Cycles: Correction to previous commit

The change didn't fix difference render result on CUDA as i've hoped,
so reverting change for GPU rendering for now.

Sorry for the noise.

intern/cycles/kernel/kernel_subsurface.h

@@ -85,7 +85,12 @@ ccl_device ShaderClosure *subsurface_scatter_pick_closure(KernelGlobals *kg, Sha
 	return NULL;
 }
 
-ccl_device_noinline float3 subsurface_scatter_eval(ShaderData *sd,
+#ifndef __KERNEL_GPU__
+ccl_device_noinline
+#else
+ccl_device_inline
+#endif
+float3 subsurface_scatter_eval(ShaderData *sd,
                                ShaderClosure *sc,
                                float disk_r,
                                float r,