diff --git a/loopy/auto_test.py b/loopy/auto_test.py
index 690e2bec2492eb19e029b5ad1b4f4e9919ceb3ba..c299df4e511603647de13d70614e886dad3257f1 100644
--- a/loopy/auto_test.py
+++ b/loopy/auto_test.py
@@ -324,8 +324,7 @@ def _enumerate_cl_devices_for_ref_test():
 
 # {{{ main automatic testing entrypoint
 
-def _complain_about_deferred_types(knl):
-    print "COMPL"
+def _check_for_deferred_types(knl):
     for arg in knl.args:
         if arg.dtype is None:
             raise RuntimeError("Automatic testing requires that all "
@@ -384,7 +383,7 @@ def auto_test_vs_ref(
 
     # {{{ compile and run reference code
 
-    _complain_about_deferred_types(ref_knl)
+    _check_for_deferred_types(ref_knl)
 
     found_ref_device = False
 
@@ -486,7 +485,7 @@ def auto_test_vs_ref(
             test_kernels = [test_knl]
 
     for i, kernel in enumerate(test_kernels):
-        _complain_about_deferred_types(kernel)
+        _check_for_deferred_types(kernel)
         compiled = CompiledKernel(ctx, kernel, options=options,
                 codegen_kwargs=codegen_kwargs)
 
diff --git a/loopy/codegen/__init__.py b/loopy/codegen/__init__.py
index 3541e8f222c292bd1f90e6b2e0bbcafc9b787f4b..5ebce957c00099d8dfa524c16b09ec559275a363 100644
--- a/loopy/codegen/__init__.py
+++ b/loopy/codegen/__init__.py
@@ -272,8 +272,22 @@ class ImplementedDataInfo(Record):
 
 # {{{ main code generation entrypoint
 
+def _check_for_deferred_types(knl):
+    for arg in knl.args:
+        if arg.dtype is None:
+            raise RuntimeError("Code generation requires that all "
+                    "argument types are known.  Argument '%s' has "
+                    "an unknown/deferred type." % arg.name)
+
+
 def generate_code(kernel, with_annotation=False,
         allow_complex=None):
+    if kernel.schedule is None:
+        from loopy.schedule import get_one_scheduled_kernel
+        kernel = get_one_scheduled_kernel(kernel)
+
+    _check_for_deferred_types(kernel)
+
     from cgen import (FunctionBody, FunctionDeclaration,
             Value, Module, Block,
             Line, Const, LiteralLines, Initializer)
diff --git a/loopy/compiled.py b/loopy/compiled.py
index c81f8dcdd900eb62d18575ce8e40b05808b67276..d799186f5811b1e0d8c7ae3120ea3358fbe018f3 100644
--- a/loopy/compiled.py
+++ b/loopy/compiled.py
@@ -694,25 +694,6 @@ def generate_invoker(kernel, impl_arg_info, flags):
 
 # {{{ compiled kernel object
 
-def _get_kernel_from_iterable(iterable):
-    kernel_count = 0
-
-    for scheduled_kernel in iterable:
-        kernel_count += 1
-
-        if kernel_count == 1:
-            # use the first schedule
-            result = scheduled_kernel
-
-        if kernel_count == 2:
-            from warnings import warn
-            warn("kernel scheduling was ambiguous--more than one "
-                    "schedule found, ignoring", stacklevel=2)
-            break
-
-    return result
-
-
 class _CLKernelInfo(Record):
     pass
 
@@ -748,7 +729,6 @@ class CompiledKernel:
     def get_kernel(self, arg_to_dtype_set):
         kernel = self.kernel
 
-        import loopy as lp
         from loopy.kernel.tools import add_argument_dtypes
 
         if arg_to_dtype_set:
@@ -762,8 +742,8 @@ class CompiledKernel:
             kernel = infer_unknown_types(kernel, expect_completion=True)
 
         if kernel.schedule is None:
-            kernel = _get_kernel_from_iterable(
-                    lp.generate_loop_schedules(kernel))
+            from loopy.schedule import get_one_scheduled_kernel
+            kernel = get_one_scheduled_kernel(kernel)
 
         return kernel
 
diff --git a/loopy/schedule.py b/loopy/schedule.py
index 4d5c0314ae6e53922391e35c2d2c9604b10119ed..52d001ed46f94c30f24a7711471cd5034c591672 100644
--- a/loopy/schedule.py
+++ b/loopy/schedule.py
@@ -879,4 +879,24 @@ def generate_loop_schedules(kernel, debug_args={}):
 
 # }}}
 
+
+def get_one_scheduled_kernel(kernel):
+    kernel_count = 0
+
+    for scheduled_kernel in generate_loop_schedules(kernel):
+        kernel_count += 1
+
+        if kernel_count == 1:
+            # use the first schedule
+            result = scheduled_kernel
+
+        if kernel_count == 2:
+            from warnings import warn
+            warn("kernel scheduling was ambiguous--more than one "
+                    "schedule found, ignoring", stacklevel=2)
+            break
+
+    return result
+
+
 # vim: foldmethod=marker
diff --git a/test/test_loopy.py b/test/test_loopy.py
index 57e7476877d1ed85b543ec4a8a020b32426b1d50..272b0130b53edc600c599e566856308fe108ba1e 100644
--- a/test/test_loopy.py
+++ b/test/test_loopy.py
@@ -1110,6 +1110,28 @@ def test_offsets_and_slicing(ctx_factory):
     assert la.norm(b_full.get() - b_full_h) < 1e-13
 
 
+def test_vector_ilp_with_prefetch(ctx_factory):
+    ctx = ctx_factory()
+
+    knl = lp.make_kernel(ctx.devices[0],
+            "{ [i]: 0<=i<n }",
+            "out[i] = 2*a[i]",
+            [
+                # Tests that comma-d arguments interoperate with
+                # argument guessing.
+                lp.GlobalArg("out,a", np.float32, shape=lp.auto),
+                "..."
+                ])
+
+    knl = lp.split_iname(knl, "i", 128, inner_tag="l.0")
+    knl = lp.split_iname(knl, "i_outer", 4, outer_tag="g.0", inner_tag="ilp")
+    knl = lp.add_prefetch(knl, "a", ["i_inner", "i_outer_inner"])
+
+    code, info = lp.generate_code(knl)
+    import re
+    assert len(list(re.finditer("barrier", code))) == 1
+
+
 if __name__ == "__main__":
     import sys
     if len(sys.argv) > 1: