diff --git a/pyopencl/elementwise.py b/pyopencl/elementwise.py
index 9a6d794eeaeffde8a60f7a44aa4ee3f80cca9a79..2375c6ee90b685efce1021f21dbaa80f2073540d 100644
--- a/pyopencl/elementwise.py
+++ b/pyopencl/elementwise.py
@@ -381,7 +381,9 @@ def get_take_kernel(context, dtype, idx_dtype, vec_count=1):
                 "dest%d[i] = src%d[src_idx];" % (i, i)
                 for i in range(vec_count)))
 
-    return get_elwise_kernel(context, args, body, name="take")
+    return get_elwise_kernel(context, args, body,
+            preamble=dtype_to_c_struct(context.devices[0], dtype),
+            name="take")
 
 
 @context_dependent_memoize
@@ -419,7 +421,9 @@ def get_take_put_kernel(context, dtype, idx_dtype, with_offsets, vec_count=1):
                 "%(idx_tp)s dest_idx = gmem_dest_idx[i];\n" % ctx)
             + "\n".join(get_copy_insn(i) for i in range(vec_count)))
 
-    return get_elwise_kernel(context, args, body, name="take_put")
+    return get_elwise_kernel(context, args, body,
+            preamble=dtype_to_c_struct(context.devices[0], dtype),
+            name="take_put")
 
 
 @context_dependent_memoize
@@ -444,7 +448,9 @@ def get_put_kernel(context, dtype, idx_dtype, vec_count=1):
             + "\n".join("dest%d[dest_idx] = src%d[i];" % (i, i)
                 for i in range(vec_count)))
 
-    return get_elwise_kernel(context, args, body, name="put")
+    return get_elwise_kernel(context, args, body,
+            preamble=dtype_to_c_struct(context.devices[0], dtype),
+            name="put")
 
 
 @context_dependent_memoize
@@ -462,6 +468,7 @@ def get_copy_kernel(context, dtype_dest, dtype_src):
                 "tp_src": dtype_to_ctype(dtype_src),
                 },
             "dest[i] = %s" % src,
+            preamble=dtype_to_c_struct(context.devices[0], dtype),
             name="copy")
 
 
diff --git a/test/test_array.py b/test/test_array.py
index d1e0c082ca99c884aaa2a95800d3fc3f24e0492c..3a2e88c09908cc0061275ed25e810c7b7e117408 100644
--- a/test/test_array.py
+++ b/test/test_array.py
@@ -289,6 +289,35 @@ def test_custom_type_fill(ctx_factory):
 
     assert np.array_equal(np.zeros(n, dtype), z)
 
+
+def test_custom_type_take_put(ctx_factory):
+    context = ctx_factory()
+    queue = cl.CommandQueue(context)
+
+    dtype = np.dtype([
+        ("cur_min", np.int32),
+        ("cur_max", np.int32),
+        ])
+
+    from pyopencl.tools import get_or_register_dtype, match_dtype_to_c_struct
+
+    name = "tp_type"
+    dtype, c_decl = match_dtype_to_c_struct(queue.device, name, dtype)
+    dtype = get_or_register_dtype(name, dtype)
+
+    n = 100
+    z = np.empty(100, dtype)
+    z["cur_min"] = np.arange(n)
+    z["cur_max"] = np.arange(n)**2
+
+    z_dev = cl.array.to_device(queue, z)
+    ind = cl.array.arange(queue, n, step=3, dtype=np.int32)
+
+    z_ind_ref = z[ind.get()]
+    z_ind = z_dev[ind]
+
+    assert np.array_equal(z_ind.get(), z_ind_ref)
+
 # }}}