From: David Majnemer <david.majnemer@gmail.com>
Date: Wed, 27 Aug 2014 05:16:04 +0000 (+0000)
Subject: InstCombine: Optimize GEP's involving ptrtoint better
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=dd5456bd019fce7bd2da29e970a54633883e39b1;p=oota-llvm.git

InstCombine: Optimize GEP's involving ptrtoint better

We supported transforming:
(gep i8* X, -(ptrtoint Y))

to:
(inttoptr (sub (ptrtoint X), (ptrtoint Y)))

However, this only fired if 'X' had type i8*.  Generalize this to
support various types of different sizes.  This results in much better
CodeGen, especially for pointers to packed structs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216523 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
index f01dca7fab8..46789711bc8 100644
--- a/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1508,19 +1508,37 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
         GetElementPtrInst::Create(Src->getOperand(0), Indices, GEP.getName());
   }
 
-  // Canonicalize (gep i8* X, -(ptrtoint Y)) to (sub (ptrtoint X), (ptrtoint Y))
-  // The GEP pattern is emitted by the SCEV expander for certain kinds of
-  // pointer arithmetic.
-  if (DL && GEP.getNumIndices() == 1 &&
-      match(GEP.getOperand(1), m_Neg(m_PtrToInt(m_Value())))) {
+  if (DL && GEP.getNumIndices() == 1) {
     unsigned AS = GEP.getPointerAddressSpace();
-    if (GEP.getType() == Builder->getInt8PtrTy(AS) &&
-        GEP.getOperand(1)->getType()->getScalarSizeInBits() ==
+    if (GEP.getOperand(1)->getType()->getScalarSizeInBits() ==
         DL->getPointerSizeInBits(AS)) {
-      Operator *Index = cast<Operator>(GEP.getOperand(1));
-      Value *PtrToInt = Builder->CreatePtrToInt(PtrOp, Index->getType());
-      Value *NewSub = Builder->CreateSub(PtrToInt, Index->getOperand(1));
-      return CastInst::Create(Instruction::IntToPtr, NewSub, GEP.getType());
+      Type *PtrTy = GEP.getPointerOperandType();
+      Type *Ty = PtrTy->getPointerElementType();
+      uint64_t TyAllocSize = DL->getTypeAllocSize(Ty);
+
+      // Canonicalize (gep i8* X, -(ptrtoint Y)) to (sub (ptrtoint X), (ptrtoint Y))
+      // The GEP pattern is emitted by the SCEV expander for certain kinds of
+      // pointer arithmetic.
+      uint64_t C;
+      Value *NegPtrToInt = nullptr;
+      if (TyAllocSize == 1) {
+          NegPtrToInt = GEP.getOperand(1);
+      } else if (match(GEP.getOperand(1),
+                       m_AShr(m_Value(NegPtrToInt), m_ConstantInt(C)))) {
+        if (TyAllocSize != 1ULL << C)
+          NegPtrToInt = nullptr;
+      } else if (match(GEP.getOperand(1),
+                       m_SDiv(m_Value(NegPtrToInt), m_ConstantInt(C)))) {
+        if (TyAllocSize != C)
+          NegPtrToInt = nullptr;
+      }
+
+      if (NegPtrToInt && match(NegPtrToInt, m_Neg(m_PtrToInt(m_Value())))) {
+        Operator *Index = cast<Operator>(NegPtrToInt);
+        Value *PtrToInt = Builder->CreatePtrToInt(PtrOp, Index->getType());
+        Value *NewSub = Builder->CreateSub(PtrToInt, Index->getOperand(1));
+        return CastInst::Create(Instruction::IntToPtr, NewSub, GEP.getType());
+      }
     }
   }
 
diff --git a/test/Transforms/InstCombine/getelementptr.ll b/test/Transforms/InstCombine/getelementptr.ll
index da91ada0805..71546ae3328 100644
--- a/test/Transforms/InstCombine/getelementptr.ll
+++ b/test/Transforms/InstCombine/getelementptr.ll
@@ -6,6 +6,7 @@ target datalayout = "e-p:64:64-p1:16:16-p2:32:32:32-p3:64:64:64"
 %pair = type { i32, i32 }
 %struct.B = type { double }
 %struct.A = type { %struct.B, i32, i32 }
+%struct.C = type { [7 x i8] }
 
 
 @Global = constant [10 x i8] c"helloworld"
@@ -813,6 +814,50 @@ define i16 @test41([3 x i32] addrspace(1)* %array) {
 ; CHECK-NEXT: ret i16 8
 }
 
+define i8* @test42(i8* %c1, i8* %c2) {
+  %ptrtoint = ptrtoint i8* %c1 to i64
+  %sub = sub i64 0, %ptrtoint
+  %gep = getelementptr inbounds i8* %c2, i64 %sub
+  ret i8* %gep
+
+; CHECK-LABEL: @test42(
+; CHECK-NEXT:  [[PTRTOINT1:%.*]] = ptrtoint i8* %c1 to i64
+; CHECK-NEXT:  [[PTRTOINT2:%.*]] = ptrtoint i8* %c2 to i64
+; CHECK-NEXT:  [[SUB:%.*]] = sub i64 [[PTRTOINT2]], [[PTRTOINT1]]
+; CHECK-NEXT:  [[INTTOPTR:%.*]] = inttoptr i64 [[SUB]] to i8*
+; CHECK-NEXT:  ret i8* [[INTTOPTR]]
+}
+
+define i16* @test43(i16* %c1, i16* %c2) {
+  %ptrtoint = ptrtoint i16* %c1 to i64
+  %sub = sub i64 0, %ptrtoint
+  %shr = ashr i64 %sub, 1
+  %gep = getelementptr inbounds i16* %c2, i64 %shr
+  ret i16* %gep
+
+; CHECK-LABEL: @test43(
+; CHECK-NEXT:  [[PTRTOINT1:%.*]] = ptrtoint i16* %c1 to i64
+; CHECK-NEXT:  [[PTRTOINT2:%.*]] = ptrtoint i16* %c2 to i64
+; CHECK-NEXT:  [[SUB:%.*]] = sub i64 [[PTRTOINT2]], [[PTRTOINT1]]
+; CHECK-NEXT:  [[INTTOPTR:%.*]] = inttoptr i64 [[SUB]] to i16*
+; CHECK-NEXT:  ret i16* [[INTTOPTR]]
+}
+
+define %struct.C* @test44(%struct.C* %c1, %struct.C* %c2) {
+  %ptrtoint = ptrtoint %struct.C* %c1 to i64
+  %sub = sub i64 0, %ptrtoint
+  %shr = sdiv i64 %sub, 7
+  %gep = getelementptr inbounds %struct.C* %c2, i64 %shr
+  ret %struct.C* %gep
+
+; CHECK-LABEL: @test44(
+; CHECK-NEXT:  [[PTRTOINT1:%.*]] = ptrtoint %struct.C* %c1 to i64
+; CHECK-NEXT:  [[PTRTOINT2:%.*]] = ptrtoint %struct.C* %c2 to i64
+; CHECK-NEXT:  [[SUB:%.*]] = sub i64 [[PTRTOINT2]], [[PTRTOINT1]]
+; CHECK-NEXT:  [[INTTOPTR:%.*]] = inttoptr i64 [[SUB]] to %struct.C*
+; CHECK-NEXT:  ret %struct.C* [[INTTOPTR]]
+}
+
 define i32 addrspace(1)* @ascast_0_gep(i32* %p) nounwind {
 ; CHECK-LABEL: @ascast_0_gep(
 ; CHECK-NOT: getelementptr