use crate::common::*; use std::alloc::Layout; /// Given an iterator over BitChunk Items, iterates over the indices (each represented as a u32) for which the bit is SET, /// treating the bits in the BitChunk as a contiguous array. /// e.g. input [0b111000, 0b11] gives output [3, 4, 5, 32, 33]. /// observe that the bits per chunk are ordered from least to most significant bits, yielding smaller to larger usizes. /// assumes chunk_iter will yield no more than std::u32::MAX / 32 chunks pub const fn usize_bytes() -> usize { std::mem::size_of::() } pub const fn usize_bits() -> usize { usize_bytes() * 8 } pub const fn usizes_for_bits(bits: usize) -> usize { (bits + (usize_bits() - 1)) / usize_bits() } type Chunk = usize; type BitIndex = usize; pub(crate) struct BitChunkIter> { cached: usize, chunk_iter: I, next_bit_index: BitIndex, } impl> BitChunkIter { pub fn new(chunk_iter: I) -> Self { // first chunk is always a dummy zero, as if chunk_iter yielded Some(FALSE_CHUNK). // Consequences: // 1. our next_bit_index is always off by usize_bits() (we correct for it in Self::next) (no additional overhead) // 2. we cache Chunk and not Option, because chunk_iter.next() is only called in Self::next. Self { chunk_iter, next_bit_index: 0, cached: 0 } } } impl> Iterator for BitChunkIter { type Item = BitIndex; fn next(&mut self) -> Option { let mut chunk = self.cached; // loop until either: // 1. there are no more Items to return, or // 2. chunk encodes 1+ Items, one of which we will return. while chunk == 0 { // chunk has no bits set! get the next one... chunk = self.chunk_iter.next()?; // ... and jump self.next_bit_index to the next multiple of usize_bits(). self.next_bit_index = (self.next_bit_index + usize_bits()) & !(usize_bits() - 1); } // there exists 1+ set bits in chunk // assert(chunk > 0); // Until the least significant bit of chunk is 1: // 1. shift chunk to the right, // 2. and increment self.next_bit_index accordingly // effectively performs a little binary search, shifting 32, then 16, ... // TODO perhaps there is a more efficient SIMD op for this? const N_INIT: BitIndex = usize_bits() / 2; let mut n = N_INIT; while n >= 1 { // n is [32,16,8,4,2,1] on 64-bit machine // this loop is unrolled with release optimizations let n_least_significant_mask = (1 << n) - 1; if chunk & n_least_significant_mask == 0 { // no 1 set within 0..n least significant bits. self.next_bit_index += n; chunk >>= n; } n /= 2; } // least significant bit of chunk is 1. Item to return is known. // assert(chunk & 1 == 1) // prepare our state for the next time Self::next is called. // Overwrite self.cached such that its shifted state is retained, // and jump over the bit whose index we are about to return. self.next_bit_index += 1; self.cached = chunk >> 1; // returned index is usize_bits() smaller than self.next_bit_index because we use an // off-by-usize_bits() encoding to avoid having to cache an Option. Some(self.next_bit_index - 1 - usize_bits()) } } pub(crate) struct BitChunkIterRev> { cached: usize, chunk_iter: I, next_bit_index: BitIndex, } impl> BitChunkIterRev { pub fn new(chunk_iter: I) -> Self { let next_bit_index = chunk_iter.len() * usize_bits(); Self { chunk_iter, next_bit_index, cached: 0 } } } impl> Iterator for BitChunkIterRev { type Item = BitIndex; fn next(&mut self) -> Option { let mut chunk = self.cached; if chunk == 0 { self.next_bit_index += usize_bits(); loop { self.next_bit_index -= usize_bits(); chunk = self.chunk_iter.next()?; if chunk != 0 { break; } } } const N_INIT: BitIndex = usize_bits() / 2; let mut n = N_INIT; while n >= 1 { let n_most_significant_mask = !0 << (usize_bits() - n); if chunk & n_most_significant_mask == 0 { self.next_bit_index -= n; chunk <<= n; } n /= 2; } self.cached = chunk << 1; self.next_bit_index -= 1; Some(self.next_bit_index) } } /* --properties--> ___ ___ ___ ___ |___|___|___|___| | |___|___|___|___| | |___|___|___|___| | |___|___|___|___| | V entity chunks (groups of size usize_bits()) */ // TODO newtypes Entity and Property #[derive(Debug, Copy, Clone, Eq, PartialEq)] pub struct Pair { pub entity: u32, pub property: u32, } impl From<[u32; 2]> for Pair { fn from([entity, property]: [u32; 2]) -> Self { Pair { entity, property } } } impl Default for BitMatrix { fn default() -> Self { Self::new(Pair { entity: 0, property: 0 }) } } pub struct BitMatrix { buffer: *mut usize, bounds: Pair, layout: Layout, // layout of the currently-allocated buffer } impl Drop for BitMatrix { fn drop(&mut self) { unsafe { // ? std::alloc::dealloc(self.buffer as *mut u8, self.layout); } } } impl Debug for BitMatrix { fn fmt(&self, f: &mut Formatter) -> std::fmt::Result { struct FmtRow<'a> { me: &'a BitMatrix, property: usize, }; impl Debug for FmtRow<'_> { fn fmt(&self, f: &mut Formatter) -> std::fmt::Result { let row_chunks = BitMatrix::row_chunks(self.me.bounds.property as usize); let column_chunks = BitMatrix::column_chunks(self.me.bounds.entity as usize); write!(f, "|")?; for entity_chunk in 0..column_chunks { let mut chunk = unsafe { *self.me.buffer.add(row_chunks * entity_chunk + self.property) }; let end = if entity_chunk + 1 == column_chunks { self.me.bounds.entity % usize_bits() as u32 } else { usize_bits() as u32 }; for _ in 0..end { let c = match chunk & 1 { 0 => '0', _ => '1', }; write!(f, "{}", c)?; chunk >>= 1; } write!(f, "_")?; } Ok(()) } } let row_chunks = BitMatrix::row_chunks(self.bounds.property as usize); let iter = (0..row_chunks).map(move |property| FmtRow { me: self, property }); f.debug_list().entries(iter).finish() } } impl BitMatrix { #[inline] const fn row_of(entity: usize) -> usize { entity / usize_bits() } #[inline] const fn row_chunks(property_bound: usize) -> usize { property_bound } #[inline] const fn column_chunks(entity_bound: usize) -> usize { usizes_for_bits(entity_bound) } #[inline] fn offsets_unchecked(&self, at: Pair) -> [usize; 2] { let o_in = at.entity as usize % usize_bits(); let row = Self::row_of(at.entity as usize); let row_chunks = self.bounds.property as usize; let o_of = row * row_chunks + at.property as usize; [o_of, o_in] } // returns a u32 which has bits 000...000111...111 // for the last JAGGED chunk given the column size // if the last chunk is not jagged (when entity_bound % 32 == 0) // None is returned, // otherwise Some(x) is returned such that x & chunk would mask out // the bits NOT in 0..entity_bound fn last_row_chunk_mask(entity_bound: u32) -> Option { let zero_prefix_len = entity_bound as usize % usize_bits(); if zero_prefix_len == 0 { None } else { Some(!0 >> (usize_bits() - zero_prefix_len)) } } fn assert_within_bounds(&self, at: Pair) { assert!(at.entity < self.bounds.entity); assert!(at.property < self.bounds.property); } fn layout_for(total_chunks: usize) -> std::alloc::Layout { unsafe { // this layout is ALWAYS valid: // 1. size is always nonzero // 2. size is always a multiple of 4 and 4-aligned Layout::from_size_align_unchecked(usize_bytes() * total_chunks.max(1), usize_bytes()) } } ///////// pub fn get_bounds(&self) -> &Pair { &self.bounds } pub fn grow_to(&mut self, bounds: Pair) { assert!(bounds.entity >= self.bounds.entity); assert!(bounds.property >= self.bounds.property); let old_row_chunks = Self::row_chunks(self.bounds.property as usize); let old_col_chunks = Self::column_chunks(self.bounds.entity as usize); let new_row_chunks = Self::row_chunks(bounds.property as usize); let new_col_chunks = Self::column_chunks(bounds.entity as usize); let new_layout = Self::layout_for(new_row_chunks * new_col_chunks); let new_buffer = unsafe { let new_buffer = std::alloc::alloc(new_layout) as *mut usize; let mut src: *mut usize = self.buffer; let mut dest: *mut usize = new_buffer; let row_chunk_diff = new_row_chunks - old_row_chunks; for _col_idx in 0..old_col_chunks { src.copy_to_nonoverlapping(dest, old_row_chunks); src = src.add(old_row_chunks); dest = dest.add(old_row_chunks); if row_chunk_diff > 0 { dest.write_bytes(0u8, row_chunk_diff); dest = dest.add(row_chunk_diff); } } let last_zero_chunks = (new_col_chunks - old_col_chunks) * new_row_chunks; dest.write_bytes(0u8, last_zero_chunks); new_buffer }; self.layout = new_layout; self.buffer = new_buffer; self.bounds = bounds; } pub fn clear(&mut self) { let total_chunks = Self::row_chunks(self.bounds.property as usize) * Self::column_chunks(self.bounds.entity as usize); unsafe { self.buffer.write_bytes(0u8, total_chunks); } } pub fn new(bounds: Pair) -> Self { let total_chunks = Self::row_chunks(bounds.property as usize) * Self::column_chunks(bounds.entity as usize); let layout = Self::layout_for(total_chunks); let buffer; unsafe { buffer = std::alloc::alloc(layout) as *mut usize; buffer.write_bytes(0u8, total_chunks); }; Self { buffer, bounds, layout } } pub fn set(&mut self, at: Pair) { self.assert_within_bounds(at); let [o_of, o_in] = self.offsets_unchecked(at); unsafe { *self.buffer.add(o_of) |= 1 << o_in }; } pub fn unset(&mut self, at: Pair) { self.assert_within_bounds(at); let [o_of, o_in] = self.offsets_unchecked(at); unsafe { *self.buffer.add(o_of) &= !(1 << o_in) }; } pub fn test(&self, at: Pair) -> bool { self.assert_within_bounds(at); let [o_of, o_in] = self.offsets_unchecked(at); unsafe { *self.buffer.add(o_of) & 1 << o_in != 0 } } pub fn batch_mut<'a, 'b>(&mut self, mut chunk_mut_fn: impl FnMut(&'b mut [BitChunk])) { let row_chunks = Self::row_chunks(self.bounds.property as usize); let column_chunks = Self::column_chunks(self.bounds.entity as usize); let mut ptr = self.buffer; for _row in 0..column_chunks { let slice; unsafe { let slicey = std::slice::from_raw_parts_mut(ptr, row_chunks); slice = std::mem::transmute(slicey); ptr = ptr.add(row_chunks); } chunk_mut_fn(slice); } if let Some(mask) = Self::last_row_chunk_mask(self.bounds.entity) { // TODO TEST let mut ptr = unsafe { self.buffer.add((column_chunks - 1) * row_chunks) }; for _ in 0..row_chunks { unsafe { *ptr &= mask; ptr = ptr.add(1); } } } } /// given: /// 1. a buffer to work with /// 2. a _fold function_ for combining the properties of a given entity /// and returning a new derived property (working ) pub fn iter_entities_where<'a, 'b>( &'a self, buf: &'b mut Vec, mut fold_fn: impl FnMut(&'b [BitChunk]) -> BitChunk, ) -> impl Iterator + 'b { let buf_start = buf.len(); let row_chunks = Self::row_chunks(self.bounds.property as usize); let column_chunks = Self::column_chunks(self.bounds.entity as usize); let mut ptr = self.buffer; for _row in 0..column_chunks { let slice; unsafe { let slicey = std::slice::from_raw_parts(ptr, row_chunks); slice = std::mem::transmute(slicey); ptr = ptr.add(row_chunks); } let chunk = fold_fn(slice); buf.push(chunk.0); } if let Some(mask) = Self::last_row_chunk_mask(self.bounds.entity) { *buf.iter_mut().last().unwrap() &= mask; } BitChunkIter::new(buf.drain(buf_start..)).map(|x| x as u32) } pub fn iter_entities_where_rev<'a, 'b>( &'a self, buf: &'b mut Vec, mut fold_fn: impl FnMut(&'b [BitChunk]) -> BitChunk, ) -> impl Iterator + 'b { let buf_start = buf.len(); let row_chunks = Self::row_chunks(self.bounds.property as usize); let column_chunks = Self::column_chunks(self.bounds.entity as usize); let mut ptr = self.buffer; for _row in 0..column_chunks { let slice; unsafe { let slicey = std::slice::from_raw_parts(ptr, row_chunks); slice = std::mem::transmute(slicey); ptr = ptr.add(row_chunks); } let chunk = fold_fn(slice); buf.push(chunk.0); } if let Some(mask) = Self::last_row_chunk_mask(self.bounds.entity) { *buf.iter_mut().last().unwrap() &= mask; } BitChunkIterRev::new(buf.drain(buf_start..).rev()).map(|x| x as u32) } } use derive_more::*; #[derive( Debug, Copy, Clone, BitAnd, Not, BitOr, BitXor, BitAndAssign, BitOrAssign, BitXorAssign, )] #[repr(transparent)] pub struct BitChunk(usize); impl BitChunk { const fn any(self) -> bool { self.0 != FALSE_CHUNK.0 } const fn all(self) -> bool { self.0 == TRUE_CHUNK.0 } } pub const TRUE_CHUNK: BitChunk = BitChunk(!0); pub const FALSE_CHUNK: BitChunk = BitChunk(0); #[test] fn matrix_test() { let mut m = BitMatrix::new(Pair { entity: 70, property: 3 }); m.set([2, 0].into()); m.set([40, 1].into()); m.set([40, 2].into()); m.set([40, 0].into()); println!("{:#?}", &m); m.batch_mut(|p| p[0] = TRUE_CHUNK); println!("{:#?}", &m); for i in (0..40).step_by(7) { m.unset([i, 0].into()); } m.unset([62, 0].into()); println!("{:#?}", &m); m.batch_mut(move |p| p[1] = p[0] ^ TRUE_CHUNK); println!("{:#?}", &m); let mut buf = vec![]; for index in m.iter_entities_where(&mut buf, move |p| p[1]) { println!("index {}", index); } for index in m.iter_entities_where_rev(&mut buf, move |p| p[1]) { println!("index {}", index); } } #[test] fn bit_chunk_iter_rev() { let x = &[0b1, 0b1000011, 0, 0, 0b101]; for i in BitChunkIterRev::new(x.iter().copied()) { println!("i = {:?}", i); } }