Index: lib/Transforms/InstCombine/InstCombineCasts.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -1719,7 +1719,7 @@
   // cast to be exposed to other transforms.
   unsigned AS = CI.getAddressSpace();
   if (CI.getOperand(0)->getType()->getScalarSizeInBits() !=
-      DL.getPointerSizeInBits(AS)) {
+      DL.getIndexSizeInBits(AS)) {
     Type *Ty = DL.getIntPtrType(CI.getContext(), AS);
     if (CI.getType()->isVectorTy()) // Handle vectors of pointers.
       Ty = VectorType::get(Ty, CI.getType()->getVectorNumElements());
Index: test/Transforms/InstCombine/ptr-int-cast_custom_gep.ll
===================================================================
--- /dev/null
+++ test/Transforms/InstCombine/ptr-int-cast_custom_gep.ll
@@ -0,0 +1,77 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+target datalayout = "E-p:40:64:64:32-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
+
+define i1 @test1(i32 *%x) nounwind {
+; CHECK-LABEL: @test1(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = ptrtoint i32* [[X:%.*]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[TMP0]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; CHECK-NEXT:    ret i1 [[TMP2]]
+;
+entry:
+  %0 = ptrtoint i32* %x to i1
+  ret i1 %0
+}
+
+define i32* @test2(i128 %x) nounwind {
+; CHECK-LABEL: @test2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = trunc i128 [[X:%.*]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = inttoptr i32 [[TMP0]] to i32*
+; CHECK-NEXT:    ret i32* [[TMP1]]
+;
+entry:
+  %0 = inttoptr i128 %x to i32*
+  ret i32* %0
+}
+
+; PR3574
+define i64 @f0(i32 %a0) nounwind {
+; CHECK-LABEL: @f0(
+; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[A0:%.*]] to i64
+; CHECK-NEXT:    ret i64 [[T1]]
+;
+  %t0 = inttoptr i32 %a0 to i8*
+  %t1 = ptrtoint i8* %t0 to i64
+  ret i64 %t1
+}
+
+define <4 x i32> @test4(<4 x i8*> %arg) nounwind {
+; CHECK-LABEL: @test4(
+; CHECK-NEXT:    [[P1:%.*]] = ptrtoint <4 x i8*> [[ARG:%.*]] to <4 x i32>
+; CHECK-NEXT:    ret <4 x i32> [[P1]]
+;
+  %p1 = ptrtoint <4 x i8*> %arg to <4 x i32>
+  ret <4 x i32> %p1
+}
+
+define <4 x i128> @test5(<4 x i8*> %arg) nounwind {
+; CHECK-LABEL: @test5(
+; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint <4 x i8*> [[ARG:%.*]] to <4 x i32>
+; CHECK-NEXT:    [[P1:%.*]] = zext <4 x i32> [[TMP1]] to <4 x i128>
+; CHECK-NEXT:    ret <4 x i128> [[P1]]
+;
+  %p1 = ptrtoint <4 x i8*> %arg to <4 x i128>
+  ret <4 x i128> %p1
+}
+
+define <4 x i8*> @test6(<4 x i32> %arg) nounwind {
+; CHECK-LABEL: @test6(
+; CHECK-NEXT:    [[P1:%.*]] = inttoptr <4 x i32> [[ARG:%.*]] to <4 x i8*>
+; CHECK-NEXT:    ret <4 x i8*> [[P1]]
+;
+  %p1 = inttoptr <4 x i32> %arg to <4 x i8*>
+  ret <4 x i8*> %p1
+}
+
+define <4 x i8*> @test7(<4 x i128> %arg) nounwind {
+; CHECK-LABEL: @test7(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc <4 x i128> [[ARG:%.*]] to <4 x i32>
+; CHECK-NEXT:    [[P1:%.*]] = inttoptr <4 x i32> [[TMP1]] to <4 x i8*>
+; CHECK-NEXT:    ret <4 x i8*> [[P1]]
+;
+  %p1 = inttoptr <4 x i128> %arg to <4 x i8*>
+  ret <4 x i8*> %p1
+}
Index: test/Transforms/SeparateConstOffsetFromGEP/custom_dl.ll
===================================================================
--- /dev/null
+++ test/Transforms/SeparateConstOffsetFromGEP/custom_dl.ll
@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt  -S -separate-const-offset-from-gep -reassociate-geps-verify-no-dead-code -gvn < %s | FileCheck %s
+
+target datalayout = "e-m:m-p:40:64:64:32-i32:32-i16:16-i8:8-n32"
+
+@array = internal addrspace(4) constant [4096 x [32 x float]] zeroinitializer, align 4
+
+define  void @sum_of_array(i32 %x, i32 %y, float addrspace(1)* nocapture %output) {
+; CHECK-LABEL: @sum_of_array(
+; CHECK-NEXT:    [[TMP:%.*]] = sext i32 [[Y:%.*]] to i64
+; CHECK-NEXT:    [[TMP1:%.*]] = sext i32 [[X:%.*]] to i64
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i32 0, i32 [[X]], i32 [[Y]]
+; CHECK-NEXT:    [[TMP4:%.*]] = load float, float addrspace(4)* [[TMP2]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = fadd float [[TMP4]], 0.000000e+00
+; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[Y]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = sext i32 [[TMP6]] to i64
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i32 0, i32 [[X]], i32 [[TMP6]]
+; CHECK-NEXT:    [[TMP10:%.*]] = load float, float addrspace(4)* [[TMP8]], align 4
+; CHECK-NEXT:    [[TMP11:%.*]] = fadd float [[TMP5]], [[TMP10]]
+; CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[X]], 1
+; CHECK-NEXT:    [[TMP13:%.*]] = sext i32 [[TMP12]] to i64
+; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i32 0, i32 [[TMP12]], i32 [[Y]]
+; CHECK-NEXT:    [[TMP16:%.*]] = load float, float addrspace(4)* [[TMP14]], align 4
+; CHECK-NEXT:    [[TMP17:%.*]] = fadd float [[TMP11]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i32 0, i32 [[TMP12]], i32 [[TMP6]]
+; CHECK-NEXT:    [[TMP20:%.*]] = load float, float addrspace(4)* [[TMP18]], align 4
+; CHECK-NEXT:    [[TMP21:%.*]] = fadd float [[TMP17]], [[TMP20]]
+; CHECK-NEXT:    store float [[TMP21]], float addrspace(1)* [[OUTPUT:%.*]], align 4
+; CHECK-NEXT:    ret void
+;
+  %tmp = sext i32 %y to i64
+  %tmp1 = sext i32 %x to i64
+  %tmp2 = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i64 0, i64 %tmp1, i64 %tmp
+  %tmp4 = load float, float addrspace(4)* %tmp2, align 4
+  %tmp5 = fadd float %tmp4, 0.000000e+00
+  %tmp6 = add i32 %y, 1
+  %tmp7 = sext i32 %tmp6 to i64
+  %tmp8 = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i64 0, i64 %tmp1, i64 %tmp7
+  %tmp10 = load float, float addrspace(4)* %tmp8, align 4
+  %tmp11 = fadd float %tmp5, %tmp10
+  %tmp12 = add i32 %x, 1
+  %tmp13 = sext i32 %tmp12 to i64
+  %tmp14 = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i64 0, i64 %tmp13, i64 %tmp
+  %tmp16 = load float, float addrspace(4)* %tmp14, align 4
+  %tmp17 = fadd float %tmp11, %tmp16
+  %tmp18 = getelementptr inbounds [4096 x [32 x float]], [4096 x [32 x float]] addrspace(4)* @array, i64 0, i64 %tmp13, i64 %tmp7
+  %tmp20 = load float, float addrspace(4)* %tmp18, align 4
+  %tmp21 = fadd float %tmp17, %tmp20
+  store float %tmp21, float addrspace(1)* %output, align 4
+  ret void
+}