Cycles: Implement Mitchell-Netravali pixel filter

It's the same filter which is used by default by Blender Internal renderer
and it gives crispier edges than gaussian filter.

Default filter for Cycles is unchanged because it's unclear if new filter
gives more noise or not. After some further real production tests we can
consider making Mitchell filter default for Cycles as well.

1 files changed, 26 insertions, 0 deletions

diff --git a/intern/cycles/render/film.cpp b/intern/cycles/render/film.cpp
index 19f959d4ea1..76a92d3d704 100644
--- a/intern/cycles/render/film.cpp
+++ b/intern/cycles/render/film.cpp
@@ -198,6 +198,29 @@ static float filter_func_gaussian(float v, float width)
 	return expf(-2.0f*v*v);
 }
 
+static float filter_func_mitchell(float v, float width)
+{
+	(void)width;  /* Ignored. */
+
+	/* B = 1,   C = 0   - cubic B-spline */
+	/* B = 1/3, C = 1/3 - recommended */
+	/* B = 0,   C = 1/2 - Catmull-Rom spline */
+	const float B = 1.0f / 3.0f;
+	const float C = 1.0f / 3.0f;
+
+	const float ax = fabsf(v);
+	if (ax < 1.0f) {
+		return ((12.0f - 9.0f * B - 6 * C) * ax * ax * ax +
+		        (-18.0f + 12.0f * B + 6.0f * C) * ax * ax +
+		        (6.0f - 2.0f * B)) / 6.0f;
+	} else if ((ax >= 1.0f) && (ax < 2.0f)) {
+		return ((-B - 6.0f * C) * ax * ax * ax +
+		        (6.0f * B + 30.0f * C) * ax * ax + (-12.0f * B - 48.0f * C) *
+		        ax + (8.0f * B + 24.0f * C)) / 6.0f;
+	}
+	return 0.0f;
+}
+
 static vector<float> filter_table(FilterType type, float width)
 {
 	const int filter_table_size = FILTER_TABLE_SIZE-1;
@@ -213,6 +236,9 @@ static vector<float> filter_table(FilterType type, float width)
 		case FILTER_GAUSSIAN:
 			filter_func = filter_func_gaussian;
 			break;
+		case FILTER_MITCHELL:
+			filter_func = filter_func_mitchell;
+			break;
 		default:
 			assert(0);
 	}