Index: llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1989,9 +1989,6 @@
                                Indices, "", IsInBounds));
   }
 
-  if (Src->getResultElementType() != GEP.getSourceElementType())
-    return nullptr;
-
   SmallVector<Value*, 8> Indices;
 
   // Find out whether the last index in the source GEP is a sequential idx.
@@ -2001,7 +1998,8 @@
     EndsWithSequential = I.isSequential();
 
   // Can we combine the two pointer arithmetics offsets?
-  if (EndsWithSequential) {
+  if (Src->getResultElementType() == GEP.getSourceElementType() && 
+      EndsWithSequential) {
     // Replace: gep (gep %P, long B), long A, ...
     // With:    T = long A+B; gep %P, T, ...
     Value *SO1 = Src->getOperand(Src->getNumOperands()-1);
@@ -2031,12 +2029,31 @@
     Indices.append(Src->op_begin()+1, Src->op_end()-1);
     Indices.push_back(Sum);
     Indices.append(GEP.op_begin()+2, GEP.op_end());
-  } else if (isa<Constant>(*GEP.idx_begin()) &&
+  } else if (Src->getResultElementType() == GEP.getSourceElementType() &&
+             isa<Constant>(*GEP.idx_begin()) &&
              cast<Constant>(*GEP.idx_begin())->isNullValue() &&
              Src->getNumOperands() != 1) {
     // Otherwise we can do the fold if the first index of the GEP is a zero
     Indices.append(Src->op_begin()+1, Src->op_end());
     Indices.append(GEP.idx_begin()+1, GEP.idx_end());
+  } else { 
+    Indices.append(Src->op_begin()+1, Src->op_end());
+
+    // The indices of GEP may have been folded; in such case try to get
+    // the types to match by appending zeros to the indices of Src
+    Type *SrcType = Src->getResultElementType();
+    while (SrcType && SrcType->isAggregateType() && 
+           SrcType != GEP.getSourceElementType()) {
+        Indices.push_back(Builder.getInt32(0));
+        SrcType = GetElementPtrInst::getIndexedType(
+          Src->getSourceElementType(), Indices);
+    }
+    if (SrcType == Src->getResultElementType() || 
+        SrcType != GEP.getSourceElementType())
+      return nullptr;
+
+    Indices.pop_back();
+    Indices.append(GEP.idx_begin(), GEP.idx_end());
   }
 
   if (!Indices.empty())
Index: llvm/test/Transforms/InstCombine/gepgep-after-fold.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/InstCombine/gepgep-after-fold.ll
@@ -0,0 +1,65 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=instcombine -S < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32-S128"
+
+%s1 = type [3 x [3 x i32]]
+%s2 = type [3 x [3 x [3 x i32]]]
+
+define void @test_unfold_one(ptr %x, i32 %i, i32 %j) {
+; CHECK-LABEL: @test_unfold_one(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [3 x [3 x i32]], ptr [[X:%.*]], i32 [[I:%.*]], i32 2, i32 [[J:%.*]]
+; CHECK-NEXT:    store i32 1, ptr [[GEP]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %src = getelementptr inbounds %s1, ptr %x, i32 %i, i32 2
+  %gep = getelementptr inbounds i32, ptr %src, i32 %j
+  store i32 1, ptr %gep
+  ret void
+}
+
+define void @test_unfold_one2(ptr %x, i32 %i, i32 %j) {
+; CHECK-LABEL: @test_unfold_one2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [3 x [3 x [3 x i32]]], ptr [[X:%.*]], i32 [[I:%.*]], i32 2, i32 [[J:%.*]]
+; CHECK-NEXT:    store i32 1, ptr [[GEP]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %src = getelementptr inbounds %s2, ptr %x, i32 %i, i32 2
+  %gep = getelementptr inbounds [3 x i32], ptr %src, i32 %j
+  store i32 1, ptr %gep
+  ret void
+}
+
+define void @test_unfold_two(ptr %x, i32 %i, i32 %j) {
+; CHECK-LABEL: @test_unfold_two(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [3 x [3 x [3 x i32]]], ptr [[X:%.*]], i32 [[I:%.*]], i32 2, i32 0, i32 [[J:%.*]]
+; CHECK-NEXT:    store i32 1, ptr [[GEP]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %src = getelementptr inbounds %s2, ptr %x, i32 %i, i32 2
+  %gep = getelementptr inbounds i32, ptr %src, i32 %j
+  store i32 1, ptr %gep
+  ret void
+}
+
+define void @test_negative(ptr %x, i32 %i, i32 %j) {
+; CHECK-LABEL: @test_negative(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[SRC:%.*]] = getelementptr inbounds [3 x [3 x i32]], ptr [[X:%.*]], i32 [[I:%.*]], i32 2
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [3 x i32], ptr [[SRC]], i32 [[J:%.*]]
+; CHECK-NEXT:    store i32 1, ptr [[GEP]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %src = getelementptr inbounds %s1, ptr %x, i32 %i, i32 2
+  ; should be %gep = getelementptr inbounds [3 x i32], ptr %src, i32 0, i32 %j
+  %gep = getelementptr inbounds [3 x i32], ptr %src, i32 %j
+  store i32 1, ptr %gep
+  ret void
+}