diff --git a/intern/cycles/kernel/kernel_compat_cuda.h b/intern/cycles/kernel/kernel_compat_cuda.h
index 38708f7ff0b..6d1cf055f2c 100644
--- a/intern/cycles/kernel/kernel_compat_cuda.h
+++ b/intern/cycles/kernel/kernel_compat_cuda.h
@@ -53,6 +53,10 @@
 #define ccl_may_alias
 #define ccl_addr_space
 #define ccl_restrict __restrict__
+/* TODO(sergey): In theory we might use references with CUDA, however
+ * performance impact yet to be investigated.
+ */
+#define ccl_ref
 #define ccl_align(n) __align__(n)
 
 #define ATTR_FALLTHROUGH
diff --git a/intern/cycles/kernel/kernel_compat_opencl.h b/intern/cycles/kernel/kernel_compat_opencl.h
index 21eba971688..36d6031d042 100644
--- a/intern/cycles/kernel/kernel_compat_opencl.h
+++ b/intern/cycles/kernel/kernel_compat_opencl.h
@@ -42,6 +42,7 @@
 #define ccl_local_param __local
 #define ccl_private __private
 #define ccl_restrict restrict
+#define ccl_ref
 #define ccl_align(n) __attribute__((aligned(n)))
 
 #ifdef __SPLIT_KERNEL__
diff --git a/intern/cycles/util/util_defines.h b/intern/cycles/util/util_defines.h
index d0d87e74332..ae654092c87 100644
--- a/intern/cycles/util/util_defines.h
+++ b/intern/cycles/util/util_defines.h
@@ -35,6 +35,7 @@
 #  define ccl_local_param
 #  define ccl_private
 #  define ccl_restrict __restrict
+#  define ccl_ref &
 #  define __KERNEL_WITH_SSE_ALIGN__
 
 #  if defined(_WIN32) && !defined(FREE_WINDOWS)