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

4 files changed, 16 insertions, 16 deletions

diff --git a/intern/cycles/kernel/closure/bsdf_microfacet.h b/intern/cycles/kernel/closure/bsdf_microfacet.h
index ca68b1eb940..6a50bbed3b3 100644
--- a/intern/cycles/kernel/closure/bsdf_microfacet.h
+++ b/intern/cycles/kernel/closure/bsdf_microfacet.h
@@ -235,7 +235,7 @@ ccl_device_inline float3 microfacet_sample_stretched(
 
 ccl_device int bsdf_microfacet_ggx_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
 	sc->data1 = sc->data0; /* alpha_y */
 	
 	sc->type = CLOSURE_BSDF_MICROFACET_GGX_ID;
@@ -245,8 +245,8 @@ ccl_device int bsdf_microfacet_ggx_setup(ShaderClosure *sc)
 
 ccl_device int bsdf_microfacet_ggx_aniso_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
-	sc->data1 = clamp(sc->data1, 0.0f, 1.0f); /* alpha_y */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
+	sc->data1 = saturate(sc->data1); /* alpha_y */
 	
 	sc->type = CLOSURE_BSDF_MICROFACET_GGX_ANISO_ID;
 
@@ -255,7 +255,7 @@ ccl_device int bsdf_microfacet_ggx_aniso_setup(ShaderClosure *sc)
 
 ccl_device int bsdf_microfacet_ggx_refraction_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
 	sc->data1 = sc->data0; /* alpha_y */
 
 	sc->type = CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID;
@@ -588,7 +588,7 @@ ccl_device int bsdf_microfacet_ggx_sample(KernelGlobals *kg, const ShaderClosure
 
 ccl_device int bsdf_microfacet_beckmann_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
 	sc->data1 = sc->data0; /* alpha_y */
 
 	sc->type = CLOSURE_BSDF_MICROFACET_BECKMANN_ID;
@@ -597,8 +597,8 @@ ccl_device int bsdf_microfacet_beckmann_setup(ShaderClosure *sc)
 
 ccl_device int bsdf_microfacet_beckmann_aniso_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
-	sc->data1 = clamp(sc->data1, 0.0f, 1.0f); /* alpha_y */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
+	sc->data1 = saturate(sc->data1); /* alpha_y */
 
 	sc->type = CLOSURE_BSDF_MICROFACET_BECKMANN_ANISO_ID;
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
@@ -606,7 +606,7 @@ ccl_device int bsdf_microfacet_beckmann_aniso_setup(ShaderClosure *sc)
 
 ccl_device int bsdf_microfacet_beckmann_refraction_setup(ShaderClosure *sc)
 {
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f); /* alpha_x */
+	sc->data0 = saturate(sc->data0); /* alpha_x */
 	sc->data1 = sc->data0; /* alpha_y */
 
 	sc->type = CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID;
diff --git a/intern/cycles/kernel/closure/bsdf_oren_nayar.h b/intern/cycles/kernel/closure/bsdf_oren_nayar.h
index c476d4ca4e2..61b7cb11b02 100644
--- a/intern/cycles/kernel/closure/bsdf_oren_nayar.h
+++ b/intern/cycles/kernel/closure/bsdf_oren_nayar.h
@@ -37,7 +37,7 @@ ccl_device int bsdf_oren_nayar_setup(ShaderClosure *sc)
 
 	sc->type = CLOSURE_BSDF_OREN_NAYAR_ID;
 
-	sigma = clamp(sigma, 0.0f, 1.0f);
+	sigma = saturate(sigma);
 
 	float div = 1.0f / (M_PI_F + ((3.0f * M_PI_F - 4.0f) / 6.0f) * sigma);
 
diff --git a/intern/cycles/kernel/closure/bsdf_toon.h b/intern/cycles/kernel/closure/bsdf_toon.h
index df03942638f..e5b6ab93a64 100644
--- a/intern/cycles/kernel/closure/bsdf_toon.h
+++ b/intern/cycles/kernel/closure/bsdf_toon.h
@@ -40,8 +40,8 @@ CCL_NAMESPACE_BEGIN
 ccl_device int bsdf_diffuse_toon_setup(ShaderClosure *sc)
 {
 	sc->type = CLOSURE_BSDF_DIFFUSE_TOON_ID;
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f);
-	sc->data1 = clamp(sc->data1, 0.0f, 1.0f);
+	sc->data0 = saturate(sc->data0);
+	sc->data1 = saturate(sc->data1);
 
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
@@ -120,8 +120,8 @@ ccl_device int bsdf_diffuse_toon_sample(const ShaderClosure *sc, float3 Ng, floa
 ccl_device int bsdf_glossy_toon_setup(ShaderClosure *sc)
 {
 	sc->type = CLOSURE_BSDF_GLOSSY_TOON_ID;
-	sc->data0 = clamp(sc->data0, 0.0f, 1.0f);
-	sc->data1 = clamp(sc->data1, 0.0f, 1.0f);
+	sc->data0 = saturate(sc->data0);
+	sc->data1 = saturate(sc->data1);
 
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
diff --git a/intern/cycles/kernel/closure/bssrdf.h b/intern/cycles/kernel/closure/bssrdf.h
index b2848262001..f817dcd5f2d 100644
--- a/intern/cycles/kernel/closure/bssrdf.h
+++ b/intern/cycles/kernel/closure/bssrdf.h
@@ -30,8 +30,8 @@ ccl_device int bssrdf_setup(ShaderClosure *sc, ClosureType type)
 		return flag;
 	}
 	else {
-		sc->data1 = clamp(sc->data1, 0.0f, 1.0f); /* texture blur */
-		sc->T.x = clamp(sc->T.x, 0.0f, 1.0f); /* sharpness */
+		sc->data1 = saturate(sc->data1); /* texture blur */
+		sc->T.x = saturate(sc->T.x); /* sharpness */
 		sc->type = type;
 
 		return SD_BSDF|SD_BSDF_HAS_EVAL|SD_BSSRDF;
@@ -168,7 +168,7 @@ ccl_device float bssrdf_cubic_quintic_root_find(float xi)
 		if(fabsf(f) < tolerance || f_ == 0.0f)
 			break;
 
-		x = clamp(x - f/f_, 0.0f, 1.0f);
+		x = saturate(x - f/f_);
 	}
 
 	return x;