instcombine: Migrate memcmp optimizations

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167683 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 140 insertions, 91 deletions

diff --git a/lib/Transforms/Scalar/SimplifyLibCalls.cpp b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
index 3f39df9456..4d54e13c77 100644
--- a/lib/Transforms/Scalar/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Scalar/SimplifyLibCalls.cpp
@@ -105,54 +105,6 @@ static bool CallHasFloatingPointArgument(const CallInst *CI) {
 
 namespace {
 //===---------------------------------------===//
-// 'memcmp' Optimizations
-
-struct MemCmpOpt : public LibCallOptimization {
-  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
-    FunctionType *FT = Callee->getFunctionType();
-    if (FT->getNumParams() != 3 || !FT->getParamType(0)->isPointerTy() ||
-        !FT->getParamType(1)->isPointerTy() ||
-        !FT->getReturnType()->isIntegerTy(32))
-      return 0;
-
-    Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);
-
-    if (LHS == RHS)  // memcmp(s,s,x) -> 0
-      return Constant::getNullValue(CI->getType());
-
-    // Make sure we have a constant length.
-    ConstantInt *LenC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
-    if (!LenC) return 0;
-    uint64_t Len = LenC->getZExtValue();
-
-    if (Len == 0) // memcmp(s1,s2,0) -> 0
-      return Constant::getNullValue(CI->getType());
-
-    // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
-    if (Len == 1) {
-      Value *LHSV = B.CreateZExt(B.CreateLoad(CastToCStr(LHS, B), "lhsc"),
-                                 CI->getType(), "lhsv");
-      Value *RHSV = B.CreateZExt(B.CreateLoad(CastToCStr(RHS, B), "rhsc"),
-                                 CI->getType(), "rhsv");
-      return B.CreateSub(LHSV, RHSV, "chardiff");
-    }
-
-    // Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)
-    StringRef LHSStr, RHSStr;
-    if (getConstantStringInfo(LHS, LHSStr) &&
-        getConstantStringInfo(RHS, RHSStr)) {
-      // Make sure we're not reading out-of-bounds memory.
-      if (Len > LHSStr.size() || Len > RHSStr.size())
-        return 0;
-      uint64_t Ret = memcmp(LHSStr.data(), RHSStr.data(), Len);
-      return ConstantInt::get(CI->getType(), Ret);
-    }
-
-    return 0;
-  }
-};
-
-//===---------------------------------------===//
 // 'memcpy' Optimizations
 
 struct MemCpyOpt : public LibCallOptimization {
@@ -887,7 +839,7 @@ namespace {
 
     StringMap<LibCallOptimization*> Optimizations;
     // Memory LibCall Optimizations
-    MemCmpOpt MemCmp; MemCpyOpt MemCpy; MemMoveOpt MemMove; MemSetOpt MemSet;
+    MemCpyOpt MemCpy; MemMoveOpt MemMove; MemSetOpt MemSet;
     // Math Library Optimizations
     CosOpt Cos; PowOpt Pow; Exp2Opt Exp2;
     UnaryDoubleFPOpt UnaryDoubleFP, UnsafeUnaryDoubleFP;
@@ -954,7 +906,6 @@ void SimplifyLibCalls::AddOpt(LibFunc::Func F1, LibFunc::Func F2,
 /// we know.
 void SimplifyLibCalls::InitOptimizations() {
   // Memory LibCall Optimizations
-  Optimizations["memcmp"] = &MemCmp;
   AddOpt(LibFunc::memcpy, &MemCpy);
   Optimizations["memmove"] = &MemMove;
   AddOpt(LibFunc::memset, &MemSet);
diff --git a/lib/Transforms/Utils/SimplifyLibCalls.cpp b/lib/Transforms/Utils/SimplifyLibCalls.cpp
index b5ea46dc0d..805fcbe3cb 100644
--- a/lib/Transforms/Utils/SimplifyLibCalls.cpp
+++ b/lib/Transforms/Utils/SimplifyLibCalls.cpp
@@ -914,6 +914,51 @@ struct StrStrOpt : public LibCallOptimization {
   }
 };
 
+struct MemCmpOpt : public LibCallOptimization {
+  virtual Value *callOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
+    FunctionType *FT = Callee->getFunctionType();
+    if (FT->getNumParams() != 3 || !FT->getParamType(0)->isPointerTy() ||
+        !FT->getParamType(1)->isPointerTy() ||
+        !FT->getReturnType()->isIntegerTy(32))
+      return 0;
+
+    Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);
+
+    if (LHS == RHS)  // memcmp(s,s,x) -> 0
+      return Constant::getNullValue(CI->getType());
+
+    // Make sure we have a constant length.
+    ConstantInt *LenC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
+    if (!LenC) return 0;
+    uint64_t Len = LenC->getZExtValue();
+
+    if (Len == 0) // memcmp(s1,s2,0) -> 0
+      return Constant::getNullValue(CI->getType());
+
+    // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
+    if (Len == 1) {
+      Value *LHSV = B.CreateZExt(B.CreateLoad(CastToCStr(LHS, B), "lhsc"),
+                                 CI->getType(), "lhsv");
+      Value *RHSV = B.CreateZExt(B.CreateLoad(CastToCStr(RHS, B), "rhsc"),
+                                 CI->getType(), "rhsv");
+      return B.CreateSub(LHSV, RHSV, "chardiff");
+    }
+
+    // Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)
+    StringRef LHSStr, RHSStr;
+    if (getConstantStringInfo(LHS, LHSStr) &&
+        getConstantStringInfo(RHS, RHSStr)) {
+      // Make sure we're not reading out-of-bounds memory.
+      if (Len > LHSStr.size() || Len > RHSStr.size())
+        return 0;
+      uint64_t Ret = memcmp(LHSStr.data(), RHSStr.data(), Len);
+      return ConstantInt::get(CI->getType(), Ret);
+    }
+
+    return 0;
+  }
+};
+
 } // End anonymous namespace.
 
 namespace llvm {
@@ -932,7 +977,7 @@ class LibCallSimplifierImpl {
   StpCpyChkOpt StpCpyChk;
   StrNCpyChkOpt StrNCpyChk;
 
-  // String and memory library call optimizations.
+  // String library call optimizations.
   StrCatOpt StrCat;
   StrNCatOpt StrNCat;
   StrChrOpt StrChr;
@@ -949,6 +994,9 @@ class LibCallSimplifierImpl {
   StrCSpnOpt StrCSpn;
   StrStrOpt StrStr;
 
+  // Memory library call optimizations.
+  MemCmpOpt MemCmp;
+
   void initOptimizations();
   void addOpt(LibFunc::Func F, LibCallOptimization* Opt);
 public:
@@ -972,7 +1020,7 @@ void LibCallSimplifierImpl::initOptimizations() {
   Optimizations["__strncpy_chk"] = &StrNCpyChk;
   Optimizations["__stpncpy_chk"] = &StrNCpyChk;
 
-  // String and memory library call optimizations.
+  // String library call optimizations.
   addOpt(LibFunc::strcat, &StrCat);
   addOpt(LibFunc::strncat, &StrNCat);
   addOpt(LibFunc::strchr, &StrChr);
@@ -994,6 +1042,9 @@ void LibCallSimplifierImpl::initOptimizations() {
   addOpt(LibFunc::strspn, &StrSpn);
   addOpt(LibFunc::strcspn, &StrCSpn);
   addOpt(LibFunc::strstr, &StrStr);
+
+  // Memory library call optimizations.
+  addOpt(LibFunc::memcmp, &MemCmp);
 }
 
 Value *LibCallSimplifierImpl::optimizeCall(CallInst *CI) {
diff --git a/test/Transforms/InstCombine/memcmp-1.ll b/test/Transforms/InstCombine/memcmp-1.ll
new file mode 100644
index 0000000000..862fd3a85a
--- /dev/null
+++ b/test/Transforms/InstCombine/memcmp-1.ll
@@ -0,0 +1,63 @@
+; Test that the memcmp library call 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"
+
+@foo = constant [4 x i8] c"foo\00"
+@hel = constant [4 x i8] c"hel\00"
+@hello_u = constant [8 x i8] c"hello_u\00"
+
+declare i32 @memcmp(i8*, i8*, i32)
+
+; Check memcmp(mem, mem, size) -> 0.
+
+define i32 @test_simplify1(i8* %mem, i32 %size) {
+; CHECK: @test_simplify1
+  %ret = call i32 @memcmp(i8* %mem, i8* %mem, i32 %size)
+  ret i32 %ret
+; CHECK: ret i32 0
+}
+
+; Check memcmp(mem1, mem2, 0) -> 0.
+
+define i32 @test_simplify2(i8* %mem1, i8* %mem2) {
+; CHECK: @test_simplify2
+  %ret = call i32 @memcmp(i8* %mem1, i8* %mem2, i32 0)
+  ret i32 %ret
+; CHECK: ret i32 0
+}
+
+;; Check memcmp(mem1, mem2, 1) -> *(unsigned char*)mem1 - *(unsigned char*)mem2.
+
+define i32 @test_simplify3(i8* %mem1, i8* %mem2) {
+; CHECK: @test_simplify3
+  %ret = call i32 @memcmp(i8* %mem1, i8* %mem2, i32 1)
+; CHECK: [[LOAD1:%[a-z]+]] = load i8* %mem1, align 1
+; CHECK: [[ZEXT1:%[a-z]+]] = zext i8 [[LOAD1]] to i32
+; CHECK: [[LOAD2:%[a-z]+]] = load i8* %mem2, align 1
+; CHECK: [[ZEXT2:%[a-z]+]] = zext i8 [[LOAD2]] to i32
+; CHECK: [[RET:%[a-z]+]] = sub i32 [[ZEXT1]], [[ZEXT2]]
+  ret i32 %ret
+; CHECK: ret i32 [[RET]]
+}
+
+; Check memcmp(mem1, mem2, size) -> cnst, where all arguments are constants.
+
+define i32 @test_simplify4() {
+; CHECK: @test_simplify4
+  %mem1 = getelementptr [4 x i8]* @hel, i32 0, i32 0
+  %mem2 = getelementptr [8 x i8]* @hello_u, i32 0, i32 0
+  %ret = call i32 @memcmp(i8* %mem1, i8* %mem2, i32 3)
+  ret i32 %ret
+; CHECK: ret i32 0
+}
+
+define i32 @test_simplify5() {
+; CHECK: @test_simplify5
+  %mem1 = getelementptr [4 x i8]* @hel, i32 0, i32 0
+  %mem2 = getelementptr [4 x i8]* @foo, i32 0, i32 0
+  %ret = call i32 @memcmp(i8* %mem1, i8* %mem2, i32 3)
+  ret i32 %ret
+; CHECK: ret i32 2
+}
diff --git a/test/Transforms/InstCombine/memcmp-2.ll b/test/Transforms/InstCombine/memcmp-2.ll
new file mode 100644
index 0000000000..3796117bc2
--- /dev/null
+++ b/test/Transforms/InstCombine/memcmp-2.ll
@@ -0,0 +1,17 @@
+; Test that the memcmp library call 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"
+
+declare i32* @memcmp(i8*, i8*, i32)
+
+; Check that memcmp functions with the wrong prototype aren't simplified.
+
+define i32* @test_no_simplify1(i8* %mem, i32 %size) {
+; CHECK: @test_no_simplify1
+  %ret = call i32* @memcmp(i8* %mem, i8* %mem, i32 %size)
+; CHECK-NEXT: call i32* @memcmp
+  ret i32* %ret
+; CHECK-NEXT: ret i32* %ret
+}
diff --git a/test/Transforms/InstCombine/strncmp-1.ll b/test/Transforms/InstCombine/strncmp-1.ll
index 48b26d1a5f..187c2fa50e 100644
--- a/test/Transforms/InstCombine/strncmp-1.ll
+++ b/test/Transforms/InstCombine/strncmp-1.ll
@@ -67,12 +67,14 @@ define i32 @test5() {
 }
 
 ; strncmp(x,y,1) -> memcmp(x,y,1)
-; TODO: Once the memcmp simplifier gets moved into the instcombine pass
-; the following memcmp will be folded into two loads and a subtract.
 define i32 @test6(i8* %str1, i8* %str2) {
 ; CHECK: @test6
-; CHECK: call i32 @memcmp
-; CHECK: ret i32 %memcmp
+; CHECK: [[LOAD1:%[a-z]+]] = load i8* %str1, align 1
+; CHECK: [[ZEXT1:%[a-z]+]] = zext i8 [[LOAD1]] to i32
+; CHECK: [[LOAD2:%[a-z]+]] = load i8* %str2, align 1
+; CHECK: [[ZEXT2:%[a-z]+]] = zext i8 [[LOAD2]] to i32
+; CHECK: [[RET:%[a-z]+]] = sub i32 [[ZEXT1]], [[ZEXT2]]
+; CHECK: ret i32 [[RET]]
 
   %temp1 = call i32 @strncmp(i8* %str1, i8* %str2, i32 1)
   ret i32 %temp1
diff --git a/test/Transforms/SimplifyLibCalls/memcmp.ll b/test/Transforms/SimplifyLibCalls/memcmp.ll
deleted file mode 100644
index 6ca4dc97a1..0000000000
--- a/test/Transforms/SimplifyLibCalls/memcmp.ll
+++ /dev/null
@@ -1,35 +0,0 @@
-; Test that the memcmpOptimizer works correctly
-; RUN: opt < %s -simplify-libcalls -S | FileCheck %s
-
-@h = constant [2 x i8] c"h\00"		; <[2 x i8]*> [#uses=0]
-@hel = constant [4 x i8] c"hel\00"		; <[4 x i8]*> [#uses=0]
-@hello_u = constant [8 x i8] c"hello_u\00"		; <[8 x i8]*> [#uses=0]
-
-declare i32 @memcmp(i8*, i8*, i32)
-
-define void @test(i8* %P, i8* %Q, i32 %N, i32* %IP, i1* %BP) {
-	%A = call i32 @memcmp( i8* %P, i8* %P, i32 %N )		; <i32> [#uses=1]
-; CHECK-NOT: call {{.*}} memcmp
-; CHECK: store volatile
-	store volatile i32 %A, i32* %IP
-	%B = call i32 @memcmp( i8* %P, i8* %Q, i32 0 )		; <i32> [#uses=1]
-; CHECK-NOT: call {{.*}} memcmp
-; CHECK: store volatile
-	store volatile i32 %B, i32* %IP
-	%C = call i32 @memcmp( i8* %P, i8* %Q, i32 1 )		; <i32> [#uses=1]
-; CHECK: load
-; CHECK: zext
-; CHECK: load
-; CHECK: zext
-; CHECK: sub
-; CHECK: store volatile
-	store volatile i32 %C, i32* %IP
-  %F = call i32 @memcmp(i8* getelementptr ([4 x i8]* @hel, i32 0, i32 0),
-                        i8* getelementptr ([8 x i8]* @hello_u, i32 0, i32 0),
-                        i32 3)
-; CHECK-NOT: call {{.*}} memcmp
-; CHECK: store volatile
-  store volatile i32 %F, i32* %IP
-	ret void
-}
-