Cycles: Use native saturate function for CUDA

This more a workaround for CUDA optimizer which can't optimize clamp(x, 0, 1)
into a single instruction and uses 4 instructions instead.

Original patch by @lockal with own modification:

  Don't make changes outside of the kernel. They don't make any difference
  anyway and term saturate() has a bit different meaning outside of kernel.

This gives around 2% of speedup in Barcelona file, but in more complex shader
setups with lots of math nodes with clamping speedup could be much nicer.

Subscribers: dingto

Projects: #cycles

Differential Revision: https://developer.blender.org/D1224

1 files changed, 3 insertions, 3 deletions

diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index ccc1e3df730..08a6c01162c 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -433,17 +433,17 @@ ccl_device void svm_node_tex_image_box(KernelGlobals *kg, ShaderData *sd, float
 		/* in case of blending, test for mixes between two textures */
 		if(N.z < (1.0f - limit)*(N.y + N.x)) {
 			weight.x = N.x/(N.x + N.y);
-			weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
+			weight.x = saturate((weight.x - 0.5f*(1.0f - blend))/blend);
 			weight.y = 1.0f - weight.x;
 		}
 		else if(N.x < (1.0f - limit)*(N.y + N.z)) {
 			weight.y = N.y/(N.y + N.z);
-			weight.y = clamp((weight.y - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
+			weight.y = saturate((weight.y - 0.5f*(1.0f - blend))/blend);
 			weight.z = 1.0f - weight.y;
 		}
 		else if(N.y < (1.0f - limit)*(N.x + N.z)) {
 			weight.x = N.x/(N.x + N.z);
-			weight.x = clamp((weight.x - 0.5f*(1.0f - blend))/blend, 0.0f, 1.0f);
+			weight.x = saturate((weight.x - 0.5f*(1.0f - blend))/blend);
 			weight.z = 1.0f - weight.x;
 		}
 		else {