Index: lib/Transforms/InstCombine/InstCombineCasts.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -2243,6 +2243,12 @@
     Type *DstElTy = DstPTy->getElementType();
     Type *SrcElTy = SrcPTy->getElementType();
 
+    // Casting pointers between the same type, but with different address spaces
+    // is an addrspace cast rather than a bitcast.
+    if ((DstElTy == SrcElTy) &&
+        (DstPTy->getAddressSpace() != SrcPTy->getAddressSpace()))
+      return new AddrSpaceCastInst(Src, DestTy);
+
     // If we are casting a alloca to a pointer to a type of the same
     // size, rewrite the allocation instruction to allocate the "right" type.
     // There is no need to modify malloc calls because it is their bitcast that
Index: test/Transforms/InstCombine/gep-addrspace.ll
===================================================================
--- test/Transforms/InstCombine/gep-addrspace.ll
+++ test/Transforms/InstCombine/gep-addrspace.ll
@@ -51,3 +51,21 @@
   %.fca.1.insert = insertvalue { i8, i8 } zeroinitializer, i8 %3, 1
   ret { i8, i8 } %.fca.1.insert
 }
+
+declare spir_func <16 x i32> @my_extern_func()
+
+; check that a bitcast is not generated when we need an addrspace cast
+define void @bitcast_after_gep(<16 x i32>*) {
+entry:
+; CHECK-LABEL: @bitcast_after_gep
+  %1 = bitcast <16 x i32>* %0 to [16 x i32]*
+  %2 = addrspacecast [16 x i32]* %1 to [16 x i32] addrspace(3)*
+  %3 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* %2, i64 0, i64 0
+; CHECK:  %1 = addrspacecast <16 x i32>* %0 to <16 x i32> addrspace(3)*
+  %4 = bitcast i32 addrspace(3)* %3 to <16 x i32> addrspace(3)*
+; CHECK:  %call = call spir_func <16 x i32> @my_extern_func()
+  %call = call spir_func <16 x i32> @my_extern_func()
+; CHECK:  store <16 x i32> %call, <16 x i32> addrspace(3)* %1
+  store <16 x i32> %call, <16 x i32> addrspace(3)* %4
+  ret void
+}