Code cleanup: move rng into path state.

Also pass by value and don't write back now that it is just a hash for seeding
and no longer an LCG state. Together this makes CUDA a tiny bit faster in my
tests, but mainly simplifies code.

intern/cycles/kernel/kernel_subsurface.h

@@ -219,7 +219,7 @@ ccl_device void subsurface_color_bump_blur(KernelGlobals *kg,
 
 	if(bump || texture_blur > 0.0f) {
 		/* average color and normal at incoming point */
-		shader_eval_surface(kg, sd, NULL, state, 0.0f, state_flag, SHADER_CONTEXT_SSS);
+		shader_eval_surface(kg, sd, state, 0.0f, state_flag, SHADER_CONTEXT_SSS);
 		float3 in_color = shader_bssrdf_sum(sd, (bump)? N: NULL, NULL);
 
 		/* we simply divide out the average color and multiply with the average
@@ -243,7 +243,7 @@ ccl_device_inline int subsurface_scatter_multi_intersect(
         SubsurfaceIntersection *ss_isect,
         ShaderData *sd,
         ShaderClosure *sc,
-        RNG *lcg_state,
+        uint *lcg_state,
         float disk_u,
         float disk_v,
         bool all)