Cycles: Use curve approximation for blackbody instead of lookup table

Now we calculate color in range 800..12000 using an approximation a/x+bx+c for R and G and ((at + b)t + c)t + d) for B.
Max absolute error for RGB for non-lut function is less than 0.0001, which is enough to get the same 8 bit/channel color as for OSL with a noticeable performance difference.
However there is a slight visible difference between previous non-OSL implementation because of lookup table interpolation and offset-by-one mistake.
The previous implementation gave black color outside of soft range (t > 12000), now it gives the same color as for 12000.

Also blackbody node without input connected is being converted to value input at shader compile time.

Reviewers: dingto, sergey

Reviewed By: dingto

Subscribers: nutel, brecht, juicyfruit

Differential Revision: https://developer.blender.org/D1280

4 files changed, 66 insertions, 46 deletions

diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index b948f7de2f4..0491b8ddec7 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -38,12 +38,6 @@ CCL_NAMESPACE_BEGIN
 #define BSSRDF_MIN_RADIUS			1e-8f
 #define BSSRDF_MAX_HITS				4
 
-#define BB_DRAPER				800.0f
-#define BB_MAX_TABLE_RANGE		12000.0f
-#define BB_TABLE_XPOWER			1.5f
-#define BB_TABLE_YPOWER			5.0f
-#define BB_TABLE_SPACING		2.0f
-
 #define BECKMANN_TABLE_SIZE		256
 
 #define TEX_NUM_FLOAT_IMAGES	5
@@ -989,9 +983,8 @@ typedef struct KernelCurves {
 } KernelCurves;
 
 typedef struct KernelTables {
-	int blackbody_offset;
 	int beckmann_offset;
-	int pad1, pad2;
+	int pad1, pad2, pad3;
 } KernelTables;
 
 typedef struct KernelData {
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index dd9173d92b3..1598019856a 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -142,6 +142,8 @@ CCL_NAMESPACE_END
 #include "svm_noise.h"
 #include "svm_texture.h"
 
+#include "svm_math_util.h"
+
 #include "svm_attribute.h"
 #include "svm_gradient.h"
 #include "svm_blackbody.h"
@@ -164,7 +166,6 @@ CCL_NAMESPACE_END
 #include "svm_mapping.h"
 #include "svm_normal.h"
 #include "svm_wave.h"
-#include "svm_math_util.h"
 #include "svm_math.h"
 #include "svm_mix.h"
 #include "svm_ramp.h"
diff --git a/intern/cycles/kernel/svm/svm_blackbody.h b/intern/cycles/kernel/svm/svm_blackbody.h
index b2ff97d71e6..b750ad87b7f 100644
--- a/intern/cycles/kernel/svm/svm_blackbody.h
+++ b/intern/cycles/kernel/svm/svm_blackbody.h
@@ -36,46 +36,10 @@ CCL_NAMESPACE_BEGIN
 
 ccl_device void svm_node_blackbody(KernelGlobals *kg, ShaderData *sd, float *stack, uint temperature_offset, uint col_offset)
 {
-	/* Output */
-	float3 color_rgb = make_float3(0.0f, 0.0f, 0.0f);
-
 	/* Input */
 	float temperature = stack_load_float(stack, temperature_offset);
 
-	if(temperature < BB_DRAPER) {
-		/* just return very very dim red */
-		color_rgb = make_float3(1.0e-6f,0.0f,0.0f);
-	}
-	else if(temperature <= BB_MAX_TABLE_RANGE) {
-		/* This is the overall size of the table */
-		const int lookuptablesize = 956;
-		const float lookuptablenormalize = 1.0f/956.0f;
-
-		/* reconstruct a proper index for the table lookup, compared to OSL we don't look up two colors
-		just one (the OSL-lerp is also automatically done for us by "lookup_table_read") */
-		float t = powf((temperature - BB_DRAPER) * (1.0f / BB_TABLE_SPACING), (1.0f / BB_TABLE_XPOWER));
-
-		int blackbody_table_offset = kernel_data.tables.blackbody_offset;
-
-		/* Retrieve colors from the lookup table */
-		float lutval = t*lookuptablenormalize;
-		float R = lookup_table_read(kg, lutval, blackbody_table_offset, lookuptablesize);
-		lutval = (t + 319.0f*1.0f)*lookuptablenormalize;
-		float G = lookup_table_read(kg, lutval, blackbody_table_offset, lookuptablesize);
-		lutval = (t + 319.0f*2.0f)*lookuptablenormalize;
-		float B = lookup_table_read(kg, lutval, blackbody_table_offset, lookuptablesize);
-
-		R = powf(R, BB_TABLE_YPOWER);
-		G = powf(G, BB_TABLE_YPOWER);
-		B = powf(B, BB_TABLE_YPOWER);
-
-		color_rgb = make_float3(R, G, B);
-	}
-
-	/* Luminance */
-	float l = linear_rgb_to_gray(color_rgb);
-	if(l != 0.0f)
-		color_rgb /= l;
+	float3 color_rgb = svm_math_blackbody_color(temperature);
 
 	if(stack_valid(col_offset))
 		stack_store_float3(stack, col_offset, color_rgb);
diff --git a/intern/cycles/kernel/svm/svm_math_util.h b/intern/cycles/kernel/svm/svm_math_util.h
index ff9e662e931..645cbd3fc73 100644
--- a/intern/cycles/kernel/svm/svm_math_util.h
+++ b/intern/cycles/kernel/svm/svm_math_util.h
@@ -104,5 +104,67 @@ ccl_device float svm_math(NodeMath type, float Fac1, float Fac2)
 	return Fac;
 }
 
+ccl_device float3 svm_math_blackbody_color(float t) {
+	/* Calculate color in range 800..12000 using an approximation
+	 * a/x+bx+c for R and G and ((at + b)t + c)t + d) for B
+	 * Max absolute error for RGB is (0.00095, 0.00077, 0.00057),
+	 * which is enough to get the same 8 bit/channel color.
+	 */
+
+	const float rc[6][3] = {
+		{  2.52432244e+03f, -1.06185848e-03f, 3.11067539e+00f },
+		{  3.37763626e+03f, -4.34581697e-04f, 1.64843306e+00f },
+		{  4.10671449e+03f, -8.61949938e-05f, 6.41423749e-01f },
+		{  4.66849800e+03f,  2.85655028e-05f, 1.29075375e-01f },
+		{  4.60124770e+03f,  2.89727618e-05f, 1.48001316e-01f },
+		{  3.78765709e+03f,  9.36026367e-06f, 3.98995841e-01f },
+	};
+
+	const float gc[6][3] = {
+		{ -7.50343014e+02f,  3.15679613e-04f, 4.73464526e-01f },
+		{ -1.00402363e+03f,  1.29189794e-04f, 9.08181524e-01f },
+		{ -1.22075471e+03f,  2.56245413e-05f, 1.20753416e+00f },
+		{ -1.42546105e+03f, -4.01730887e-05f, 1.44002695e+00f },
+		{ -1.18134453e+03f, -2.18913373e-05f, 1.30656109e+00f },
+		{ -5.00279505e+02f, -4.59745390e-06f, 1.09090465e+00f },
+	};
+
+	const float bc[6][4] = {
+		{ 0.0f, 0.0f, 0.0f, 0.0f }, /* zeros should be optimized by compiler */
+		{ 0.0f, 0.0f, 0.0f, 0.0f },
+		{ 0.0f, 0.0f, 0.0f, 0.0f },
+		{ -2.02524603e-11f,  1.79435860e-07f, -2.60561875e-04f, -1.41761141e-02f },
+		{ -2.22463426e-13f, -1.55078698e-08f,  3.81675160e-04f, -7.30646033e-01f },
+		{  6.72595954e-13f, -2.73059993e-08f,  4.24068546e-04f, -7.52204323e-01f },
+	};
+
+	if(t >= 12000.0f)
+		return make_float3(0.826270103f, 0.994478524f, 1.56626022f);
+
+	/* Define a macro to reduce stack usage for nvcc */
+#define MAKE_BB_RGB(i) make_float3(\
+		rc[i][0] / t + rc[i][1] * t + rc[i][2],\
+		gc[i][0] / t + gc[i][1] * t + gc[i][2],\
+		((bc[i][0] * t + bc[i][1]) * t + bc[i][2]) * t + bc[i][3])
+
+	if(t >= 6365.0f)
+		return MAKE_BB_RGB(5);
+	if(t >= 3315.0f)
+		return MAKE_BB_RGB(4);
+	if(t >= 1902.0f)
+		return MAKE_BB_RGB(3);
+	if(t >= 1449.0f)
+		return MAKE_BB_RGB(2);
+	if(t >= 1167.0f)
+		return MAKE_BB_RGB(1);
+	if(t >= 965.0f)
+		return MAKE_BB_RGB(0);
+
+#undef MAKE_BB_RGB
+
+	/* For 800 <= t < 965 color does not change in OSL implementation, so keep color the same */
+	return make_float3(4.70366907f, 0.0f, 0.0f);
+}
+
 CCL_NAMESPACE_END