cessen/psychopath
1
0
Fork 0
Code Issues 1 Pull Requests Projects Releases Wiki Activity

Use golden ratio sampling for wavelength again.

Browse Source 
Now that the Sobol sequence can be decorrelated, this doesn't
introduce sampling correlation problems anymore, so it's mostly
just a win.
This commit is contained in:
Nathan Vegdahl 2020-04-23 14:46:21 +09:00
parent 3ffaf7bee5
commit 2c68e18bd6
1 changed files with 21 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
src/renderer.rs
View File

@ -15,6 +15,7 @@ use crate::{
accel::ACCEL_NODE_RAY_TESTS,
color::{map_0_1_to_wavelength, SpectralSample, XYZ},
fp_utils::robust_ray_origin,
hash::hash_u32,
hilbert,
image::Image,
math::{probit, upper_power_of_two},
@ -243,10 +244,10 @@ impl<'a> Renderer<'a> {
// Calculate image plane x and y coordinates
let (img_x, img_y) = {
let filter_x =
probit(get_sample(4, si as u32, (x, y), self.seed), 2.0 / 6.0)
probit(get_sample(3, si as u32, (x, y), self.seed), 2.0 / 6.0)
+ 0.5;
let filter_y =
probit(get_sample(5, si as u32, (x, y), self.seed), 2.0 / 6.0)
probit(get_sample(4, si as u32, (x, y), self.seed), 2.0 / 6.0)
+ 0.5;
let samp_x = (filter_x + x as f32) * cmpx;
let samp_y = (filter_y + y as f32) * cmpy;
@ -260,11 +261,14 @@ impl<'a> Renderer<'a> {
(x, y),
(img_x, img_y),
(
get_sample(1, si as u32, (x, y), self.seed),
get_sample(2, si as u32, (x, y), self.seed),
get_sample(3, si as u32, (x, y), self.seed),
),
get_sample(1, si as u32, (x, y), self.seed),
map_0_1_to_wavelength(get_sample(0, si as u32, (x, y), self.seed)),
get_sample(0, si as u32, (x, y), self.seed),
map_0_1_to_wavelength(golden_ratio_sample(
si as u32,
hash_u32((x << 16) ^ y, self.seed),
)),
si as u32,
);
paths.push(path);
@ -717,6 +721,18 @@ fn get_sample(dimension: u32, i: u32, pixel_co: (u32, u32), seed: u32) -> f32 {
}
}
/// Golden ratio sampler.
fn golden_ratio_sample(i: u32, scramble: u32) -> f32 {
// NOTE: use this for the wavelength dimension, because
// due to the nature of hero wavelength sampling this ends up
// being crazily more efficient than pretty much any other sampler,
// and reduces variance by a huge amount.
let n = i
.wrapping_add(hash_u32(scramble, 0))
.wrapping_mul(2654435769);
n as f32 * (1.0 / (1u64 << 32) as f32)
}
#[derive(Debug)]
struct BucketJob {
x: u32,