Hi!
I'm attaching a patch to use clang-3.8 as host compiler when building
the CUDA kernel. It's hard coded, so maybe it should be a parameter from
somewhere instead.
/Gustaf
Index: blender-2.78.a+dfsg0/intern/cycles/device/device_cuda.cpp
===================================================================
--- blender-2.78.a+dfsg0.orig/intern/cycles/device/device_cuda.cpp
+++ blender-2.78.a+dfsg0/intern/cycles/device/device_cuda.cpp
@@ -384,6 +384,7 @@ public:
string command = string_printf("\"%s\" "
"-arch=sm_%d%d "
"--cubin \"%s\" "
+ "-ccbin clang-3.8 "
"-o \"%s\" "
"%s ",
nvcc,
