsertBefore, NULL, IntPtrTy, AllocTy, ArraySize, Name);
 }
 
 /// CreateMalloc - Generate the IR for a call to malloc:
@@ -547,9 +544,8 @@
 /// responsibility of the caller.
 Value *CallInst::CreateMalloc(BasicBlock *InsertAtEnd, const Type *IntPtrTy,
                               const Type *AllocTy, Value *ArraySize, 
-                              Function* MallocF, const Twine &Name) {
-  return createMalloc(NULL, InsertAtEnd, IntPtrTy, AllocTy,
-                      ArraySize, MallocF, Name);
+                              const Twine &Name) {
+  return createMalloc(NULL, InsertAtEnd, IntPtrTy, AllocTy, ArraySize, Name);
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/test/Analysis/PointerTracking/sizes.ll b/test/Analysis/PointerTracking/sizes.ll
index c6224e0..c0b0606 100644
--- a/test/Analysis/PointerTracking/sizes.ll
+++ b/test/Analysis/PointerTracking/sizes.ll
@@ -63,7 +63,7 @@
 define i32 @foo2(i32 %n) nounwind {
 entry:
 	%call = malloc i8, i32 %n		; <i8*> [#uses=1]
-; CHECK: %malloccall =
+; CHECK: %call =
 ; CHECK: ==> %n elements, %n bytes allocated
 	%call2 = tail call i8* @calloc(i64 2, i64 4) nounwind		; <i8*> [#uses=1]
 ; CHECK: %call2 =
diff --git a/test/Transforms/GlobalOpt/malloc-promote-2.ll b/test/Transforms/GlobalOpt/malloc-promote-2.ll
index d3d2252..0d03835 100644
--- a/test/Transforms/GlobalOpt/malloc-promote-2.ll
+++ b/test/Transforms/GlobalOpt/malloc-promote-2.ll
@@ -1,6 +1,4 @@
-; RUN: opt < %s -globalopt -globaldce -S | not grep malloc
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-target triple = "i686-apple-darwin8"
+; RUN: opt < %s -globalopt -S | not grep malloc
 
 @G = internal global i32* null          ; <i32**> [#uses=3]
 
diff --git a/test/Transforms/GlobalOpt/malloc-promote-3.ll b/test/Transforms/GlobalOpt/malloc-promote-3.ll
index a920b61..d4ee4e8 100644
--- a/test/Transforms/GlobalOpt/malloc-promote-3.ll
+++ b/test/Transforms/GlobalOpt/malloc-promote-3.ll
@@ -1,6 +1,4 @@
-; RUN: opt < %s -globalopt -globaldce -S | not grep malloc
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-target triple = "i686-apple-darwin8"
+; RUN: opt < %s -globalopt -S | not grep malloc
 
 @G = internal global i32* null          ; <i32**> [#uses=4]
 
diff --git a/test/Transforms/InstCombine/cast-malloc.ll b/test/Transforms/InstCombine/cast-malloc.ll
index 43a1489..3754032 100644
--- a/test/Transforms/InstCombine/cast-malloc.ll
+++ b/test/Transforms/InstCombine/cast-malloc.ll
@@ -1,6 +1,6 @@
 ; test that casted mallocs get converted to malloc of the right type
 ; RUN: opt < %s -instcombine -S | \
-; RUN:    grep bitcast | count 1
+; RUN:    not grep bitcast
 
 ; The target datalayout is important for this test case. We have to tell 
 ; instcombine that the ABI alignment for a long is 4-bytes, not 8, otherwise
diff --git a/test/Transforms/InstCombine/cast.ll b/test/Transforms/InstCombine/cast.ll
index 2722267..59d7cd0 100644
--- a/test/Transforms/InstCombine/cast.ll
+++ b/test/Transforms/InstCombine/cast.ll
@@ -79,9 +79,9 @@
 }
 
 define i32* @test12() {
-        %c = malloc [4 x i8]            ; <[4 x i8]*> [#uses=1]
-        %p = bitcast [4 x i8]* %c to i32*               ; <i32*> [#uses=1]
-        ret i32* %p
+        %p = malloc [4 x i8]            ; <[4 x i8]*> [#uses=1]
+        %c = bitcast [4 x i8]* %p to i32*               ; <i32*> [#uses=1]
+        ret i32* %c
 }
 define i8* @test13(i64 %A) {
         %c = getelementptr [0 x i8]* bitcast ([32832 x i8]* @inbuf to [0 x i8]*), i64 0, i64 %A             ; <i8*> [#uses=1]
diff --git a/test/Transforms/InstCombine/getelementptr.ll b/test/Transforms/InstCombine/getelementptr.ll
index 285e0ba..ffaa6af 100644
--- a/test/Transforms/InstCombine/getelementptr.ll
+++ b/test/Transforms/InstCombine/getelementptr.ll
@@ -58,7 +58,7 @@
         %B = getelementptr i32* %A, i64 2             
         ret i32* %B
 ; CHECK: @test6
-; CHECK: getelementptr i8* %malloccall, i64 8
+; CHECK: getelementptr [4 x i32]* %M, i64 0, i64 2
 }
 
 define i32* @test7(i32* %I, i64 %C, i64 %D) {
diff --git a/test/Transforms/InstCombine/malloc-free-delete.ll b/test/Transforms/InstCombine/malloc-free-delete.ll
index fd91e44..2ed5ec6 100644
--- a/test/Transforms/InstCombine/malloc-free-delete.ll
+++ b/test/Transforms/InstCombine/malloc-free-delete.ll
@@ -1,5 +1,5 @@
 ; RUN: opt < %s -instcombine -S | grep {ret i32 0}
-; RUN: opt < %s -instcombine -globaldce -S | not grep malloc
+; RUN: opt < %s -instcombine -S | not grep malloc
 ; PR1201
 define i32 @main(i32 %argc, i8** %argv) {
         %c_19 = alloca i8*              ; <i8**> [#uses=2]
diff --git a/test/Transforms/InstCombine/malloc2.ll b/test/Transforms/InstCombine/malloc2.ll
index cc1506b..102422e 100644
--- a/test/Transforms/InstCombine/malloc2.ll
+++ b/test/Transforms/InstCombine/malloc2.ll
@@ -1,4 +1,5 @@
 ; RUN: opt < %s -instcombine -S | grep {ret i32 0}
+; RUN: opt < %s -instcombine -S | not grep malloc
 ; PR1313
 
 define i32 @test1(i32 %argc, i8* %argv, i8* %envp) {