diff --git a/setup.py b/setup.py
index 68d51a08b2d9e1cf1cb180ce2cc6f4e582a7e492..1e43659bee4fc208be5d83be616d46a758f74d76 100644
--- a/setup.py
+++ b/setup.py
@@ -26,6 +26,7 @@ def get_config_schema():
 
         Switch("CL_TRACE", False, "Enable OpenCL API tracing"),
         Switch("CL_ENABLE_GL", False, "Enable OpenCL<->OpenGL interoperability"),
+        Switch("CL_ENABLE_DEVICE_FISSION", True, "Enable device fission extension, if present"),
 
         IncludeDir("CL", []),
         LibraryDir("CL", []),
@@ -65,13 +66,16 @@ def main():
     if conf["CL_TRACE"]:
         EXTRA_DEFINES["PYOPENCL_TRACE"] = 1
 
-    INCLUDE_DIRS = ['src/cpp'] + conf["BOOST_INC_DIR"] + conf["CL_INC_DIR"]
+    INCLUDE_DIRS = conf["BOOST_INC_DIR"] + conf["CL_INC_DIR"]
 
     ext_kwargs = dict()
 
     if conf["CL_ENABLE_GL"]:
         EXTRA_DEFINES["HAVE_GL"] = 1
 
+    if conf["CL_ENABLE_DEVICE_FISSION"]:
+        EXTRA_DEFINES["PYOPENCL_USE_DEVICE_FISSION"] = 1
+
     ver_dic = {}
     exec(compile(open("pyopencl/version.py").read(), "pyopencl/version.py", 'exec'), ver_dic)
 
diff --git a/src/wrapper/wrap_cl.hpp b/src/wrapper/wrap_cl.hpp
index 855deb924b811ad27385036629ea9ee1837e6fc2..2356d7df291d749860911087d9afdc837ef90d58 100644
--- a/src/wrapper/wrap_cl.hpp
+++ b/src/wrapper/wrap_cl.hpp
@@ -20,7 +20,6 @@
 
 // elsewhere ------------------------------------------------------------------
 #include <CL/cl.h>
-// TBD: Nvidia used to not install cl_ext.h by default. Grr.
 #include <CL/cl_ext.h>
 
 #if defined(_WIN32)
@@ -369,7 +368,7 @@ namespace pyopencl
       {
         if (ownable_reference)
         {
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
           if (retain)
           {
             clRetainDeviceEXT_fn cl_retain_device
@@ -391,7 +390,7 @@ namespace pyopencl
 
       ~device()
       {
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
         if (m_ownable_reference)
         {
           clReleaseDeviceEXT_fn cl_release_device
@@ -523,7 +522,7 @@ namespace pyopencl
           case CL_DEVICE_INTEGRATED_MEMORY_NV:
             DEV_GET_INT_INF(cl_bool);
 #endif
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
           case CL_DEVICE_PARENT_DEVICE_EXT:
             PYOPENCL_GET_OPAQUE_INFO(Device, m_device, param_name, cl_device_id, device);
           case CL_DEVICE_PARTITION_TYPES_EXT:
@@ -542,7 +541,7 @@ namespace pyopencl
         }
       }
 
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
       py::list create_sub_devices(py::object py_properties)
       {
         std::vector<cl_device_partition_property_ext> properties;
diff --git a/src/wrapper/wrap_cl_part_1.cpp b/src/wrapper/wrap_cl_part_1.cpp
index 7121fb7cee5500618869be7e79a0eb0394bd5be4..701b5099325437ce43674ab6e2fbe2b01c73cb9b 100644
--- a/src/wrapper/wrap_cl_part_1.cpp
+++ b/src/wrapper/wrap_cl_part_1.cpp
@@ -37,7 +37,7 @@ void pyopencl_expose_part_1()
       .add_property("obj_ptr", &cls::obj_ptr)
       .def(py::self == py::self)
       .def(py::self != py::self)
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
       .DEF_SIMPLE_METHOD(create_sub_devices)
 #endif
       ;
diff --git a/src/wrapper/wrap_constants.cpp b/src/wrapper/wrap_constants.cpp
index cc8cc52b02678273c046aaacb4fda198325b7569..7375647bf27390dbb159aa50a736e80e10f87ea6 100644
--- a/src/wrapper/wrap_constants.cpp
+++ b/src/wrapper/wrap_constants.cpp
@@ -178,7 +178,7 @@ void pyopencl_expose_constants()
     ADD_ATTR(, INVALID_GLOBAL_WORK_SIZE);
 #endif
 
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
     ADD_ATTR(, DEVICE_PARTITION_FAILED_EXT);
     ADD_ATTR(, INVALID_PARTITION_COUNT_EXT);
     ADD_ATTR(, INVALID_PARTITION_NAME_EXT);
@@ -290,7 +290,7 @@ void pyopencl_expose_constants()
 #ifdef CL_DEVICE_PROFILING_TIMER_OFFSET_AMD
     ADD_ATTR(DEVICE_, PROFILING_TIMER_OFFSET_AMD);
 #endif
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
     ADD_ATTR(DEVICE_, PARENT_DEVICE_EXT);
     ADD_ATTR(DEVICE_, PARTITION_TYPES_EXT);
     ADD_ATTR(DEVICE_, AFFINITY_DOMAINS_EXT);
@@ -600,7 +600,7 @@ void pyopencl_expose_constants()
   {
     py::class_<device_partition_property_ext> cls(
         "device_partition_property_ext", py::no_init);
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
     ADD_ATTR_SUFFIX(DEVICE_PARTITION_, EQUALLY, _EXT);
     ADD_ATTR_SUFFIX(DEVICE_PARTITION_, BY_COUNTS, _EXT);
     ADD_ATTR_SUFFIX(DEVICE_PARTITION_, BY_NAMES, _EXT);
@@ -613,7 +613,7 @@ void pyopencl_expose_constants()
 
   {
     py::class_<affinity_domain_ext> cls("affinity_domain_ext", py::no_init);
-#ifdef cl_ext_device_fission
+#if defined(cl_ext_device_fission) && defined(PYOPENCL_USE_DEVICE_FISSION)
     ADD_ATTR_SUFFIX(AFFINITY_DOMAIN_, L1_CACHE, _EXT);
     ADD_ATTR_SUFFIX(AFFINITY_DOMAIN_, L2_CACHE, _EXT);
     ADD_ATTR_SUFFIX(AFFINITY_DOMAIN_, L3_CACHE, _EXT);