instcombine: Migrate strcat and strncat optimizations

This patch migrates the strcat and strncat optimizations from the
simplify-libcalls pass into the instcombine library call simplifier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165874 91177308-0d34-0410-b5e6-96231b3b80d8

10 files changed, 311 insertions, 165 deletions

diff --git a/lib/Transforms/Scalar/SimplifyLibCalls.cpp b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
index 73f53b7cec..22a01fc968 100644
--- a/lib/Transforms/Scalar/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
@@ -133,105 +133,7 @@ static bool IsOnlyUsedInEqualityComparison(Value *V, Value *With) {
 // String and Memory LibCall Optimizations
 //===----------------------------------------------------------------------===//
 
-//===---------------------------------------===//
-// 'strcat' Optimizations
 namespace {
-struct StrCatOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strcat" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 2 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        FT->getParamType(1) != FT->getReturnType())
-      return 0;
-
-    // Extract some information from the instruction
-    Value *Dst = CI->getArgOperand(0);
-    Value *Src = CI->getArgOperand(1);
-
-    // See if we can get the length of the input string.
-    uint64_t Len = GetStringLength(Src);
-    if (Len == 0) return 0;
-    --Len;  // Unbias length.
-
-    // Handle the simple, do-nothing case: strcat(x, "") -> x
-    if (Len == 0)
-      return Dst;
-
-    // These optimizations require DataLayout.
-    if (!TD) return 0;
-
-    return EmitStrLenMemCpy(Src, Dst, Len, B);
-  }
-
-  Value *EmitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len, IRBuilder<> &B) {
-    // We need to find the end of the destination string.  That's where the
-    // memory is to be moved to. We just generate a call to strlen.
-    Value *DstLen = EmitStrLen(Dst, B, TD, TLI);
-    if (!DstLen)
-      return 0;
-
-    // Now that we have the destination's length, we must index into the
-    // destination's pointer to get the actual memcpy destination (end of
-    // the string .. we're concatenating).
-    Value *CpyDst = B.CreateGEP(Dst, DstLen, "endptr");
-
-    // We have enough information to now generate the memcpy call to do the
-    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
-    B.CreateMemCpy(CpyDst, Src,
-                   ConstantInt::get(TD->getIntPtrType(*Context), Len + 1), 1);
-    return Dst;
-  }
-};
-
-//===---------------------------------------===//
-// 'strncat' Optimizations
-
-struct StrNCatOpt : public StrCatOpt {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    // Verify the "strncat" function prototype.
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 ||
-        FT->getReturnType() != B.getInt8PtrTy() ||
-        FT->getParamType(0) != FT->getReturnType() ||
-        FT->getParamType(1) != FT->getReturnType() ||
-        !FT->getParamType(2)->isIntegerTy())
-      return 0;
-
-    // Extract some information from the instruction
-    Value *Dst = CI->getArgOperand(0);
-    Value *Src = CI->getArgOperand(1);
-    uint64_t Len;
-
-    // We don't do anything if length is not constant
-    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
-      Len = LengthArg->getZExtValue();
-    else
-      return 0;
-
-    // See if we can get the length of the input string.
-    uint64_t SrcLen = GetStringLength(Src);
-    if (SrcLen == 0) return 0;
-    --SrcLen;  // Unbias length.
-
-    // Handle the simple, do-nothing cases:
-    // strncat(x, "", c) -> x
-    // strncat(x,  c, 0) -> x
-    if (SrcLen == 0 || Len == 0) return Dst;
-
-    // These optimizations require DataLayout.
-    if (!TD) return 0;
-
-    // We don't optimize this case
-    if (Len < SrcLen) return 0;
-
-    // strncat(x, s, c) -> strcat(x, s)
-    // s is constant so the strcat can be optimized further
-    return EmitStrLenMemCpy(Src, Dst, SrcLen, B);
-  }
-};
-
 //===---------------------------------------===//
 // 'strchr' Optimizations
 
@@ -1564,7 +1466,7 @@ namespace {
 
     StringMap<LibCallOptimization*> Optimizations;
     // String and Memory LibCall Optimizations
-    StrCatOpt StrCat; StrNCatOpt StrNCat; StrChrOpt StrChr; StrRChrOpt StrRChr;
+    StrChrOpt StrChr; StrRChrOpt StrRChr;
     StrCmpOpt StrCmp; StrNCmpOpt StrNCmp;
     StrCpyOpt StrCpy; StrCpyOpt StrCpyChk;
     StpCpyOpt StpCpy; StpCpyOpt StpCpyChk;
@@ -1639,8 +1541,6 @@ void SimplifyLibCalls::AddOpt(LibFunc::Func F1, LibFunc::Func F2,
 /// we know.
 void SimplifyLibCalls::InitOptimizations() {
   // String and Memory LibCall Optimizations
-  Optimizations["strcat"] = &StrCat;
-  Optimizations["strncat"] = &StrNCat;
   Optimizations["strchr"] = &StrChr;
   Optimizations["strrchr"] = &StrRChr;
   Optimizations["strcmp"] = &StrCmp;
diff --git a/lib/Transforms/Utils/SimplifyLibCalls.cpp b/lib/Transforms/Utils/SimplifyLibCalls.cpp
index 7e4c1e70e2..2c8fb8b896 100644
--- a/lib/Transforms/Utils/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Utils/SimplifyLibCalls.cpp
@@ -222,6 +222,104 @@ struct StrNCpyChkOpt : public InstFortifiedLibCallOptimization {
   }
 };
 
+//===----------------------------------------------------------------------===//
+// String and Memory Library Call Optimizations
+//===----------------------------------------------------------------------===//
+
+struct StrCatOpt : public LibCallOptimization {
+  virtual Value *callOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    // Verify the "strcat" function prototype.
+    FunctionType *FT = Callee->getFunctionType();
+    if (FT->getNumParams() != 2 ||
+        FT->getReturnType() != B.getInt8PtrTy() ||
+        FT->getParamType(0) != FT->getReturnType() ||
+        FT->getParamType(1) != FT->getReturnType())
+      return 0;
+
+    // Extract some information from the instruction
+    Value *Dst = CI->getArgOperand(0);
+    Value *Src = CI->getArgOperand(1);
+
+    // See if we can get the length of the input string.
+    uint64_t Len = GetStringLength(Src);
+    if (Len == 0) return 0;
+    --Len;  // Unbias length.
+
+    // Handle the simple, do-nothing case: strcat(x, "") -> x
+    if (Len == 0)
+      return Dst;
+
+    // These optimizations require DataLayout.
+    if (!TD) return 0;
+
+    return emitStrLenMemCpy(Src, Dst, Len, B);
+  }
+
+  Value *emitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len,
+                          IRBuilder<> &B) {
+    // We need to find the end of the destination string.  That's where the
+    // memory is to be moved to. We just generate a call to strlen.
+    Value *DstLen = EmitStrLen(Dst, B, TD, TLI);
+    if (!DstLen)
+      return 0;
+
+    // Now that we have the destination's length, we must index into the
+    // destination's pointer to get the actual memcpy destination (end of
+    // the string .. we're concatenating).
+    Value *CpyDst = B.CreateGEP(Dst, DstLen, "endptr");
+
+    // We have enough information to now generate the memcpy call to do the
+    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
+    B.CreateMemCpy(CpyDst, Src,
+                   ConstantInt::get(TD->getIntPtrType(*Context), Len + 1), 1);
+    return Dst;
+  }
+};
+
+struct StrNCatOpt : public StrCatOpt {
+  virtual Value *callOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    // Verify the "strncat" function prototype.
+    FunctionType *FT = Callee->getFunctionType();
+    if (FT->getNumParams() != 3 ||
+        FT->getReturnType() != B.getInt8PtrTy() ||
+        FT->getParamType(0) != FT->getReturnType() ||
+        FT->getParamType(1) != FT->getReturnType() ||
+        !FT->getParamType(2)->isIntegerTy())
+      return 0;
+
+    // Extract some information from the instruction
+    Value *Dst = CI->getArgOperand(0);
+    Value *Src = CI->getArgOperand(1);
+    uint64_t Len;
+
+    // We don't do anything if length is not constant
+    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
+      Len = LengthArg->getZExtValue();
+    else
+      return 0;
+
+    // See if we can get the length of the input string.
+    uint64_t SrcLen = GetStringLength(Src);
+    if (SrcLen == 0) return 0;
+    --SrcLen;  // Unbias length.
+
+    // Handle the simple, do-nothing cases:
+    // strncat(x, "", c) -> x
+    // strncat(x,  c, 0) -> x
+    if (SrcLen == 0 || Len == 0) return Dst;
+
+    // These optimizations require DataLayout.
+    if (!TD) return 0;
+
+    // We don't optimize this case
+    if (Len < SrcLen) return 0;
+
+    // strncat(x, s, c) -> strcat(x, s)
+    // s is constant so the strcat can be optimized further
+    return emitStrLenMemCpy(Src, Dst, SrcLen, B);
+  }
+};
+
 } // End anonymous namespace.
 
 namespace llvm {
@@ -239,6 +337,10 @@ class LibCallSimplifierImpl {
   StrCpyChkOpt StrCpyChk;
   StrNCpyChkOpt StrNCpyChk;
 
+  // String and memory library call optimizations.
+  StrCatOpt StrCat;
+  StrNCatOpt StrNCat;
+
   void initOptimizations();
 public:
   LibCallSimplifierImpl(const DataLayout *TD, const TargetLibraryInfo *TLI) {
@@ -258,6 +360,10 @@ void LibCallSimplifierImpl::initOptimizations() {
   Optimizations["__stpcpy_chk"] = &StrCpyChk;
   Optimizations["__strncpy_chk"] = &StrNCpyChk;
   Optimizations["__stpncpy_chk"] = &StrNCpyChk;
+
+  // String and memory library call optimizations.
+  Optimizations["strcat"] = &StrCat;
+  Optimizations["strncat"] = &StrNCat;
 }
 
 Value *LibCallSimplifierImpl::optimizeCall(CallInst *CI) {
diff --git a/test/Transforms/InstCombine/strcat-1.ll b/test/Transforms/InstCombine/strcat-1.ll
new file mode 100644
index 0000000000..3c05d6b06f
--- /dev/null
+++ b/test/Transforms/InstCombine/strcat-1.ll
@@ -0,0 +1,38 @@
+; Test that the strcat libcall simplifier works correctly per the
+; bug found in PR3661.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@null = constant [1 x i8] zeroinitializer
+@null_hello = constant [7 x i8] c"\00hello\00"
+
+declare i8* @strcat(i8*, i8*)
+declare i32 @puts(i8*)
+
+define i32 @main() {
+; CHECK: @main
+; CHECK-NOT: call i8* @strcat
+; CHECK: call i32 @puts
+
+  %target = alloca [1024 x i8]
+  %arg1 = getelementptr [1024 x i8]* %target, i32 0, i32 0
+  store i8 0, i8* %arg1
+
+  ; rslt1 = strcat(target, "hello\00")
+  %arg2 = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  %rslt1 = call i8* @strcat(i8* %arg1, i8* %arg2)
+
+  ; rslt2 = strcat(rslt1, "\00")
+  %arg3 = getelementptr [1 x i8]* @null, i32 0, i32 0
+  %rslt2 = call i8* @strcat(i8* %rslt1, i8* %arg3)
+
+  ; rslt3 = strcat(rslt2, "\00hello\00")
+  %arg4 = getelementptr [7 x i8]* @null_hello, i32 0, i32 0
+  %rslt3 = call i8* @strcat(i8* %rslt2, i8* %arg4)
+
+  call i32 @puts( i8* %rslt3 )
+  ret i32 0
+}
diff --git a/test/Transforms/InstCombine/strcat-2.ll b/test/Transforms/InstCombine/strcat-2.ll
new file mode 100644
index 0000000000..379ee74953
--- /dev/null
+++ b/test/Transforms/InstCombine/strcat-2.ll
@@ -0,0 +1,32 @@
+; Test that the strcat libcall simplifier works correctly.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@empty = constant [1 x i8] c"\00"
+@a = common global [32 x i8] zeroinitializer, align 1
+
+declare i8* @strcat(i8*, i8*)
+
+define void @test_simplify1() {
+; CHECK: @test_simplify1
+; CHECK-NOT: call i8* @strcat
+; CHECK: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i8* @strcat(i8* %dst, i8* %src)
+  ret void
+}
+
+define void @test_simplify2() {
+; CHECK: @test_simplify2
+; CHECK-NEXT: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [1 x i8]* @empty, i32 0, i32 0
+  call i8* @strcat(i8* %dst, i8* %src)
+  ret void
+}
diff --git a/test/Transforms/InstCombine/strcat-3.ll b/test/Transforms/InstCombine/strcat-3.ll
new file mode 100644
index 0000000000..15aff2f1aa
--- /dev/null
+++ b/test/Transforms/InstCombine/strcat-3.ll
@@ -0,0 +1,22 @@
+; Test that the strcat libcall simplifier works correctly.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@empty = constant [1 x i8] c"\00"
+@a = common global [32 x i8] zeroinitializer, align 1
+
+declare i16* @strcat(i8*, i8*)
+
+define void @test_nosimplify1() {
+; CHECK: @test_nosimplify1
+; CHECK: call i16* @strcat
+; CHECK: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i16* @strcat(i8* %dst, i8* %src)
+  ret void
+}
diff --git a/test/Transforms/InstCombine/strncat-1.ll b/test/Transforms/InstCombine/strncat-1.ll
new file mode 100644
index 0000000000..ad2a18b146
--- /dev/null
+++ b/test/Transforms/InstCombine/strncat-1.ll
@@ -0,0 +1,37 @@
+; Test that the strncat libcall simplifier works correctly.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@null = constant [1 x i8] zeroinitializer
+@null_hello = constant [7 x i8] c"\00hello\00"
+
+declare i8* @strncat(i8*, i8*, i32)
+declare i32 @puts(i8*)
+
+define i32 @main() {
+; CHECK: @main
+; CHECK-NOT: call i8* @strncat
+; CHECK: call i32 @puts
+
+  %target = alloca [1024 x i8]
+  %arg1 = getelementptr [1024 x i8]* %target, i32 0, i32 0
+  store i8 0, i8* %arg1
+
+  ; rslt1 = strncat(target, "hello\00")
+  %arg2 = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  %rslt1 = call i8* @strncat(i8* %arg1, i8* %arg2, i32 6)
+
+  ; rslt2 = strncat(rslt1, "\00")
+  %arg3 = getelementptr [1 x i8]* @null, i32 0, i32 0
+  %rslt2 = call i8* @strncat(i8* %rslt1, i8* %arg3, i32 42)
+
+  ; rslt3 = strncat(rslt2, "\00hello\00")
+  %arg4 = getelementptr [7 x i8]* @null_hello, i32 0, i32 0
+  %rslt3 = call i8* @strncat(i8* %rslt2, i8* %arg4, i32 42)
+
+  call i32 @puts(i8* %rslt3)
+  ret i32 0
+}
diff --git a/test/Transforms/InstCombine/strncat-2.ll b/test/Transforms/InstCombine/strncat-2.ll
new file mode 100644
index 0000000000..c56deacd39
--- /dev/null
+++ b/test/Transforms/InstCombine/strncat-2.ll
@@ -0,0 +1,53 @@
+; Test that the strncat libcall simplifier works correctly.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@empty = constant [1 x i8] c"\00"
+@a = common global [32 x i8] zeroinitializer, align 1
+
+declare i8* @strncat(i8*, i8*, i32)
+
+define void @test_simplify1() {
+; CHECK: @test_simplify1
+; CHECK-NOT: call i8* @strncat
+; CHECK: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i8* @strncat(i8* %dst, i8* %src, i32 13)
+  ret void
+}
+
+define void @test_simplify2() {
+; CHECK: @test_simplify2
+; CHECK-NEXT: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [1 x i8]* @empty, i32 0, i32 0
+  call i8* @strncat(i8* %dst, i8* %src, i32 13)
+  ret void
+}
+
+define void @test_simplify3() {
+; CHECK: @test_simplify3
+; CHECK-NEXT: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i8* @strncat(i8* %dst, i8* %src, i32 0)
+  ret void
+}
+
+define void @test_nosimplify1() {
+; CHECK: @test_nosimplify1
+; CHECK: call i8* @strncat
+; CHECK: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i8* @strncat(i8* %dst, i8* %src, i32 1)
+  ret void
+}
diff --git a/test/Transforms/InstCombine/strncat-3.ll b/test/Transforms/InstCombine/strncat-3.ll
new file mode 100644
index 0000000000..3cd7971687
--- /dev/null
+++ b/test/Transforms/InstCombine/strncat-3.ll
@@ -0,0 +1,22 @@
+; Test that the strncat libcall simplifier works correctly.
+;
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+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"
+
+@hello = constant [6 x i8] c"hello\00"
+@empty = constant [1 x i8] c"\00"
+@a = common global [32 x i8] zeroinitializer, align 1
+
+declare i16* @strncat(i8*, i8*, i32)
+
+define void @test_nosimplify1() {
+; CHECK: @test_nosimplify1
+; CHECK: call i16* @strncat
+; CHECK: ret void
+
+  %dst = getelementptr [32 x i8]* @a, i32 0, i32 0
+  %src = getelementptr [6 x i8]* @hello, i32 0, i32 0
+  call i16* @strncat(i8* %dst, i8* %src, i32 13)
+  ret void
+}
diff --git a/test/Transforms/SimplifyLibCalls/StrCat.ll b/test/Transforms/SimplifyLibCalls/StrCat.ll
deleted file mode 100644
index 3ea691a3cf..0000000000
--- a/test/Transforms/SimplifyLibCalls/StrCat.ll
+++ /dev/null
@@ -1,33 +0,0 @@
-; Test that the StrCatOptimizer works correctly
-; PR3661
-; RUN: opt < %s -simplify-libcalls -S | \
-; RUN:   not grep "call.*strcat"
-; RUN: opt < %s -simplify-libcalls -S | \
-; RUN:   grep "puts.*%arg1"
-
-; This transformation requires the pointer size, as it assumes that size_t is
-; the size of a pointer.
-target datalayout = "-p:64:64:64"
-
-@hello = constant [6 x i8] c"hello\00"		; <[6 x i8]*> [#uses=1]
-@null = constant [1 x i8] zeroinitializer		; <[1 x i8]*> [#uses=1]
-@null_hello = constant [7 x i8] c"\00hello\00"		; <[7 x i8]*> [#uses=1]
-
-declare i8* @strcat(i8*, i8*)
-
-declare i32 @puts(i8*)
-
-define i32 @main() {
-	%target = alloca [1024 x i8]		; <[1024 x i8]*> [#uses=1]
-	%arg1 = getelementptr [1024 x i8]* %target, i32 0, i32 0		; <i8*> [#uses=2]
-	store i8 0, i8* %arg1
-	%arg2 = getelementptr [6 x i8]* @hello, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt1 = call i8* @strcat( i8* %arg1, i8* %arg2 )		; <i8*> [#uses=1]
-	%arg3 = getelementptr [1 x i8]* @null, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt2 = call i8* @strcat( i8* %rslt1, i8* %arg3 )		; <i8*> [#uses=1]
-	%arg4 = getelementptr [7 x i8]* @null_hello, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt3 = call i8* @strcat( i8* %rslt2, i8* %arg4 )		; <i8*> [#uses=1]
-	call i32 @puts( i8* %rslt3 )		; <i32>:1 [#uses=0]
-	ret i32 0
-}
-
diff --git a/test/Transforms/SimplifyLibCalls/StrNCat.ll b/test/Transforms/SimplifyLibCalls/StrNCat.ll
deleted file mode 100644
index 073792b96a..0000000000
--- a/test/Transforms/SimplifyLibCalls/StrNCat.ll
+++ /dev/null
@@ -1,31 +0,0 @@
-; Test that the StrNCatOptimizer works correctly
-; RUN: opt < %s -simplify-libcalls -S | \
-; RUN:   not grep "call.*strncat"
-; RUN: opt < %s -simplify-libcalls -S | \
-; RUN:   grep "puts.*%arg1"
-
-; This transformation requires the pointer size, as it assumes that size_t is
-; the size of a pointer.
-target datalayout = "-p:64:64:64"
-
-@hello = constant [6 x i8] c"hello\00"		; <[6 x i8]*> [#uses=1]
-@null = constant [1 x i8] zeroinitializer		; <[1 x i8]*> [#uses=1]
-@null_hello = constant [7 x i8] c"\00hello\00"		; <[7 x i8]*> [#uses=1]
-
-declare i8* @strncat(i8*, i8*, i32)
-
-declare i32 @puts(i8*)
-
-define i32 @main() {
-	%target = alloca [1024 x i8]		; <[1024 x i8]*> [#uses=1]
-	%arg1 = getelementptr [1024 x i8]* %target, i32 0, i32 0		; <i8*> [#uses=2]
-	store i8 0, i8* %arg1
-	%arg2 = getelementptr [6 x i8]* @hello, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt1 = call i8* @strncat( i8* %arg1, i8* %arg2, i32 6 )		; <i8*> [#uses=1]
-	%arg3 = getelementptr [1 x i8]* @null, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt2 = call i8* @strncat( i8* %rslt1, i8* %arg3, i32 42 )		; <i8*> [#uses=1]
-	%arg4 = getelementptr [7 x i8]* @null_hello, i32 0, i32 0		; <i8*> [#uses=1]
-	%rslt3 = call i8* @strncat( i8* %rslt2, i8* %arg4, i32 42 )		; <i8*> [#uses=1]
-	call i32 @puts( i8* %rslt3 )		; <i32>:1 [#uses=0]
-	ret i32 0
-}