Cycles: experimental correlated multi-jittered sampling pattern that can be used

instead of sobol. So far one doesn't seem to be consistently better or worse than
the other for the same number of samples but more testing is needed.

The random number generator itself is slower than sobol for most number of samples,
except 16, 64, 256, .. because they can be computed faster. This can probably be
optimized, but we can do that when/if this actually turns out to be useful.

Paper this implementation is based on:
http://graphics.pixar.com/library/MultiJitteredSampling/

Also includes some refactoring of RNG code, fixing a Sobol correlation issue with
the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a
simpler unit square to unit disk function.

1 files changed, 4 insertions, 28 deletions

diff --git a/intern/cycles/kernel/kernel_montecarlo.h b/intern/cycles/kernel/kernel_montecarlo.h
index 2ae95084162..f608429da36 100644
--- a/intern/cycles/kernel/kernel_montecarlo.h
+++ b/intern/cycles/kernel/kernel_montecarlo.h
@@ -36,36 +36,12 @@
 CCL_NAMESPACE_BEGIN
 
 /// Given values x and y on [0,1], convert them in place to values on
-/// [-1,1] uniformly distributed over a unit sphere.  This code is
-/// derived from Peter Shirley, "Realistic Ray Tracing", p. 103.
+/// [-1,1] uniformly distributed over a unit sphere.
 __device void to_unit_disk(float *x, float *y)
 {
-	float r, phi;
-	float a = 2.0f * (*x) - 1.0f;
-	float b = 2.0f * (*y) - 1.0f;
-	if(a > -b) {
-		if(a > b) {
-			r = a;
-			phi = M_PI_4_F *(b/a);
-		}
-		else {
-			r = b;
-			phi = M_PI_4_F *(2.0f - a/b);
-		}
-	}
-	else {
-		if(a < b) {
-			r = -a;
-			phi = M_PI_4_F *(4.0f + b/a);
-		}
-		else {
-			r = -b;
-			if(b != 0.0f)
-				phi = M_PI_4_F *(6.0f - a/b);
-			else
-				phi = 0.0f;
-		}
-	}
+	float phi = 2.0f * M_PI_F * (*x);
+	float r = sqrtf(*y);
+
 	*x = r * cosf(phi);
 	*y = r * sinf(phi);
 }