Revert just IBLT, remove some cruft, add boxed new function.

This commit is contained in:
Adam Ierymenko 2022-04-20 14:10:29 -04:00
parent dcfdd036cf
commit bb701f6e61
No known key found for this signature in database
GPG key ID: C8877CF2D7A5D7F3
6 changed files with 80 additions and 441 deletions

View file

@ -13,26 +13,3 @@ panic = 'abort'
[dependencies] [dependencies]
crc32fast = "^1" crc32fast = "^1"
zerocopy = { version = "0.6.1", features = ["alloc"] }
[dev-dependencies]
rand = ">=0"
criterion = ">=0"
[lib]
[[bench]]
name = "to_from_bytes"
harness = false
[[bench]]
name = "clone"
harness = false
[[bench]]
name = "list"
harness = false
[[bench]]
name = "merge"
harness = false

View file

@ -1,19 +0,0 @@
use criterion::{criterion_group, criterion_main, Criterion};
use iblt::IBLT;
const CAPACITY: usize = 4096;
type OurIBLT = IBLT<[u8; 32], CAPACITY, 3>;
pub fn criterion_benchmark(c: &mut Criterion) {
let mut iblt = OurIBLT::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 32];
v.fill_with(rand::random);
iblt.insert(&v);
}
c.bench_function("clone", |b| b.iter(|| iblt.clone()));
}
criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);

View file

@ -1,39 +0,0 @@
use criterion::{criterion_group, criterion_main, Criterion};
use iblt::IBLT;
const CAPACITY: usize = 4096;
type IBLT32 = IBLT<[u8; 32], CAPACITY, 3>;
type IBLT16 = IBLT<[u8; 16], CAPACITY, 3>;
type IBLT8 = IBLT<[u8; 8], CAPACITY, 3>;
pub fn criterion_benchmark(c: &mut Criterion) {
let mut iblt = IBLT32::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 32];
v.fill_with(rand::random);
iblt.insert(&v);
}
c.bench_function("list 32", |b| b.iter(|| iblt.list(|_, _| {})));
let mut iblt = IBLT16::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 16];
v.fill_with(rand::random);
iblt.insert(&v);
}
c.bench_function("list 16", |b| b.iter(|| iblt.list(|_, _| {})));
let mut iblt = IBLT8::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 8];
v.fill_with(rand::random);
iblt.insert(&v);
}
c.bench_function("list 8", |b| b.iter(|| iblt.list(|_, _| {})));
}
criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);

View file

@ -1,78 +0,0 @@
use criterion::{criterion_group, criterion_main, Criterion};
use iblt::IBLT;
const CAPACITY: usize = 4096;
type IBLT32 = IBLT<[u8; 32], CAPACITY, 3>;
type IBLT16 = IBLT<[u8; 16], CAPACITY, 3>;
type IBLT8 = IBLT<[u8; 8], CAPACITY, 3>;
pub fn criterion_benchmark(c: &mut Criterion) {
let mut orig = IBLT32::new();
let mut new = IBLT32::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 32];
v.fill_with(rand::random);
orig.insert(&v);
}
for _ in 0..CAPACITY {
let mut v = [0u8; 32];
v.fill_with(rand::random);
new.insert(&v);
}
c.bench_function("merge 32", |b| {
b.iter(|| {
let mut new2 = new.clone();
orig.subtract(&new);
new2.subtract(&orig);
})
});
let mut orig = IBLT16::new();
let mut new = IBLT16::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 16];
v.fill_with(rand::random);
orig.insert(&v);
}
for _ in 0..CAPACITY {
let mut v = [0u8; 16];
v.fill_with(rand::random);
new.insert(&v);
}
c.bench_function("merge 16", |b| {
b.iter(|| {
let mut new2 = new.clone();
orig.subtract(&new);
new2.subtract(&orig);
})
});
let mut orig = IBLT8::new();
let mut new = IBLT8::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 8];
v.fill_with(rand::random);
orig.insert(&v);
}
for _ in 0..CAPACITY {
let mut v = [0u8; 8];
v.fill_with(rand::random);
new.insert(&v);
}
c.bench_function("merge 8", |b| {
b.iter(|| {
let mut new2 = new.clone();
orig.subtract(&new);
new2.subtract(&orig);
})
});
}
criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);

View file

@ -1,19 +0,0 @@
use criterion::{criterion_group, criterion_main, Criterion};
use iblt::IBLT;
const CAPACITY: usize = 4096;
type OurIBLT = IBLT<[u8; 32], CAPACITY, 3>;
pub fn criterion_benchmark(c: &mut Criterion) {
let mut iblt = OurIBLT::new();
for _ in 0..CAPACITY {
let mut v = [0u8; 32];
v.fill_with(rand::random);
iblt.insert(&v);
}
c.bench_function("to_from_bytes", |b| b.iter(|| OurIBLT::from_bytes(iblt.as_bytes())));
}
criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);

View file

@ -6,41 +6,7 @@
* https://www.zerotier.com/ * https://www.zerotier.com/
*/ */
use zerocopy::{AsBytes, FromBytes}; use std::borrow::Cow;
#[cfg(not(any(target_arch = "aarch64", target_arch = "x86", target_arch = "x86_64", target_arch = "powerpc64")))]
#[inline(always)]
fn xor_with<T>(x: &mut T, y: &T)
where
T: FromBytes + AsBytes + Sized,
{
x.as_bytes_mut().iter_mut().zip(y.as_bytes().iter()).for_each(|(a, b)| *a ^= *b);
}
#[cfg(any(target_arch = "aarch64", target_arch = "x86", target_arch = "x86_64", target_arch = "powerpc64"))]
#[inline(always)]
fn xor_with<T>(x: &mut T, y: &T)
where
T: FromBytes + AsBytes + Sized,
{
let size = std::mem::size_of::<T>();
if size >= 16 {
for i in 0..(size / 16) {
unsafe { *x.as_bytes_mut().as_mut_ptr().cast::<u128>().add(i) ^= *y.as_bytes().as_ptr().cast::<u128>().add(i) };
}
for i in (size - (size % 16))..size {
unsafe { *x.as_bytes_mut().as_mut_ptr().add(i) ^= *y.as_bytes().as_ptr().add(i) };
}
} else {
for i in 0..(size / 8) {
unsafe { *x.as_bytes_mut().as_mut_ptr().cast::<u64>().add(i) ^= *y.as_bytes().as_ptr().cast::<u64>().add(i) };
}
for i in (size - (size % 8))..size {
unsafe { *x.as_bytes_mut().as_mut_ptr().add(i) ^= *y.as_bytes().as_ptr().add(i) };
}
}
}
#[inline(always)] #[inline(always)]
fn murmurhash32_mix32(mut x: u32) -> u32 { fn murmurhash32_mix32(mut x: u32) -> u32 {
@ -52,6 +18,13 @@ fn murmurhash32_mix32(mut x: u32) -> u32 {
x x
} }
#[inline(always)]
fn xor_with<const L: usize>(x: &mut [u8; L], y: &[u8; L]) {
for i in 0..L {
x[i] ^= y[i];
}
}
/// An Invertible Bloom Lookup Table for set reconciliation. /// An Invertible Bloom Lookup Table for set reconciliation.
/// ///
/// Usage inspired by this paper: /// Usage inspired by this paper:
@ -69,38 +42,31 @@ fn murmurhash32_mix32(mut x: u32) -> u32 {
/// ///
/// The best value for HASHES seems to be 3 for an optimal fill of 75%. /// The best value for HASHES seems to be 3 for an optimal fill of 75%.
#[repr(C)] #[repr(C)]
pub struct IBLT<T, const BUCKETS: usize, const HASHES: usize> pub struct IBLT<const BUCKETS: usize, const ITEM_BYTES: usize, const HASHES: usize> {
where check_hash: [u32; BUCKETS],
T: FromBytes + AsBytes + Default + Sized + Clone, count: [i8; BUCKETS],
{ key: [[u8; ITEM_BYTES]; BUCKETS],
check_hash: Box<Vec<u32>>,
count: Box<Vec<i8>>,
key: Box<Vec<T>>,
} }
impl<T, const BUCKETS: usize, const HASHES: usize> Clone for IBLT<T, BUCKETS, HASHES> impl<const BUCKETS: usize, const ITEM_BYTES: usize, const HASHES: usize> Clone for IBLT<BUCKETS, ITEM_BYTES, HASHES> {
where #[inline(always)]
T: FromBytes + AsBytes + Default + Sized + Clone,
{
fn clone(&self) -> Self { fn clone(&self) -> Self {
// Intentionally designed to work on any platform even if unaligned. The default way
// of implementing clone() may not since check_hash[] is u32.
unsafe { unsafe {
let mut tmp = Self::new(); let mut tmp: Self = std::mem::MaybeUninit::uninit().assume_init();
std::ptr::copy_nonoverlapping(self.check_hash.as_ptr(), tmp.check_hash.as_mut_ptr(), BUCKETS); std::ptr::copy_nonoverlapping((self as *const Self).cast::<u8>(), (&mut tmp as *mut Self).cast::<u8>(), Self::SIZE_BYTES);
std::ptr::copy_nonoverlapping(self.count.as_ptr(), tmp.count.as_mut_ptr(), BUCKETS);
std::ptr::copy_nonoverlapping(self.key.as_ptr(), tmp.key.as_mut_ptr(), BUCKETS);
tmp tmp
} }
} }
} }
impl<T, const BUCKETS: usize, const HASHES: usize> IBLT<T, BUCKETS, HASHES> impl<const BUCKETS: usize, const ITEM_BYTES: usize, const HASHES: usize> IBLT<BUCKETS, ITEM_BYTES, HASHES> {
where
T: FromBytes + AsBytes + Default + Sized + Clone,
{
/// Number of bytes each bucket consumes (not congituously, but doesn't matter). /// Number of bytes each bucket consumes (not congituously, but doesn't matter).
const BUCKET_SIZE_BYTES: usize = std::mem::size_of::<T>() + 4 + 1; const BUCKET_SIZE_BYTES: usize = ITEM_BYTES + 4 + 1;
/// Number of buckets in this IBLT. /// Number of buckets in this IBLT.
#[allow(unused)]
pub const BUCKETS: usize = BUCKETS; pub const BUCKETS: usize = BUCKETS;
/// Size of this IBLT in bytes. /// Size of this IBLT in bytes.
@ -109,34 +75,24 @@ where
/// Create a new zeroed IBLT. /// Create a new zeroed IBLT.
#[inline(always)] #[inline(always)]
pub fn new() -> Self { pub fn new() -> Self {
assert!(BUCKETS < (i32::MAX as usize)); assert!(Self::SIZE_BYTES <= std::mem::size_of::<Self>());
unsafe { std::mem::zeroed() }
let mut s = Self {
check_hash: Box::new(Vec::with_capacity(BUCKETS)),
count: Box::new(Vec::with_capacity(BUCKETS)),
key: Box::new(Vec::with_capacity(BUCKETS)),
};
s.reset();
s
} }
/// Create a new zeroed IBLT on the heap.
///
/// This is useful to create and use IBLT instances too large to fit on the stack.
#[inline(always)] #[inline(always)]
pub fn as_bytes(&self) -> Box<Vec<u8>> { pub fn new_boxed() -> Box<Self> {
let check_hash_len = BUCKETS * 4; assert!(Self::SIZE_BYTES <= std::mem::size_of::<Self>());
let t_len = BUCKETS * std::mem::size_of::<T>(); unsafe { Box::from_raw(std::alloc::alloc_zeroed(std::alloc::Layout::new::<Self>()).cast()) }
let len = check_hash_len + BUCKETS + t_len; }
let mut buf = Box::new(Vec::with_capacity(len)); /// Get this IBLT as a byte slice (free cast operation).
buf.resize(len, 0); /// The returned slice is always SIZE_BYTES in length.
#[inline(always)]
let byt = buf.as_bytes_mut(); pub fn as_bytes(&self) -> &[u8] {
unsafe { &*std::ptr::slice_from_raw_parts((self as *const Self).cast::<u8>(), Self::SIZE_BYTES) }
byt[0..check_hash_len].copy_from_slice(self.check_hash.as_bytes());
byt[check_hash_len..BUCKETS + check_hash_len].copy_from_slice(self.count.as_bytes());
byt[len - t_len..len].copy_from_slice(self.key.as_bytes());
buf
} }
/// Obtain an IBLT from bytes in memory. /// Obtain an IBLT from bytes in memory.
@ -145,33 +101,21 @@ where
/// Cow to 'b' that is just a cast. If re-alignment is necessary it returns an owned Cow containing a properly /// Cow to 'b' that is just a cast. If re-alignment is necessary it returns an owned Cow containing a properly
/// aligned copy. This makes conversion a nearly free cast when alignment adjustment isn't needed. /// aligned copy. This makes conversion a nearly free cast when alignment adjustment isn't needed.
#[inline(always)] #[inline(always)]
pub fn from_bytes(b: Box<Vec<u8>>) -> Option<Self> { pub fn from_bytes<'a>(b: &'a [u8]) -> Option<Cow<'a, Self>> {
if b.len() == Self::SIZE_BYTES { if b.len() == Self::SIZE_BYTES {
// FIXME I commented this out because I do not have access to the architectures needed.
// #[cfg(not(any(target_arch = "x86_64", target_arch = "x86", target_arch = "powerpc64", target_arch = "aarch64")))]
// {
// if b.as_ptr().align_offset(8) == 0 {
// Some(Cow::Borrowed(unsafe { &*b.as_ptr().cast() }))
// } else {
// // NOTE: clone() is implemented above using a raw copy so that alignment doesn't matter.
// Some(Cow::Owned(unsafe { &*b.as_ptr().cast::<Self>() }.clone()))
// }
// }
#[cfg(any(target_arch = "x86_64", target_arch = "x86", target_arch = "powerpc64", target_arch = "aarch64"))] #[cfg(any(target_arch = "x86_64", target_arch = "x86", target_arch = "powerpc64", target_arch = "aarch64"))]
{ {
let mut tmp = Self::new(); Some(Cow::Borrowed(unsafe { &*b.as_ptr().cast() }))
}
let mut i = 0; #[cfg(not(any(target_arch = "x86_64", target_arch = "x86", target_arch = "powerpc64", target_arch = "aarch64")))]
{
tmp.check_hash.as_bytes_mut().copy_from_slice(&b[0..BUCKETS * 4]); if b.as_ptr().align_offset(4) == 0 {
i += BUCKETS * 4; Some(Cow::Borrowed(unsafe { &*b.as_ptr().cast() }))
} else {
tmp.count.as_bytes_mut().copy_from_slice(&b[i..i + BUCKETS]); // NOTE: clone() is implemented above using a raw copy so that alignment doesn't matter.
i += BUCKETS; Some(Cow::Owned(unsafe { &*b.as_ptr().cast::<Self>() }.clone()))
}
tmp.key.as_bytes_mut().copy_from_slice(&b[i..i + std::mem::size_of::<T>() * BUCKETS]);
Some(tmp)
} }
} else { } else {
None None
@ -181,45 +125,42 @@ where
/// Zero this IBLT. /// Zero this IBLT.
#[inline(always)] #[inline(always)]
pub fn reset(&mut self) { pub fn reset(&mut self) {
self.check_hash.clear(); unsafe { std::ptr::write_bytes((self as *mut Self).cast::<u8>(), 0, std::mem::size_of::<Self>()) };
self.count.clear();
self.key.clear();
self.check_hash.resize(BUCKETS, 0);
self.count.resize(BUCKETS, 0);
self.key.resize(BUCKETS, Default::default());
} }
pub(crate) fn ins_rem(&mut self, key: &T, delta: i8) { pub(crate) fn ins_rem(&mut self, key: &[u8; ITEM_BYTES], delta: i8) {
let check_hash = crc32fast::hash(key.as_bytes()); let check_hash = crc32fast::hash(key);
let mut iteration_index = u32::from_le(check_hash).wrapping_add(1); let mut iteration_index = u32::from_le(check_hash).wrapping_add(1);
for _ in 0..(HASHES as u64) { for _ in 0..(HASHES as u64) {
iteration_index = murmurhash32_mix32(iteration_index); iteration_index = murmurhash32_mix32(iteration_index);
let i = (iteration_index as usize) % BUCKETS; let i = (iteration_index as usize) % BUCKETS;
self.check_hash[i] ^= check_hash; self.check_hash[i] ^= check_hash;
self.count[i] = self.count[i].wrapping_add(delta); self.count[i] = self.count[i].wrapping_add(delta);
xor_with(&mut self.key[i], &key); xor_with(&mut self.key[i], key);
} }
} }
/// Insert a set item into this set. /// Insert a set item into this set.
/// This will panic if the slice is smaller than ITEM_BYTES. /// This will panic if the slice is smaller than ITEM_BYTES.
#[inline(always)] #[inline(always)]
pub fn insert(&mut self, key: &T) { pub fn insert(&mut self, key: &[u8]) {
self.ins_rem(key, 1); assert!(key.len() >= ITEM_BYTES);
self.ins_rem(unsafe { &*key.as_ptr().cast() }, 1);
} }
/// Insert a set item into this set. /// Insert a set item into this set.
/// This will panic if the slice is smaller than ITEM_BYTES. /// This will panic if the slice is smaller than ITEM_BYTES.
#[inline(always)] #[inline(always)]
pub fn remove(&mut self, key: &T) { pub fn remove(&mut self, key: &[u8]) {
self.ins_rem(key, -1); assert!(key.len() >= ITEM_BYTES);
self.ins_rem(unsafe { &*key.as_ptr().cast() }, -1);
} }
/// Subtract another IBLT from this one to get a set difference. /// Subtract another IBLT from this one to get a set difference.
pub fn subtract(&mut self, other: &Self) { pub fn subtract(&mut self, other: &Self) {
self.check_hash.iter_mut().zip(other.check_hash.iter()).for_each(|(a, b)| *a ^= *b); self.check_hash.iter_mut().zip(other.check_hash.iter()).for_each(|(a, b)| *a ^= *b);
self.count.iter_mut().zip(other.count.iter()).for_each(|(a, b)| *a = a.wrapping_sub(*b)); self.count.iter_mut().zip(other.count.iter()).for_each(|(a, b)| *a = a.wrapping_sub(*b));
self.key.iter_mut().zip(other.key.iter()).for_each(|(a, b)| xor_with(a, &b)); self.key.iter_mut().zip(other.key.iter()).for_each(|(a, b)| xor_with(a, b));
} }
/// List as many entries in this IBLT as can be extracted. /// List as many entries in this IBLT as can be extracted.
@ -236,12 +177,13 @@ where
/// Due to the small check hash sizes used in this IBLT there is a very small chance this will list /// Due to the small check hash sizes used in this IBLT there is a very small chance this will list
/// bogus items that were never added. This is not an issue with this protocol as it would just result /// bogus items that were never added. This is not an issue with this protocol as it would just result
/// in an unsatisfied record request. /// in an unsatisfied record request.
pub fn list<F: FnMut(T, bool)>(&mut self, mut f: F) -> bool { pub fn list<F: FnMut([u8; ITEM_BYTES], bool)>(mut self, mut f: F) -> bool {
let mut queue: Box<Vec<u32>> = Box::new(Vec::with_capacity(BUCKETS)); assert!(BUCKETS <= (u32::MAX as usize));
let mut queue: Vec<u32> = Vec::with_capacity(BUCKETS);
for i in 0..BUCKETS { for i in 0..BUCKETS {
let count = self.count[i]; let count = self.count[i];
if (count == 1 || count == -1) && crc32fast::hash(&self.key[i].as_bytes()) == self.check_hash[i] { if (count == 1 || count == -1) && crc32fast::hash(&self.key[i]) == self.check_hash[i] {
queue.push(i as u32); queue.push(i as u32);
} }
} }
@ -257,7 +199,7 @@ where
let check_hash = self.check_hash[i]; let check_hash = self.check_hash[i];
let count = self.count[i]; let count = self.count[i];
let key = &self.key[i]; let key = &self.key[i];
if (count == 1 || count == -1) && check_hash == crc32fast::hash(key.as_bytes()) { if (count == 1 || count == -1) && check_hash == crc32fast::hash(key) {
let key = key.clone(); let key = key.clone();
let mut iteration_index = u32::from_le(check_hash).wrapping_add(1); let mut iteration_index = u32::from_le(check_hash).wrapping_add(1);
@ -270,8 +212,8 @@ where
self.check_hash[i2] = check_hash2; self.check_hash[i2] = check_hash2;
self.count[i2] = count2; self.count[i2] = count2;
xor_with(key2, &key); xor_with(key2, &key);
if (count2 == 1 || count2 == -1) && check_hash2 == crc32fast::hash(key2.as_bytes()) { if (count2 == 1 || count2 == -1) && check_hash2 == crc32fast::hash(key2) {
if queue.len() > BUCKETS { if queue.len() >= BUCKETS {
// sanity check, should be impossible // sanity check, should be impossible
break 'list_main; break 'list_main;
} }
@ -287,17 +229,14 @@ where
} }
} }
impl<T, const BUCKETS: usize, const HASHES: usize> PartialEq for IBLT<T, BUCKETS, HASHES> impl<const BUCKETS: usize, const ITEM_BYTES: usize, const HASHES: usize> PartialEq for IBLT<BUCKETS, ITEM_BYTES, HASHES> {
where
T: AsBytes + FromBytes + Default + Clone,
{
#[inline(always)] #[inline(always)]
fn eq(&self, other: &Self) -> bool { fn eq(&self, other: &Self) -> bool {
self.as_bytes().eq(&other.as_bytes()) self.as_bytes().eq(other.as_bytes())
} }
} }
impl<T, const BUCKETS: usize, const HASHES: usize> Eq for IBLT<T, BUCKETS, HASHES> where T: AsBytes + FromBytes + Default + Clone {} impl<const BUCKETS: usize, const ITEM_BYTES: usize, const HASHES: usize> Eq for IBLT<BUCKETS, ITEM_BYTES, HASHES> {}
#[cfg(test)] #[cfg(test)]
mod tests { mod tests {
@ -317,68 +256,31 @@ mod tests {
const HASHES: usize = 3; const HASHES: usize = 3;
fn check_xor_with2<const L: usize>() { #[test]
let with = [17_u8; L]; fn struct_packing() {
let mut expected = [69_u8; L];
let mut actual = [69_u8; L];
expected.iter_mut().zip(with.iter()).for_each(|(a, b)| *a ^= *b);
xor_with(&mut actual, &with);
assert!(actual.eq(&expected));
}
fn typical_iblt() -> IBLT<[u8; 32], 16, 3> {
// Typical case // Typical case
let mut tmp = IBLT::<[u8; 32], 16, 3>::new(); let mut tmp = IBLT::<64, 16, 3>::new();
tmp.check_hash.fill(0x01010101); tmp.check_hash.fill(0x01010101);
tmp.count.fill(1); tmp.count.fill(1);
tmp.key.iter_mut().for_each(|x| x.fill(1)); tmp.key.iter_mut().for_each(|x| x.fill(1));
tmp
}
#[test]
fn check_xor_with() {
check_xor_with2::<128>();
check_xor_with2::<65>();
check_xor_with2::<64>();
check_xor_with2::<63>();
check_xor_with2::<33>();
check_xor_with2::<32>();
check_xor_with2::<31>();
check_xor_with2::<17>();
check_xor_with2::<16>();
check_xor_with2::<15>();
check_xor_with2::<9>();
check_xor_with2::<8>();
check_xor_with2::<7>();
check_xor_with2::<6>();
check_xor_with2::<5>();
check_xor_with2::<4>();
check_xor_with2::<3>();
check_xor_with2::<2>();
check_xor_with2::<1>();
}
#[test]
fn struct_packing() {
let tmp = typical_iblt();
assert!(tmp.as_bytes().iter().all(|x| *x == 1)); assert!(tmp.as_bytes().iter().all(|x| *x == 1));
// Pathological alignment case #1 // Pathological alignment case #1
let mut tmp = IBLT::<[u8; 17], 13, 3>::new(); let mut tmp = IBLT::<17, 13, 3>::new();
tmp.check_hash.fill(0x01010101); tmp.check_hash.fill(0x01010101);
tmp.count.fill(1); tmp.count.fill(1);
tmp.key.iter_mut().for_each(|x| x.fill(1)); tmp.key.iter_mut().for_each(|x| x.fill(1));
assert!(tmp.as_bytes().iter().all(|x| *x == 1)); assert!(tmp.as_bytes().iter().all(|x| *x == 1));
// Pathological alignment case #2 // Pathological alignment case #2
let mut tmp = IBLT::<[u8; 17], 8, 3>::new(); let mut tmp = IBLT::<17, 8, 3>::new();
tmp.check_hash.fill(0x01010101); tmp.check_hash.fill(0x01010101);
tmp.count.fill(1); tmp.count.fill(1);
tmp.key.iter_mut().for_each(|x| x.fill(1)); tmp.key.iter_mut().for_each(|x| x.fill(1));
assert!(tmp.as_bytes().iter().all(|x| *x == 1)); assert!(tmp.as_bytes().iter().all(|x| *x == 1));
// Pathological alignment case #3 // Pathological alignment case #3
let mut tmp = IBLT::<[u8; 16], 7, 3>::new(); let mut tmp = IBLT::<16, 7, 3>::new();
tmp.check_hash.fill(0x01010101); tmp.check_hash.fill(0x01010101);
tmp.count.fill(1); tmp.count.fill(1);
tmp.key.iter_mut().for_each(|x| x.fill(1)); tmp.key.iter_mut().for_each(|x| x.fill(1));
@ -387,13 +289,12 @@ mod tests {
#[test] #[test]
fn fill_list_performance() { fn fill_list_performance() {
const LENGTH: usize = 16;
const CAPACITY: usize = 4096; const CAPACITY: usize = 4096;
let mut rn: u128 = 0xd3b07384d113edec49eaa6238ad5ff00; let mut rn: u128 = 0xd3b07384d113edec49eaa6238ad5ff00;
let mut expected: HashSet<u128> = HashSet::with_capacity(CAPACITY); let mut expected: HashSet<u128> = HashSet::with_capacity(4096);
let mut count = LENGTH; let mut count = 64;
while count <= CAPACITY { while count <= CAPACITY {
let mut test = IBLT::<[u8; LENGTH], CAPACITY, HASHES>::new(); let mut test = IBLT::<CAPACITY, 16, HASHES>::new_boxed();
expected.clear(); expected.clear();
for _ in 0..count { for _ in 0..count {
@ -410,16 +311,15 @@ mod tests {
}); });
println!("inserted: {}\tlisted: {}\tcapacity: {}\tscore: {:.4}\tfill: {:.4}", count, list_count, CAPACITY, (list_count as f64) / (count as f64), (count as f64) / (CAPACITY as f64)); println!("inserted: {}\tlisted: {}\tcapacity: {}\tscore: {:.4}\tfill: {:.4}", count, list_count, CAPACITY, (list_count as f64) / (count as f64), (count as f64) / (CAPACITY as f64));
count += LENGTH; count += 32;
} }
} }
#[test] #[test]
fn merge_sets() { fn merge_sets() {
const CAPACITY: usize = 4096; // previously 16384; const CAPACITY: usize = 16384;
const REMOTE_SIZE: usize = 1024 * 1024 * 2; const REMOTE_SIZE: usize = 1024 * 1024 * 2;
const STEP: usize = 1024; const STEP: usize = 1024;
const LENGTH: usize = 16;
let mut rn: u128 = 0xd3b07384d113edec49eaa6238ad5ff00; let mut rn: u128 = 0xd3b07384d113edec49eaa6238ad5ff00;
let mut missing_count = 1024; let mut missing_count = 1024;
let mut missing: HashSet<u128> = HashSet::with_capacity(CAPACITY * 2); let mut missing: HashSet<u128> = HashSet::with_capacity(CAPACITY * 2);
@ -427,8 +327,8 @@ mod tests {
while missing_count <= CAPACITY { while missing_count <= CAPACITY {
missing.clear(); missing.clear();
all.clear(); all.clear();
let mut local = IBLT::<[u8; LENGTH], CAPACITY, HASHES>::new(); let mut local = IBLT::<CAPACITY, 16, HASHES>::new_boxed();
let mut remote = IBLT::<[u8; LENGTH], CAPACITY, HASHES>::new(); let mut remote = IBLT::<CAPACITY, 16, HASHES>::new_boxed();
let mut k = 0; let mut k = 0;
while k < REMOTE_SIZE { while k < REMOTE_SIZE {
@ -463,87 +363,4 @@ mod tests {
missing_count += STEP; missing_count += STEP;
} }
} }
#[derive(Eq, PartialEq, Clone, AsBytes, FromBytes, Debug)]
#[repr(C)]
struct TestType {
thing: [u8; 256],
other_thing: [u8; 32],
}
impl Default for TestType {
fn default() -> Self {
Self::zeroed()
}
}
impl TestType {
pub fn zeroed() -> Self {
unsafe { std::mem::zeroed() }
}
pub fn new() -> Self {
let mut newtype = Self::zeroed();
newtype.thing.fill_with(|| rand::random());
newtype.other_thing.fill_with(|| rand::random());
newtype
}
}
#[test]
fn test_polymorphism() {
const CAPACITY: usize = 4096;
let mut full = Box::new(IBLT::<TestType, CAPACITY, HASHES>::new());
let mut zero = Box::new(IBLT::<TestType, CAPACITY, HASHES>::new());
for _ in 0..CAPACITY {
zero.insert(&TestType::zeroed());
full.insert(&TestType::new());
}
zero.subtract(&full);
zero.list(|item, new| {
if new {
assert_eq!(item, TestType::zeroed());
} else {
assert_ne!(item, TestType::zeroed());
}
});
zero.reset();
full.reset();
for _ in 0..CAPACITY {
zero.insert(&TestType::zeroed());
full.insert(&TestType::new());
}
full.subtract(&zero);
full.list(|item, new| {
if new {
assert_ne!(item, TestType::zeroed());
} else {
assert_eq!(item, TestType::zeroed());
}
});
}
#[test]
fn test_to_from_bytes() {
let tmp = typical_iblt();
let mut tmp2 = IBLT::<[u8; 32], 16, 3>::from_bytes(tmp.as_bytes()).unwrap();
tmp2.subtract(&tmp);
tmp2.list(|_, new| assert!(!new));
}
#[test]
fn test_clone() {
let tmp = typical_iblt();
let mut tmp2 = tmp.clone();
tmp2.subtract(&tmp);
tmp2.list(|_, new| assert!(!new));
}
} }