Index: llvm/trunk/include/llvm/Analysis/TargetTransformInfoImpl.h
===================================================================
--- llvm/trunk/include/llvm/Analysis/TargetTransformInfoImpl.h
+++ llvm/trunk/include/llvm/Analysis/TargetTransformInfoImpl.h
@@ -483,7 +483,9 @@
     int64_t Scale = 0;
 
     auto GTI = gep_type_begin(PointeeType, Operands);
+    Type *TargetType;
     for (auto I = Operands.begin(); I != Operands.end(); ++I, ++GTI) {
+      TargetType = GTI.getIndexedType();
       // We assume that the cost of Scalar GEP with constant index and the
       // cost of Vector GEP with splat constant index are the same.
       const ConstantInt *ConstIdx = dyn_cast<ConstantInt>(*I);
@@ -513,11 +515,9 @@
     unsigned AS =
         (Ptr == nullptr ? 0 : Ptr->getType()->getPointerAddressSpace());
     if (static_cast<T *>(this)->isLegalAddressingMode(
-            PointerType::get(Type::getInt8Ty(PointeeType->getContext()), AS),
-            const_cast<GlobalValue *>(BaseGV),
-            BaseOffset, HasBaseReg, Scale, AS)) {
+            TargetType, const_cast<GlobalValue *>(BaseGV), BaseOffset,
+            HasBaseReg, Scale, AS))
       return TTI::TCC_Free;
-    }
     return TTI::TCC_Basic;
   }
 
Index: llvm/trunk/test/Analysis/CostModel/AArch64/gep.ll
===================================================================
--- llvm/trunk/test/Analysis/CostModel/AArch64/gep.ll
+++ llvm/trunk/test/Analysis/CostModel/AArch64/gep.ll
@@ -0,0 +1,196 @@
+; RUN: opt -cost-model -analyze -mtriple=aarch64--linux-gnu -mcpu=kryo < %s | FileCheck %s
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64--linux-gnu"
+
+define i8 @test1(i8* %p, i32 %i) {
+; CHECK-LABEL: test1
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 0
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test2(i16* %p, i32 %i) {
+; CHECK-LABEL: test2
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 0
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test3(i32* %p, i32 %i) {
+; CHECK-LABEL: test3
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 0
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test4(i64* %p, i32 %i) {
+; CHECK-LABEL: test4
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 0
+  %v = load i64, i64* %a
+  ret i64 %v
+}
+
+define i8 @test5(i8* %p, i32 %i) {
+; CHECK-LABEL: test5
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 1024
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test6(i16* %p, i32 %i) {
+; CHECK-LABEL: test6
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 1024
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test7(i32* %p, i32 %i) {
+; CHECK-LABEL: test7
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 1024
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test8(i64* %p, i32 %i) {
+; CHECK-LABEL: test8
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 1024
+  %v = load i64, i64* %a
+  ret i64 %v
+}
+
+define i8 @test9(i8* %p, i32 %i) {
+; CHECK-LABEL: test9
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 4096
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test10(i16* %p, i32 %i) {
+; CHECK-LABEL: test10
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 4096
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test11(i32* %p, i32 %i) {
+; CHECK-LABEL: test11
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 4096
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test12(i64* %p, i32 %i) {
+; CHECK-LABEL: test12
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 4096
+  %v = load i64, i64* %a
+  ret i64 %v
+}
+
+define i8 @test13(i8* %p, i32 %i) {
+; CHECK-LABEL: test13
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 -64
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test14(i16* %p, i32 %i) {
+; CHECK-LABEL: test14
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 -64
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test15(i32* %p, i32 %i) {
+; CHECK-LABEL: test15
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 -64
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test16(i64* %p, i32 %i) {
+; CHECK-LABEL: test16
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 -64
+  %v = load i64, i64* %a
+  ret i64 %v
+}
+
+define i8 @test17(i8* %p, i32 %i) {
+; CHECK-LABEL: test17
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 -1024
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test18(i16* %p, i32 %i) {
+; CHECK-LABEL: test18
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 -1024
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test19(i32* %p, i32 %i) {
+; CHECK-LABEL: test19
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 -1024
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test20(i64* %p, i32 %i) {
+; CHECK-LABEL: test20
+; CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 -1024
+  %v = load i64, i64* %a
+  ret i64 %v
+}
+
+define i8 @test21(i8* %p, i32 %i) {
+; CHECK-LABEL: test21
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i8, i8*
+  %a = getelementptr inbounds i8, i8* %p, i32 %i
+  %v = load i8, i8* %a
+  ret i8 %v
+}
+
+define i16 @test22(i16* %p, i32 %i) {
+; CHECK-LABEL: test22
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i16, i16*
+  %a = getelementptr inbounds i16, i16* %p, i32 %i
+  %v = load i16, i16* %a
+  ret i16 %v
+}
+
+define i32 @test23(i32* %p, i32 %i) {
+; CHECK-LABEL: test23
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i32, i32*
+  %a = getelementptr inbounds i32, i32* %p, i32 %i
+  %v = load i32, i32* %a
+  ret i32 %v
+}
+
+define i64 @test24(i64* %p, i32 %i) {
+; CHECK-LABEL: test24
+; CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i64, i64*
+  %a = getelementptr inbounds i64, i64* %p, i32 %i
+  %v = load i64, i64* %a
+  ret i64 %v
+}
Index: llvm/trunk/test/Analysis/CostModel/ARM/gep.ll
===================================================================
--- llvm/trunk/test/Analysis/CostModel/ARM/gep.ll
+++ llvm/trunk/test/Analysis/CostModel/ARM/gep.ll
@@ -44,17 +44,17 @@
   %b4 = getelementptr inbounds float, float* undef, i32 1024
 ;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds double, double*
   %b5 = getelementptr inbounds double, double* undef, i32 1024
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i8>, <4 x i8>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i8>, <4 x i8>*
   %b7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 1
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i16>, <4 x i16>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i16>, <4 x i16>*
   %b8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 1
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i32>, <4 x i32>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i32>, <4 x i32>*
   %b9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 1
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i64>, <4 x i64>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i64>, <4 x i64>*
   %b10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 1
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x float>, <4 x float>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x float>, <4 x float>*
   %b11 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 1
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x double>, <4 x double>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x double>, <4 x double>*
   %b12 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 1
 
 ;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i8, i8*
@@ -63,15 +63,15 @@
   %c1 = getelementptr inbounds i16, i16* undef, i32 %i
 ;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i32, i32*
   %c2 = getelementptr inbounds i32, i32* undef, i32 %i
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds i64, i64*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds i64, i64*
   %c3 = getelementptr inbounds i64, i64* undef, i32 %i
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds float, float*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds float, float*
   %c4 = getelementptr inbounds float, float* undef, i32 %i
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds double, double*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds double, double*
   %c5 = getelementptr inbounds double, double* undef, i32 %i
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i8>, <4 x i8>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i8>, <4 x i8>*
   %c7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 %i
-;CHECK: cost of 0 for instruction: {{.*}} getelementptr inbounds <4 x i16>, <4 x i16>*
+;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i16>, <4 x i16>*
   %c8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 %i
   ; Thumb-2 cannot fold scales larger than 8 to address computation.
 ;CHECK: cost of 1 for instruction: {{.*}} getelementptr inbounds <4 x i32>, <4 x i32>*