psychopath/src/accel/objects_split.rs

342 lines
10 KiB
Rust

#![allow(dead_code)]
use std::cmp::Ordering;
use crate::{
algorithm::{partition, quick_select},
bbox::BBox,
lerp::lerp_slice,
math::{dot, Vector},
sampling::uniform_sample_hemisphere,
};
const SAH_BIN_COUNT: usize = 13; // Prime numbers work best, for some reason
const SPLIT_PLANE_COUNT: usize = 5;
/// Takes a slice of boundable objects and partitions them based on the Surface
/// Area Heuristic, but using arbitrarily oriented planes.
///
/// Returns the index of the partition boundary and the axis that it split on
/// (0 = x, 1 = y, 2 = z).
pub fn free_sah_split<'a, T, F>(seed: u32, objects: &mut [T], bounder: &F) -> (usize, usize)
where
F: Fn(&T) -> &'a [BBox],
{
// Generate the planes for splitting
let planes = {
let mut planes = [Vector::new(0.0, 0.0, 0.0); SPLIT_PLANE_COUNT];
let offset = seed * SPLIT_PLANE_COUNT as u32;
for i in 0..SPLIT_PLANE_COUNT {
let u = halton::sample(0, offset + i as u32);
let v = halton::sample(1, offset + i as u32);
planes[i] = uniform_sample_hemisphere(u, v).normalized();
}
planes
};
// Get the extents of the objects with respect to the split planes
let extents = {
let mut extents = [(std::f32::INFINITY, std::f32::NEG_INFINITY); SPLIT_PLANE_COUNT];
for obj in &objects[..] {
let centroid = lerp_slice(bounder(obj), 0.5).center().into_vector();
for i in 0..SPLIT_PLANE_COUNT {
let dist = dot(centroid, planes[i]);
extents[i].0 = extents[i].0.min(dist);
extents[i].1 = extents[i].1.max(dist);
}
}
extents
};
// Pre-calc SAH div distances
let sah_divs = {
let mut sah_divs = [[0.0f32; SAH_BIN_COUNT - 1]; SPLIT_PLANE_COUNT];
for pi in 0..SPLIT_PLANE_COUNT {
let extent = extents[pi].1 - extents[pi].0;
for div in 0..(SAH_BIN_COUNT - 1) {
let part = extent * ((div + 1) as f32 / SAH_BIN_COUNT as f32);
sah_divs[pi][div] = extents[pi].0 + part;
}
}
sah_divs
};
// Build SAH bins
let sah_bins = {
let mut sah_bins =
[[(BBox::new(), BBox::new(), 0, 0); SAH_BIN_COUNT - 1]; SPLIT_PLANE_COUNT];
for obj in objects.iter() {
let tb = lerp_slice(bounder(obj), 0.5);
let centroid = tb.center().into_vector();
for pi in 0..SPLIT_PLANE_COUNT {
for div in 0..(SAH_BIN_COUNT - 1) {
let dist = dot(centroid, planes[pi]);
if dist <= sah_divs[pi][div] {
sah_bins[pi][div].0 |= tb;
sah_bins[pi][div].2 += 1;
} else {
sah_bins[pi][div].1 |= tb;
sah_bins[pi][div].3 += 1;
}
}
}
}
sah_bins
};
// Find best split axis and div point
let (split_plane_i, div_n) = {
let mut split_plane_i = 0;
let mut div_n = 0.0;
let mut smallest_cost = std::f32::INFINITY;
for pi in 0..SPLIT_PLANE_COUNT {
for div in 0..(SAH_BIN_COUNT - 1) {
let left_cost = sah_bins[pi][div].0.surface_area() * sah_bins[pi][div].2 as f32;
let right_cost = sah_bins[pi][div].1.surface_area() * sah_bins[pi][div].3 as f32;
let tot_cost = left_cost + right_cost;
if tot_cost < smallest_cost {
split_plane_i = pi;
div_n = sah_divs[pi][div];
smallest_cost = tot_cost;
}
}
}
(split_plane_i, div_n)
};
// Calculate the approximate axis-aligned split, along with flipping the split plane as
// appropriate.
let (plane, approx_axis, div) = {
// Find axis with largest value
let mut largest_axis = 0;
let mut n = 0.0;
for d in 0..3 {
let m = planes[split_plane_i].get_n(d).abs();
if n < m {
largest_axis = d;
n = m;
}
}
// If it's negative, flip
if planes[split_plane_i].get_n(largest_axis).is_sign_positive() {
(planes[split_plane_i], largest_axis, div_n)
} else {
(planes[split_plane_i] * -1.0, largest_axis, div_n * -1.0)
}
};
// Partition
let mut split_i = partition(&mut objects[..], |obj| {
let centroid = lerp_slice(bounder(obj), 0.5).center().into_vector();
let dist = dot(centroid, plane);
dist < div
});
if split_i < 1 {
split_i = 1;
} else if split_i >= objects.len() {
split_i = objects.len() - 1;
}
(split_i, approx_axis)
}
/// Takes a slice of boundable objects and partitions them based on the Surface
/// Area Heuristic.
///
/// Returns the index of the partition boundary and the axis that it split on
/// (0 = x, 1 = y, 2 = z).
pub fn sah_split<'a, T, F>(objects: &mut [T], bounder: &F) -> (usize, usize)
where
F: Fn(&T) -> &'a [BBox],
{
// Get combined object centroid extents
let bounds = {
let mut bb = BBox::new();
for obj in &objects[..] {
bb |= lerp_slice(bounder(obj), 0.5).center();
}
bb
};
// Pre-calc SAH div points
let sah_divs = {
let mut sah_divs = [[0.0f32; SAH_BIN_COUNT - 1]; 3];
for d in 0..sah_divs.len() {
let extent = bounds.max.get_n(d) - bounds.min.get_n(d);
for div in 0..(SAH_BIN_COUNT - 1) {
let part = extent * ((div + 1) as f32 / SAH_BIN_COUNT as f32);
sah_divs[d][div] = bounds.min.get_n(d) + part;
}
}
sah_divs
};
// Build SAH bins
let sah_bins = {
let mut sah_bins = [[(BBox::new(), BBox::new(), 0, 0); SAH_BIN_COUNT - 1]; 3];
for obj in objects.iter() {
let tb = lerp_slice(bounder(obj), 0.5);
let centroid = (tb.min.into_vector() + tb.max.into_vector()) * 0.5;
for d in 0..3 {
for div in 0..(SAH_BIN_COUNT - 1) {
if centroid.get_n(d) <= sah_divs[d][div] {
sah_bins[d][div].0 |= tb;
sah_bins[d][div].2 += 1;
} else {
sah_bins[d][div].1 |= tb;
sah_bins[d][div].3 += 1;
}
}
}
}
sah_bins
};
// Find best split axis and div point
let (split_axis, div) = {
let mut dim = 0;
let mut div_n = 0.0;
let mut smallest_cost = std::f32::INFINITY;
for d in 0..3 {
for div in 0..(SAH_BIN_COUNT - 1) {
let left_cost = sah_bins[d][div].0.surface_area() * sah_bins[d][div].2 as f32;
let right_cost = sah_bins[d][div].1.surface_area() * sah_bins[d][div].3 as f32;
let left_diag = sah_bins[d][div].0.diagonal();
let right_diag = sah_bins[d][div].1.diagonal();
let tot_cost = (left_cost * left_diag) + (right_cost * right_diag);
if tot_cost < smallest_cost {
dim = d;
div_n = sah_divs[d][div];
smallest_cost = tot_cost;
}
}
}
(dim, div_n)
};
// Partition
let mut split_i = partition(&mut objects[..], |obj| {
let tb = lerp_slice(bounder(obj), 0.5);
let centroid = (tb.min.get_n(split_axis) + tb.max.get_n(split_axis)) * 0.5;
centroid < div
});
if split_i < 1 {
split_i = 1;
} else if split_i >= objects.len() {
split_i = objects.len() - 1;
}
(split_i, split_axis)
}
/// Takes a slice of boundable objects and partitions them based on the bounds mean heuristic.
///
/// Returns the index of the partition boundary and the axis that it split on
/// (0 = x, 1 = y, 2 = z).
pub fn bounds_mean_split<'a, T, F>(objects: &mut [T], bounder: &F) -> (usize, usize)
where
F: Fn(&T) -> &'a [BBox],
{
// Get combined object bounds
let bounds = {
let mut bb = BBox::new();
for obj in &objects[..] {
bb |= lerp_slice(bounder(obj), 0.5);
}
bb
};
let split_axis = {
let mut axis = 0;
let mut largest = std::f32::NEG_INFINITY;
for i in 0..3 {
let extent = bounds.max.get_n(i) - bounds.min.get_n(i);
if extent > largest {
largest = extent;
axis = i;
}
}
axis
};
let div = (bounds.min.get_n(split_axis) + bounds.max.get_n(split_axis)) * 0.5;
// Partition
let mut split_i = partition(&mut objects[..], |obj| {
let tb = lerp_slice(bounder(obj), 0.5);
let centroid = (tb.min.get_n(split_axis) + tb.max.get_n(split_axis)) * 0.5;
centroid < div
});
if split_i < 1 {
split_i = 1;
} else if split_i >= objects.len() {
split_i = objects.len() - 1;
}
(split_i, split_axis)
}
/// Takes a slice of boundable objects and partitions them based on the median heuristic.
///
/// Returns the index of the partition boundary and the axis that it split on
/// (0 = x, 1 = y, 2 = z).
pub fn median_split<'a, T, F>(objects: &mut [T], bounder: &F) -> (usize, usize)
where
F: Fn(&T) -> &'a [BBox],
{
// Get combined object bounds
let bounds = {
let mut bb = BBox::new();
for obj in &objects[..] {
bb |= lerp_slice(bounder(obj), 0.5);
}
bb
};
let split_axis = {
let mut axis = 0;
let mut largest = std::f32::NEG_INFINITY;
for i in 0..3 {
let extent = bounds.max.get_n(i) - bounds.min.get_n(i);
if extent > largest {
largest = extent;
axis = i;
}
}
axis
};
let place = {
let place = objects.len() / 2;
if place > 0 {
place
} else {
1
}
};
quick_select(objects, place, |a, b| {
let tb_a = lerp_slice(bounder(a), 0.5);
let tb_b = lerp_slice(bounder(b), 0.5);
let centroid_a = (tb_a.min.get_n(split_axis) + tb_a.max.get_n(split_axis)) * 0.5;
let centroid_b = (tb_b.min.get_n(split_axis) + tb_b.max.get_n(split_axis)) * 0.5;
if centroid_a < centroid_b {
Ordering::Less
} else if centroid_a == centroid_b {
Ordering::Equal
} else {
Ordering::Greater
}
});
(place, split_axis)
}