Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,10 @@ num-complex = { version = "0.4", features = ["bytemuck"] }
# High-performance GEMM for Standard algebra
faer = "0.24"

# Cache-aware copies for materializing strided contraction operands
strided-perm = "0.3"
strided-view = "0.3"

# Contraction order optimization
omeco = "0.2.4"

Expand Down
179 changes: 179 additions & 0 deletions src/backend/cpu/buffer_pool.rs
Original file line number Diff line number Diff line change
@@ -1,3 +1,7 @@
use std::any::{Any, TypeId};
use std::cell::RefCell;
use std::collections::HashMap;

#[derive(Default)]
pub(crate) struct ScratchPool<T> {
free: Vec<Vec<T>>,
Expand Down Expand Up @@ -51,6 +55,97 @@ impl<T> Drop for ScratchBuffer<'_, T> {
}
}

thread_local! {
static PACKING_BUFFERS: RefCell<HashMap<TypeId, Box<dyn Any>>> =
RefCell::new(HashMap::new());
}

const MAX_PACKING_BUFFERS_PER_TYPE: usize = 2;
const MAX_PACKING_BUFFER_BYTES: usize = 64 * 1024 * 1024;

pub(crate) struct PackingBuffer<T: Copy + 'static> {
buf: Vec<T>,
pooled: bool,
}

impl<T: Copy + Default + 'static> PackingBuffer<T> {
pub(crate) fn acquire(len: usize) -> Self {
let bytes = len.saturating_mul(std::mem::size_of::<T>());
if bytes == 0 || bytes > MAX_PACKING_BUFFER_BYTES {
return Self {
buf: vec![T::default(); len],
pooled: false,
};
}

let mut buf = PACKING_BUFFERS.with(|pool| {
let mut pool = pool.borrow_mut();
let buffers = pool
.entry(TypeId::of::<T>())
.or_insert_with(|| Box::new(Vec::<Vec<T>>::new()))
.downcast_mut::<Vec<Vec<T>>>()
.expect("packing buffer pool type mismatch");
let best = buffers
.iter()
.enumerate()
.filter(|(_, buffer)| buffer.capacity() >= len)
.min_by_key(|(_, buffer)| buffer.capacity())
.map(|(index, _)| index);
best.map(|index| buffers.swap_remove(index))
.unwrap_or_else(|| Vec::with_capacity(len))
});
buf.truncate(len);
buf.resize_with(len, T::default);

Self { buf, pooled: true }
}

pub(crate) fn as_slice(&self) -> &[T] {
&self.buf
}

pub(crate) fn as_mut_vec(&mut self) -> &mut Vec<T> {
&mut self.buf
}
}

impl<T: Copy + 'static> Drop for PackingBuffer<T> {
fn drop(&mut self) {
if !self.pooled {
return;
}

let buf = std::mem::take(&mut self.buf);
let bytes = buf.capacity().saturating_mul(std::mem::size_of::<T>());
if bytes == 0 || bytes > MAX_PACKING_BUFFER_BYTES {
return;
}

let _ = PACKING_BUFFERS.try_with(|pool| {
let Ok(mut pool) = pool.try_borrow_mut() else {
return;
};
let buffers = pool
.entry(TypeId::of::<T>())
.or_insert_with(|| Box::new(Vec::<Vec<T>>::new()))
.downcast_mut::<Vec<Vec<T>>>()
.expect("packing buffer pool type mismatch");
if buffers.len() < MAX_PACKING_BUFFERS_PER_TYPE {
buffers.push(buf);
} else if let Some((smallest, capacity)) = buffers
.iter()
.enumerate()
.map(|(index, buffer)| (index, buffer.capacity()))
.min_by_key(|(_, capacity)| *capacity)
{
if capacity < buf.capacity() {
buffers[smallest] = buf;
}
}
});
}
}

#[cfg(test)]
mod tests {
use super::*;
Expand All @@ -75,4 +170,88 @@ mod tests {
let large = pool.acquire(128);
assert!(large.capacity() >= 128);
}

#[test]
fn test_packing_buffer_reuses_initialized_storage() {
#[derive(Clone, Copy, Default)]
struct Marker(u32);

let mut first = PackingBuffer::<Marker>::acquire(127);
first.as_mut_vec().fill(Marker(42));
let pointer = first.as_slice().as_ptr();
drop(first);

let second = PackingBuffer::<Marker>::acquire(127);
assert_eq!(second.as_slice().as_ptr(), pointer);
assert!(second.as_slice().iter().all(|value| value.0 == 42));
}

#[test]
fn test_packing_buffer_keeps_two_largest_and_uses_best_fit() {
#[derive(Clone, Copy, Default)]
struct Marker(u8);

let smallest = PackingBuffer::<Marker>::acquire(8);
let smallest_capacity = smallest.buf.capacity();
drop(smallest);

let middle = PackingBuffer::<Marker>::acquire(smallest_capacity + 1);
let middle_capacity = middle.buf.capacity();
drop(middle);

let largest = PackingBuffer::<Marker>::acquire(middle_capacity + 1);
let largest_capacity = largest.buf.capacity();
drop(largest);

PACKING_BUFFERS.with(|pool| {
let pool = pool.borrow();
let buffers = pool
.get(&TypeId::of::<Marker>())
.unwrap()
.downcast_ref::<Vec<Vec<Marker>>>()
.unwrap();
let mut capacities: Vec<_> = buffers.iter().map(Vec::capacity).collect();
capacities.sort_unstable();
assert_eq!(capacities, vec![middle_capacity, largest_capacity]);
});

let best_fit = PackingBuffer::<Marker>::acquire(middle_capacity);
assert_eq!(best_fit.buf.capacity(), middle_capacity);
assert!(best_fit.as_slice().iter().all(|value| value.0 == 0));
}

#[test]
fn test_packing_buffer_drop_tolerates_reentrant_pool_borrow() {
#[derive(Clone, Copy, Default)]
struct Marker(u8);

let buffer = PackingBuffer::<Marker>::acquire(8);
assert!(buffer.as_slice().iter().all(|value| value.0 == 0));
PACKING_BUFFERS.with(|pool| {
let pool = pool.borrow_mut();
drop(buffer);
let buffers = pool
.get(&TypeId::of::<Marker>())
.unwrap()
.downcast_ref::<Vec<Vec<Marker>>>()
.unwrap();
assert!(buffers.is_empty());
});
}

#[test]
fn test_packing_buffer_does_not_retain_zero_byte_storage() {
#[derive(Clone, Copy, Default)]
struct Marker;

let buffer = PackingBuffer {
buf: vec![Marker; 8],
pooled: true,
};
drop(buffer);

PACKING_BUFFERS.with(|pool| {
assert!(!pool.borrow().contains_key(&TypeId::of::<Marker>()));
});
}
}
Loading
Loading