diff --git a/builtins/util.m4 b/builtins/util.m4
index 4cb46310..c1582e51 100644
--- a/builtins/util.m4
+++ b/builtins/util.m4
@@ -815,7 +815,6 @@ forloop(i, 1, eval(WIDTH-1), `  %ret_`'i = insertelement <WIDTH x $1> %ret_`'eva
   ret <WIDTH x $1> %ret_`'eval(WIDTH-1)
 
 not_const:
-  ; store two instances of the vector into memory
   %ptr = alloca <WIDTH x $1>, i32 3
   %ptr0 = getelementptr <WIDTH x $1> * %ptr, i32 0
   store <WIDTH x $1> zeroinitializer, <WIDTH x $1> * %ptr0
@@ -824,8 +823,7 @@ not_const:
   %ptr2 = getelementptr <WIDTH x $1> * %ptr, i32 2
   store <WIDTH x $1> zeroinitializer, <WIDTH x $1> * %ptr2
 
-  ; compute offset in [0,vectorwidth-1], then index into the doubled-up vector
-  %offset = add i32 %1, 16
+  %offset = add i32 %1, WIDTH
   %ptr_as_elt_array = bitcast <WIDTH x $1> * %ptr to [eval(3*WIDTH) x $1] *
   %load_ptr = getelementptr [eval(3*WIDTH) x $1] * %ptr_as_elt_array, i32 0, i32 %offset
   %load_ptr_vec = bitcast $1 * %load_ptr to <WIDTH x $1> *
diff --git a/tests/shift-1.ispc b/tests/shift-1.ispc
new file mode 100644
index 00000000..2062e36b
--- /dev/null
+++ b/tests/shift-1.ispc
@@ -0,0 +1,14 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    int a = aFOO[programIndex]; 
+    int rot = shift(a, -1);
+    RET[programIndex] = rot;
+}
+
+export void result(uniform float RET[]) {
+    varying int val = programIndex;
+    if (val < 0) val = 0;	 
+    RET[programIndex] = val;	 
+}   
diff --git a/tests/shift-2.ispc b/tests/shift-2.ispc
new file mode 100644
index 00000000..6cb88e8a
--- /dev/null
+++ b/tests/shift-2.ispc
@@ -0,0 +1,15 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    int a = aFOO[programIndex]; 
+    uniform int delta = b - 6; // -1
+    int rot = shift(a, delta);
+    RET[programIndex] = rot;
+}
+
+export void result(uniform float RET[]) {
+    varying int val = programIndex;
+    if (val < 0) val = 0;	 
+    RET[programIndex] = val;	 
+}
diff --git a/tests/shift-3.ispc b/tests/shift-3.ispc
new file mode 100644
index 00000000..827d076f
--- /dev/null
+++ b/tests/shift-3.ispc
@@ -0,0 +1,14 @@
+
+export uniform int width() { return programCount; }
+
+export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+    int a = aFOO[programIndex]; 
+    int rot = shift(a, 1);
+    RET[programIndex] = rot;
+}
+
+export void result(uniform float RET[]) {
+    varying int val = 2 + programIndex;
+    if (val > programCount) val = 0;	 
+    RET[programIndex] = val;	 
+}