diff --git a/README.rst b/README.rst
index 7a65693e38..04c05fac52 100644
--- a/README.rst
+++ b/README.rst
@@ -206,10 +206,14 @@ Benchmarks for the low-level Cython functions::
 Slow
 ====
 
-Currently only 1d, 2d, and 3d input arrays with data type (dtype) int32,
-int64, float32, and float64 are accelerated. All other ndim/dtype
+By default, only 1d, 2d, and 3d input arrays with data type (dtype)
+int32, int64, float32, and float64 are accelerated. All other ndim/dtype
 combinations result in calls to slower, unaccelerated functions.
 
+With the development version of bottleneck, it is possible to accelerate higher
+dimensional arrays if an appropriate flag is set at compile-time (see
+`Fast functions for higher dimensions`_ below).
+
 License
 =======
 
@@ -258,7 +262,8 @@ Cython.)
 Bottleneck takes a few minutes to build on newer machines. On older machines
 it can take a lot longer (one user reported 30 minutes!).
 
-**GNU/Linux, Mac OS X, et al.**
+GNU/Linux, Mac OS X, et al.
+---------------------------
 
 To install Bottleneck::
 
@@ -271,7 +276,8 @@ Or, if you wish to specify where Bottleneck is installed, for example inside
     $ python setup.py build
     $ sudo python setup.py install --prefix=/usr/local
 
-**Windows**
+Windows
+-------
 
 You can compile Bottleneck using the instructions below or you can use the
 Windows binaries created by Christoph Gohlke:
@@ -286,7 +292,21 @@ commands::
     python setup.py build --compiler=mingw32
     python setup.py install
 
-**Post install**
+Fast functions for higher dimensions
+------------------------------------
+
+If Cython is available, it is possible to adjust the number of supported
+dimensions at *compile-time* by setting the environment variable ``NDIM_MAX``,
+which defaults to 3. For example, to build 1d, 2d, 3d and 4d versions of all
+functions::
+
+    NDIM_MAX=4 make clean pyx cfiles build
+
+The size of the generated code (and the time required to compile it) scales as
+``NDIM_MAX`` squared.
+
+Post install
+------------
 
 After you have installed Bottleneck, run the suite of unit tests::
 
diff --git a/RELEASE.rst b/RELEASE.rst
index c333b36c2b..1f52441302 100644
--- a/RELEASE.rst
+++ b/RELEASE.rst
@@ -15,6 +15,9 @@ Bottleneck 0.9.0
 
 - bn.slow.move functions with method='strides' no longer limited to ndim < 4
 - Use setuptools to enable automatic installation of numpy by pip
+- A new compile time option NDIM_MAX allows for adjusting the number of
+  dimensions for which fast functions are generated
+- Moving window functions now release Python's global interpreter lock
 
 Older versions
 ==============
diff --git a/bottleneck/src/template/template.py b/bottleneck/src/template/template.py
index 04c61d73de..6a073624bb 100644
--- a/bottleneck/src/template/template.py
+++ b/bottleneck/src/template/template.py
@@ -381,7 +381,6 @@ def _parse_product_range_line(line):
                                           'range(%s)' % range_inside)
     for_statement = for_statement.replace('INDEXN', 'INDEX%d')
     range_list = eval('range(%s)' % prod_inside)
-    # range_list = range(*[int(n) for n in prod_inside.split(',')])
     return [TAB * i + for_statement % (n, n)
             for i, n in enumerate(range_list)]
 
@@ -443,9 +442,24 @@ def loop_expand_product_range(text):
 def unindex_0dimensional(text, ydtype):
     """
     If a loop template includes assignments to 0-dimensional return variables
-    like 'y[] = ', replace them with return statemetns.
+    like 'y[] = ', replace them with return statements (cast to ydtype), after
+    removing all existing return statements.
+
+    Examples
+    --------
+
+    >>> text = '''\
+    ...     if foo:
+    ...         PyArray_FillWithScalar(y, NAN)
+    ...     y[] = a
+    ...     return y
+    ...     ''''
+    >>> print(unindex_0dimensional(text, 'float64'))
+        if foo:
+            return np.float64(NAN)
+        return np.float64(a)
     """
-    if '[] = ' in text:
+    if 'y[] = ' in text:
         text = text.replace('return y', '')
         text = re.sub(r'PyArray_FillWithScalar\(y, (\w+)\)',
                       r'return np.%s(\1)' % ydtype, text)