Skip to content

Commit

Permalink
finish frustum and fog culling, branchless checking, start work on bf…
Browse files Browse the repository at this point in the history 
…s (still wip)
  • Loading branch information
@burgerindividual
burgerindividual committed Aug 12, 2023
1 parent de9ec18 commit d97eb63
Show file tree
Hide file tree
Showing 5 changed files with 467 additions and 347 deletions.
2 changes: 1 addition & 1 deletion native/core/Cargo.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ lto = "thin"
[profile.asm]
inherits = "release"
panic = "abort"
debug = true
debug = false
lto = "off"

[profile.production]
Expand Down
27 changes: 0 additions & 27 deletions native/core/src/ffi.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -186,31 +186,4 @@ impl<T, const LEN: usize> CInlineVec<T, LEN> {
// let graph = Box::from_raw(graph.into_mut_ref());
// std::mem::drop(graph);
// }
//
// #[no_mangle]
// pub unsafe extern "C" fn Java_me_jellysquid_mods_sodium_core_CoreLibFFI_frustumCreate(
// _: *mut JEnv,
// _: *mut JClass,
// out_frustum: JPtrMut<*const LocalFrustum>,
// planes: JPtr<[[f32; 4]; 6]>,
// offset: JPtr<[f32; 3]>,
// ) {
// let planes = planes.as_ref().map(f32x3::from_array);
//
// let offset = f32x3::from_array(*offset.as_ref());
//
// let frustum = Box::new(LocalFrustum::new(planes, offset));
//
// let out_frustum = out_frustum.into_mut_ref();
// *out_frustum = Box::into_raw(frustum);
// }
//
// #[no_mangle]
// pub unsafe extern "C" fn Java_me_jellysquid_mods_sodium_core_CoreLibFFI_frustumDelete(
// _: *mut JEnv,
// _: *mut JClass,
// frustum: JPtrMut<LocalFrustum>,
// ) {
// std::mem::drop(Box::from_raw(frustum.into_mut_ref()));
// }
// }
242 changes: 225 additions & 17 deletions native/core/src/graph/local.rs
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
use std::mem::transmute;
use std::ops::Shr;
use std::thread::current;

use core_simd::simd::*;
use std_float::StdFloat;
Expand All @@ -8,6 +9,212 @@ use crate::graph::octree::{LEVEL_3_COORD_LENGTH, LEVEL_3_COORD_MASK, LEVEL_3_COO
use crate::graph::*;
use crate::math::*;

#[derive(Clone, Copy)]
#[repr(transparent)]
pub struct LocalNodeIndex<const LEVEL: u8>(u32);

// XYZXYZXYZXYZXYZXYZXYZXYZ
const LOCAL_NODE_INDEX_X_MASK: u32 = 0b10010010_01001001_00100100;
const LOCAL_NODE_INDEX_Y_MASK: u32 = 0b01001001_00100100_10010010;
const LOCAL_NODE_INDEX_Z_MASK: u32 = 0b00100100_10010010_01001001;

impl<const LEVEL: u8> LocalNodeIndex<LEVEL> {
#[inline(always)]
pub fn pack(unpacked: u8x3) -> Self {
// allocate one byte per bit for each element.
// each element is still has its individual bits in linear ordering, but the bytes in the
// vector are in morton ordering.
#[rustfmt::skip]
let expanded_linear_bits = simd_swizzle!(
unpacked,
[
// X, Y, Z
2, 1, 0,
2, 1, 0,
2, 1, 0,
2, 1, 0,
2, 1, 0,
2, 1, 0,
2, 1, 0,
2, 1, 0, // LSB
]
);

// shift each bit into the sign bit for morton ordering
#[rustfmt::skip]
let expanded_morton_bits = expanded_linear_bits << Simd::<u8, 24>::from_array(
[
7, 7, 7,
6, 6, 6,
5, 5, 5,
4, 4, 4,
3, 3, 3,
2, 2, 2,
1, 1, 1,
0, 0, 0, // LSB
],
);

// arithmetic shift to set each whole lane to its sign bit, then shrinking all lanes to bitmask
let morton_packed = unsafe {
Mask::<i8, 24>::from_int_unchecked(expanded_morton_bits.cast::<i8>() >> Simd::splat(7))
}
.to_bitmask();

Self(morton_packed)
}

#[inline(always)]
pub fn inc_x(self) -> Self {
self.inc::<{ LOCAL_NODE_INDEX_X_MASK }>()
}

#[inline(always)]
pub fn inc_y(self) -> Self {
self.inc::<{ LOCAL_NODE_INDEX_Y_MASK }>()
}

#[inline(always)]
pub fn inc_z(self) -> Self {
self.inc::<{ LOCAL_NODE_INDEX_Z_MASK }>()
}

#[inline(always)]
pub fn dec_x(self) -> Self {
self.dec::<{ LOCAL_NODE_INDEX_X_MASK }>()
}

#[inline(always)]
pub fn dec_y(self) -> Self {
self.dec::<{ LOCAL_NODE_INDEX_Y_MASK }>()
}

#[inline(always)]
pub fn dec_z(self) -> Self {
self.dec::<{ LOCAL_NODE_INDEX_Z_MASK }>()
}

#[inline(always)]
pub fn inc<const MASK: u32>(self) -> Self {
// make the other bits in the number 1
let mut masked = self.0 | !MASK;

// increment
masked = masked.wrapping_add(1_u32 << LEVEL);

// modify only the masked bits in the original number
Self((self.0 & !MASK) | (masked & MASK))
}

#[inline(always)]
pub fn dec<const MASK: u32>(self) -> Self {
// make the other bits in the number 0
let mut masked = self.0 & MASK;

// decrement
masked = masked.wrapping_sub(1_u32 << LEVEL);

// modify only the masked bits in the original number
Self((self.0 & !MASK) | (masked & MASK))
}

#[inline(always)]
pub fn as_array_offset(&self) -> usize {
self.0 as usize
}

#[inline(always)]
pub fn iter_lower_nodes<const LOWER_LEVEL: u8>(&self) -> LowerNodeIter<LEVEL, LOWER_LEVEL> {
LowerNodeIter::new(self)
}

// #[inline(always)]
// pub fn get_all_neighbors(&self) -> [Self; 6] {
// const INC_MASKS: Simd<u32, 6> = Simd::from_array([]);
// }

#[inline(always)]
pub fn unpack(&self) -> u8x3 {
// allocate one byte per bit for each element.
// each element is still has its individual bits in morton ordering, but the bytes in the
// vector are in linear ordering.
#[rustfmt::skip]
let expanded_linear_bits = simd_swizzle!(
u8x4::from_array(self.0.to_le_bytes()),
[
// X
2, 2, 2, 1, 1, 1, 0, 0,
// Y
2, 2, 2, 1, 1, 0, 0, 0,
// Z
2, 2, 1, 1, 1, 0, 0, 0, // LSB
]
);

// shift each bit into the sign bit for morton ordering
#[rustfmt::skip]
let expanded_morton_bits = expanded_linear_bits << Simd::<u8, 24>::from_array(
[
// X
0, 3, 6,
1, 4, 7,
2, 5,
// Y
1, 4, 7,
2, 5, 0,
3, 6,
// Z
2, 5, 0,
3, 6, 1,
4, 7, // LSB
],
);

// arithmetic shift to set each whole lane to its sign bit, then shrinking all lanes to bitmask
let linear_packed = unsafe {
Mask::<i8, 24>::from_int_unchecked(expanded_morton_bits.cast::<i8>() >> Simd::splat(7))
}
.to_bitmask();

u8x3::from_slice(&linear_packed.to_le_bytes()[0..=2])
}
}

pub struct LowerNodeIter<const LEVEL: u8, const LOWER_LEVEL: u8> {
current: u32,
end: u32,
}

impl<const LEVEL: u8, const LOWER_LEVEL: u8> LowerNodeIter<LEVEL, LOWER_LEVEL> {
fn new(index: &LocalNodeIndex<LEVEL>) -> Self {
assert!(LEVEL > LOWER_LEVEL);

let node_size = 1 << (LEVEL * 3);

Self {
current: index.0,
end: index.0 + node_size,
}
}
}

impl<const LEVEL: u8, const LOWER_LEVEL: u8> Iterator for LowerNodeIter<LEVEL, LOWER_LEVEL> {
type Item = LocalNodeIndex<LOWER_LEVEL>;

fn next(&mut self) -> Option<Self::Item> {
if self.current >= self.end {
None
} else {
let current = self.current;

let lower_node_size = 1 << (LOWER_LEVEL * 3);
self.current += lower_node_size;

Some(LocalNodeIndex(current))
}
}
}

pub struct LocalCoordinateContext {
frustum: LocalFrustum,

Expand All @@ -24,7 +231,7 @@ pub struct LocalCoordinateContext {
// this is the index that encompasses the corner of the view distance bounding box where the
// coordinate for each axis is closest to negative infinity, and truncated to the origin of the
// level 3 node it's contained in.
pub iter_node_origin_idx: LocalNodeIndex,
pub iter_node_origin_index: LocalNodeIndex<3>,
pub iter_node_origin_coords: u8x3,
pub level_3_node_iters: u8x3,

Expand Down Expand Up @@ -65,7 +272,7 @@ impl LocalCoordinateContext {
);

let iter_node_origin_coords = iter_section_origin_coords & u8x3::splat(LEVEL_3_COORD_MASK);
let iter_node_origin_idx = LocalNodeIndex::pack(iter_node_origin_coords);
let iter_node_origin_index = LocalNodeIndex::pack(iter_node_origin_coords);

let view_cube_length = (section_view_distance * 2) + 1;
// convert to i32 to avoid implicit wrapping, then explicitly wrap
Expand All @@ -88,7 +295,7 @@ impl LocalCoordinateContext {
fog_distance_squared,
world_bottom_section_y,
world_top_section_y,
iter_node_origin_idx,
iter_node_origin_index,
iter_node_origin_coords,
level_3_node_iters,
iter_region_origin_coords: todo!(),
Expand All @@ -102,7 +309,12 @@ impl LocalCoordinateContext {
// }

#[inline(always)]
pub fn check_node<const LEVEL: u8>(&self, local_node_pos: u8x3) -> BoundsCheckResult {
pub fn test_node<const LEVEL: u8>(
&self,
local_node_index: LocalNodeIndex<LEVEL>,
) -> BoundsCheckResult {
let local_node_pos = local_node_index.unpack();

let bounds = self.node_get_local_bounds::<LEVEL>(local_node_pos);

let mut result = self.bounds_inside_fog::<LEVEL>(&bounds);
Expand Down Expand Up @@ -232,7 +444,8 @@ impl LocalFrustum {
}
}

#[inline(always)]
// #[inline(always)]
#[no_mangle]
pub fn test_local_bounding_box(&self, bb: &LocalBoundingBox) -> BoundsCheckResult {
unsafe {
// These unsafe mask shenanigans just check if the sign bit is set for each lane.
Expand Down Expand Up @@ -271,18 +484,13 @@ impl LocalFrustum {
.fast_fma(inside_bounds_y, self.plane_zs * inside_bounds_z),
);

if outside_length_sq.simd_ge(-self.plane_ws).to_bitmask() == 0b111111 {
if inside_length_sq.simd_ge(-self.plane_ws).to_bitmask() == 0b111111 {
BoundsCheckResult::Inside
} else {
BoundsCheckResult::Partial
}
} else {
if inside_length_sq.simd_ge(-self.plane_ws).to_bitmask() == 0b111111 {
panic!("BAD!!!!!");
}
BoundsCheckResult::Outside
}
// if any outside lengths are greater than -w, return OUTSIDE
// if all inside lengths are greater than -w, return INSIDE
// otherwise, return PARTIAL
let none_outside = outside_length_sq.simd_ge(-self.plane_ws).to_bitmask() == 0b000000;
let all_inside = inside_length_sq.simd_ge(-self.plane_ws).to_bitmask() == 0b111111;

BoundsCheckResult::from_int_unchecked(none_outside as u8 + all_inside as u8)
}
}
}
Expand Down
Loading

0 comments on commit d97eb63

Please sign in to comment.