Cycles: Implement extrapolation for RGB curves

Previously RGB Curves node will clamp input to 0..1 which is rather useless
when one wants to use HDR image textures and do bit of correction on them.

Now kernel code supports extrapolation of baked LUT based on first/last two
table points and performs linear extrapolation.

The only tricky part is to guess the range to bake the LUT for. Currently
it's using simple approach -- minmax of the input curves. While this behaves
ok for the simple cases it's easy to trick the system up causing incorrect
results.

Not sure we can solve those issues in a general case and since the new code
is giving more expected results it's not that bad actually. In the worst
case artist migh always create explicit point to make sure LUT is created
for the needed HDR range.

Reviewers: brecht, juicyfruit

Subscribers: sebastian_k

Differential Revision: https://developer.blender.org/D1658

6 files changed, 125 insertions, 19 deletions

diff --git a/intern/cycles/blender/blender_shader.cpp b/intern/cycles/blender/blender_shader.cpp
index 38e019eb0c1..74dee206c0e 100644
--- a/intern/cycles/blender/blender_shader.cpp
+++ b/intern/cycles/blender/blender_shader.cpp
@@ -220,8 +220,13 @@ static ShaderNode *add_node(Scene *scene,
 	/* existing blender nodes */
 	if(b_node.is_a(&RNA_ShaderNodeRGBCurve)) {
 		BL::ShaderNodeRGBCurve b_curve_node(b_node);
+		BL::CurveMapping mapping(b_curve_node.mapping());
 		RGBCurvesNode *curves = new RGBCurvesNode();
-		curvemapping_color_to_array(b_curve_node.mapping(), curves->curves, RAMP_TABLE_SIZE, true);
+		curvemapping_color_to_array(mapping,
+		                            curves->curves,
+		                            RAMP_TABLE_SIZE,
+		                            true);
+		curvemapping_minmax(mapping, true, &curves->min_x, &curves->max_x);
 		node = curves;
 	}
 	if(b_node.is_a(&RNA_ShaderNodeVectorCurve)) {
diff --git a/intern/cycles/blender/blender_util.h b/intern/cycles/blender/blender_util.h
index 2e9e9266404..029d0af0fd2 100644
--- a/intern/cycles/blender/blender_util.h
+++ b/intern/cycles/blender/blender_util.h
@@ -62,6 +62,29 @@ static inline void colorramp_to_array(BL::ColorRamp ramp, float4 *data, int size
 	}
 }
 
+static inline void curvemap_minmax_curve(/*const*/ BL::CurveMap& curve,
+                                         float *min_x,
+                                         float *max_x)
+{
+	*min_x = min(*min_x, curve.points[0].location()[0]);
+	*max_x = max(*max_x, curve.points[curve.points.length() - 1].location()[0]);
+}
+
+static inline void curvemapping_minmax(/*const*/ BL::CurveMapping& cumap,
+                                       bool rgb_curve,
+                                       float *min_x,
+                                       float *max_x)
+{
+	/* const int num_curves = cumap.curves.length(); */  /* Gives linking error so far. */
+	const int num_curves = rgb_curve? 4: 3;
+	*min_x = FLT_MAX;
+	*max_x = -FLT_MAX;
+	for(int i = 0; i < num_curves; ++i) {
+		BL::CurveMap map(cumap.curves[i]);
+		curvemap_minmax_curve(map, min_x, max_x);
+	}
+}
+
 static inline void curvemapping_to_array(BL::CurveMapping cumap, float *data, int size)
 {
 	cumap.update();
@@ -72,8 +95,29 @@ static inline void curvemapping_to_array(BL::CurveMapping cumap, float *data, in
 	}
 }
 
-static inline void curvemapping_color_to_array(BL::CurveMapping cumap, float4 *data, int size, bool rgb_curve)
+static inline void curvemapping_color_to_array(BL::CurveMapping cumap,
+                                               float4 *data,
+                                               int size,
+                                               bool rgb_curve)
 {
+	float min_x = 0.0f, max_x = 1.0f;
+	/* TODO(sergey): Vector curve mapping is still clipping to 0..1. */
+	if(rgb_curve) {
+		/* TODO(sergey): There is no easy way to automatically guess what is
+		 * the range to be used here for the case when mapping is applied on
+		 * top of another mapping (i.e. R curve applied on top of common
+		 * one).
+		 *
+		 * Using largest possible range form all curves works correct for the
+		 * cases like vector curves and should be good enough heuristic for
+		 * the color curves as well.
+		 *
+		 * There might be some better estimations here tho.
+		 */
+		curvemapping_minmax(cumap, rgb_curve, &min_x, &max_x);
+	}
+	const float range_x = max_x - min_x;
+
 	cumap.update();
 
 	BL::CurveMap mapR = cumap.curves[0];
@@ -84,7 +128,7 @@ static inline void curvemapping_color_to_array(BL::CurveMapping cumap, float4 *d
 		BL::CurveMap mapI = cumap.curves[3];
 
 		for(int i = 0; i < size; i++) {
-			float t = (float)i/(float)(size-1);
+			float t = min_x + (float)i/(float)(size-1) * range_x;
 
 			data[i][0] = mapR.evaluate(mapI.evaluate(t));
 			data[i][1] = mapG.evaluate(mapI.evaluate(t));
@@ -93,7 +137,7 @@ static inline void curvemapping_color_to_array(BL::CurveMapping cumap, float4 *d
 	}
 	else {
 		for(int i = 0; i < size; i++) {
-			float t = (float)i/(float)(size-1);
+			float t = min_x + (float)i/(float)(size-1) * range_x;
 
 			data[i][0] = mapR.evaluate(t);
 			data[i][1] = mapG.evaluate(t);
diff --git a/intern/cycles/kernel/shaders/node_rgb_curves.osl b/intern/cycles/kernel/shaders/node_rgb_curves.osl
index 60cb273ba98..0427c001f6d 100644
--- a/intern/cycles/kernel/shaders/node_rgb_curves.osl
+++ b/intern/cycles/kernel/shaders/node_rgb_curves.osl
@@ -19,6 +19,21 @@
 
 float ramp_lookup(color ramp[RAMP_TABLE_SIZE], float at, int component)
 {
+	if (at < 0.0 || at > 1.0) {
+		float t0, dy;
+		if(at < 0.0) {
+			t0 = ramp[0][component];
+			dy = t0 - ramp[1][component];
+			at = -at;
+		}
+		else {
+			t0 = ramp[RAMP_TABLE_SIZE - 1][component];
+			dy = t0 - ramp[RAMP_TABLE_SIZE - 2][component];
+			at = at - 1.0;
+		}
+		return t0 + dy * at * (RAMP_TABLE_SIZE - 1);
+	}
+
 	float f = clamp(at, 0.0, 1.0) * (RAMP_TABLE_SIZE - 1);
 
 	/* clamp int as well in case of NaN */
@@ -37,14 +52,18 @@ float ramp_lookup(color ramp[RAMP_TABLE_SIZE], float at, int component)
 
 shader node_rgb_curves(
 	color ramp[RAMP_TABLE_SIZE] = {0.0},
+	float min_x = 0.0,
+	float max_x = 1.0,
 
 	color ColorIn = 0.0,
 	float Fac = 0.0,
 	output color ColorOut = 0.0)
 {
-	ColorOut[0] = ramp_lookup(ramp, ColorIn[0], 0);
-	ColorOut[1] = ramp_lookup(ramp, ColorIn[1], 1);
-	ColorOut[2] = ramp_lookup(ramp, ColorIn[2], 2);
+	color c = (ColorIn - color(min_x, min_x, min_x)) / (max_x - min_x);
+
+	ColorOut[0] = ramp_lookup(ramp, c[0], 0);
+	ColorOut[1] = ramp_lookup(ramp, c[1], 1);
+	ColorOut[2] = ramp_lookup(ramp, c[2], 2);
 
 	ColorOut = mix(ColorIn, ColorOut, Fac);
 }
diff --git a/intern/cycles/kernel/svm/svm_ramp.h b/intern/cycles/kernel/svm/svm_ramp.h
index 062ab013b1f..48d969947f5 100644
--- a/intern/cycles/kernel/svm/svm_ramp.h
+++ b/intern/cycles/kernel/svm/svm_ramp.h
@@ -19,8 +19,27 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device float4 rgb_ramp_lookup(KernelGlobals *kg, int offset, float f, bool interpolate)
+ccl_device float4 rgb_ramp_lookup(KernelGlobals *kg,
+                                  int offset,
+                                  float f,
+                                  bool interpolate,
+                                  bool extrapolate)
 {
+	if((f < 0.0f || f > 1.0f) && extrapolate) {
+		float4 t0, dy;
+		if(f < 0.0f) {
+			t0 = fetch_node_float(kg, offset);
+			dy = t0 - fetch_node_float(kg, offset + 1);
+			f = -f;
+		}
+		else {
+			t0 = fetch_node_float(kg, offset + RAMP_TABLE_SIZE - 1);
+			dy = t0 - fetch_node_float(kg, offset + RAMP_TABLE_SIZE - 2);
+			f = f - 1.0f;
+		}
+		return t0 + dy * f * (RAMP_TABLE_SIZE-1);
+	}
+
 	f = saturate(f)*(RAMP_TABLE_SIZE-1);
 
 	/* clamp int as well in case of NaN */
@@ -43,7 +62,7 @@ ccl_device void svm_node_rgb_ramp(KernelGlobals *kg, ShaderData *sd, float *stac
 	decode_node_uchar4(node.y, &fac_offset, &color_offset, &alpha_offset, NULL);
 
 	float fac = stack_load_float(stack, fac_offset);
-	float4 color = rgb_ramp_lookup(kg, *offset, fac, interpolate);
+	float4 color = rgb_ramp_lookup(kg, *offset, fac, interpolate, false);
 
 	if(stack_valid(color_offset))
 		stack_store_float3(stack, color_offset, float4_to_float3(color));
@@ -55,16 +74,24 @@ ccl_device void svm_node_rgb_ramp(KernelGlobals *kg, ShaderData *sd, float *stac
 
 ccl_device void svm_node_rgb_curves(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, int *offset)
 {
-	uint fac_offset = node.y;
-	uint color_offset = node.z;
-	uint out_offset = node.w;
+	uint fac_offset, color_offset, out_offset;
+	decode_node_uchar4(node.y,
+	                   &fac_offset,
+	                   &color_offset,
+	                   &out_offset,
+	                   NULL);
 
 	float fac = stack_load_float(stack, fac_offset);
 	float3 color = stack_load_float3(stack, color_offset);
 
-	float r = rgb_ramp_lookup(kg, *offset, color.x, true).x;
-	float g = rgb_ramp_lookup(kg, *offset, color.y, true).y;
-	float b = rgb_ramp_lookup(kg, *offset, color.z, true).z;
+	const float min_x = __int_as_float(node.z),
+	            max_x = __int_as_float(node.w);
+	const float range_x = max_x - min_x;
+	color = (color - make_float3(min_x, min_x, min_x)) / range_x;
+
+	float r = rgb_ramp_lookup(kg, *offset, color.x, true, true).x;
+	float g = rgb_ramp_lookup(kg, *offset, color.y, true, true).y;
+	float b = rgb_ramp_lookup(kg, *offset, color.z, true, true).z;
 
 	color = (1.0f - fac)*color + fac*make_float3(r, g, b);
 	stack_store_float3(stack, out_offset, color);
@@ -81,9 +108,9 @@ ccl_device void svm_node_vector_curves(KernelGlobals *kg, ShaderData *sd, float
 	float fac = stack_load_float(stack, fac_offset);
 	float3 color = stack_load_float3(stack, color_offset);
 
-	float r = rgb_ramp_lookup(kg, *offset, (color.x + 1.0f)*0.5f, true).x;
-	float g = rgb_ramp_lookup(kg, *offset, (color.y + 1.0f)*0.5f, true).y;
-	float b = rgb_ramp_lookup(kg, *offset, (color.z + 1.0f)*0.5f, true).z;
+	float r = rgb_ramp_lookup(kg, *offset, (color.x + 1.0f)*0.5f, true, false).x;
+	float g = rgb_ramp_lookup(kg, *offset, (color.y + 1.0f)*0.5f, true, false).y;
+	float b = rgb_ramp_lookup(kg, *offset, (color.z + 1.0f)*0.5f, true, false).z;
 
 	color = (1.0f - fac)*color + fac*make_float3(r*2.0f - 1.0f, g*2.0f - 1.0f, b*2.0f - 1.0f);
 	stack_store_float3(stack, out_offset, color);
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index cece0c37f59..fd8410ce71b 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -4316,6 +4316,9 @@ RGBCurvesNode::RGBCurvesNode()
 	add_input("Fac", SHADER_SOCKET_FLOAT);
 	add_input("Color", SHADER_SOCKET_COLOR);
 	add_output("Color", SHADER_SOCKET_COLOR);
+
+	min_x = 0.0f;
+	max_x = 1.0f;
 }
 
 void RGBCurvesNode::compile(SVMCompiler& compiler)
@@ -4328,7 +4331,12 @@ void RGBCurvesNode::compile(SVMCompiler& compiler)
 	compiler.stack_assign(color_in);
 	compiler.stack_assign(color_out);
 
-	compiler.add_node(NODE_RGB_CURVES, fac_in->stack_offset, color_in->stack_offset, color_out->stack_offset);
+	compiler.add_node(NODE_RGB_CURVES,
+	                  compiler.encode_uchar4(fac_in->stack_offset,
+	                                         color_in->stack_offset,
+	                                         color_out->stack_offset),
+	                  __float_as_int(min_x),
+	                  __float_as_int(max_x));
 	compiler.add_array(curves, RAMP_TABLE_SIZE);
 }
 
@@ -4343,6 +4351,8 @@ void RGBCurvesNode::compile(OSLCompiler& compiler)
 	}
 
 	compiler.parameter_color_array("ramp", ramp, RAMP_TABLE_SIZE);
+	compiler.parameter("min_x", min_x);
+	compiler.parameter("max_x", max_x);
 	compiler.add(this, "node_rgb_curves");
 }
 
diff --git a/intern/cycles/render/nodes.h b/intern/cycles/render/nodes.h
index 259936c0b7c..1fa4b67c6b9 100644
--- a/intern/cycles/render/nodes.h
+++ b/intern/cycles/render/nodes.h
@@ -703,6 +703,7 @@ public:
 	virtual int get_group() { return NODE_GROUP_LEVEL_3; }
 
 	float4 curves[RAMP_TABLE_SIZE];
+	float min_x, max_x;
 };
 
 class VectorCurvesNode : public ShaderNode {