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

2 files changed, 3 insertions, 3 deletions

diff --git a/intern/cycles/render/buffers.cpp b/intern/cycles/render/buffers.cpp
index 5202bf5862c..dc00da1bf88 100644
--- a/intern/cycles/render/buffers.cpp
+++ b/intern/cycles/render/buffers.cpp
@@ -187,7 +187,7 @@ bool RenderBuffers::get_pass_rect(PassType type, float exposure, int sample, int
 			else if(type == PASS_MIST) {
 				for(int i = 0; i < size; i++, in += pass_stride, pixels++) {
 					float f = *in;
-					pixels[0] = clamp(f*scale_exposure, 0.0f, 1.0f);
+					pixels[0] = saturate(f*scale_exposure);
 				}
 			}
 #ifdef WITH_CYCLES_DEBUG
@@ -298,7 +298,7 @@ bool RenderBuffers::get_pass_rect(PassType type, float exposure, int sample, int
 					pixels[2] = f.z*scale_exposure;
 
 					/* clamp since alpha might be > 1.0 due to russian roulette */
-					pixels[3] = clamp(f.w*scale, 0.0f, 1.0f);
+					pixels[3] = saturate(f.w*scale);
 				}
 			}
 		}
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index 747bafcc904..8529855ff2e 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -3752,7 +3752,7 @@ void MathNode::compile(SVMCompiler& compiler)
 		                                 value1_in->value.x,
 		                                 value2_in->value.x);
 		if(use_clamp) {
-			optimized_value = clamp(optimized_value, 0.0f, 1.0f);
+			optimized_value = saturate(optimized_value);
 		}
 		compiler.add_node(NODE_VALUE_F,
 		                  __float_as_int(optimized_value),