Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: initial implementation #1

Merged
merged 1 commit into from
Oct 3, 2024
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
3 changes: 2 additions & 1 deletion Nargo.toml
Original file line number Diff line number Diff line change
@@ -1,7 +1,8 @@
[package]
name = "noir_library"
name = "sparse_array"
type = "lib"
authors = [""]
compiler_version = ">=0.34.0"

[dependencies]
sort = {tag = "v0.1.0", git = "https://github.com/noir-lang/noir_sort"}
52 changes: 34 additions & 18 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,20 +1,7 @@
# noir-library-starter
# sparse_array

This repository is a template used by the noir-lang org when creating internally maintained libraries.
Noir library that implements efficient sparse arrays, both constant (SparseArray) and mutable (MutSparseArray)

This provides out of the box:

- A simple CI setup to test and format the library
- A canary flagging up compilation failures on nightly releases.
- A [release-please](https://github.com/googleapis/release-please) setup to ease creating releases for the library.

Feel free to use this template as a starting point to create your own Noir libraries.

---

# LIBRARY_NAME

Add a brief description of the library

## Benchmarks

Expand All @@ -27,11 +14,40 @@ In your _Nargo.toml_ file, add the version of this library you would like to ins

```
[dependencies]
LIBRARY = { tag = "v0.1.0", git = "https://github.com/noir-lang/LIBRARY_NAME" }
sparse_array = { tag = "v0.1.0", git = "https://github.com/noir-lang/sparse_array" }
```

## `library`
## `sparse_array`

### Usage

`PLACEHOLDER`
```rust
use dep::sparse_array::{SparseArray, MutSparseArray}

// a sparse array of size 10,000 with 10 nonzero values
fn example_sparse_array(nonzero_indices: [Field; 10], nonzero_values: [Field; 10]) {
let sparse_array_size = 10000;
let array: SparseArray<10, Field> = SparseArray::create(nonzero_indices, nonzero_values, sparse_array_size);

assert(array.get(999) == 12345);
}

// a mutable sparse array that can contain up to 10 nonzero values
fn example_mut_sparse_array(initial_nonzero_indices: [Field; 9], initial_nonzero_values: [Field; 9]) {
let sparse_array_size = 10000;
let mut array: MutSparseArray<10, Field> = MutSparseArray::create(nonzero_indices, nonzero_values, sparse_array_size);

// update element 1234 to contain value 9999
array.set(1234, 9999);

// error, array can only contain 10 nonzero values
array.ser(10, 888);
}
```

# Costs

Constructing arrays is proportional to the number of nonzero entries in the array and very small ~10 gates per element (plus the cost of initializing range tables if not already done so)

Reading from `SparseArray` is 14.5 gates
Reading and writing to `MutSparseArray` is ~30 gates
276 changes: 272 additions & 4 deletions src/lib.nr
Original file line number Diff line number Diff line change
@@ -1,5 +1,273 @@
/// This doesn't really do anything by ensures that there is a test for CI to run.
#[test]
fn smoke_test() {
assert(true);
mod mut_sparse_array;
use dep::sort::sort_advanced;

unconstrained fn __sort_field_as_u32(lhs: Field, rhs: Field) -> bool {
// lhs.lt(rhs)
lhs as u32 < rhs as u32
}

fn assert_sorted(lhs: Field, rhs: Field) {
let result = (rhs - lhs - 1);
result.assert_max_bit_size(32);
}

/**
* @brief MutSparseArray, a sparse array of configurable size with `N` nonzero entries.
* Can be read from and written into
*
* @param keys is size N+2 because we want to always ensure that,
* for any valid index, there is some X where `keys[X] <= index <= keys[X+1]`
* when constructing, we will set keys[0] = 0, and keys[N-1] = maximum - 1
* @param values is size N+3 because of the following:
* 1. keys[i] maps to values[i+1]
* 2. values[0] is an empty object. when calling `get(idx)`, if `idx` is not in `keys` we will return `values[0]`
**/
struct MutSparseArrayBase<let N: u32, T, ComparisonFuncs>
{
values: [T; N + 3],
keys: [Field; N + 2],
linked_keys: [Field; N + 2],
tail_ptr: Field,
maximum: Field
}

struct U32RangeTraits {
}

struct MutSparseArray<let N: u32, T>
{
inner: MutSparseArrayBase<N, T, U32RangeTraits>
}
/**
* @brief SparseArray, stores a sparse array of up to size 2^32 with `N` nonzero entries
* SparseArray is constant i.e. values canot be inserted after creation.
* See MutSparseArray for a mutable version (a bit more expensive)
* @param keys is size N+2 because we want to always ensure that,
* for any valid index, there is some X where `keys[X] <= index <= keys[X+1]`
* when constructing, we will set keys[0] = 0, and keys[N-1] = maximum - 1
* @param values is size N+3 because of the following:
* 1. keys[i] maps to values[i+1]
* 2. values[0] is an empty object. when calling `get(idx)`, if `idx` is not in `keys` we will return `values[0]`
**/
struct SparseArray<let N: u32, T> {
keys: [Field; N + 2],
values: [T; N + 3],
maximum: Field // can be up to 2^32
}
impl<let N: u32, T> SparseArray<N, T> where T : std::default::Default {

/**
* @brief construct a SparseArray
**/
fn create(_keys: [Field; N], _values: [T; N], size: Field) -> Self {
let _maximum = size - 1;
let mut r: Self = SparseArray { keys: [0; N + 2], values: [T::default(); N + 3], maximum: _maximum };

// for any valid index, we want to ensure the following is satified:
// self.keys[X] <= index <= self.keys[X+1]
// this requires us to sort hte keys, and insert a startpoint and endpoint
let sorted_keys = sort_advanced(_keys, __sort_field_as_u32, assert_sorted);

// insert start and endpoints
r.keys[0] = 0;
for i in 0..N {
r.keys[i+1] = sorted_keys.sorted[i];
}
r.keys[N+1] = _maximum;

// populate values based on the sorted keys
// note: self.keys[i] maps to self.values[i+1]
// self.values[0] does not map to any key. we use it to store the default empty value,
// which is returned when `get(idx)` is called and `idx` does not exist in `self.keys`
for i in 0..N {
r.values[i+2] = _values[sorted_keys.sort_indices[i]];
}
// insert values that map to our key start and endpoints
// if _keys[0] = 0 then values[0] must equal _values[0], so some conditional logic is required
// (same for _keys[N-1])
let mut initial_value = T::default();
if (_keys[0] == 0) {
initial_value = _values[0];
}
let mut final_value = T::default();
if (_keys[N - 1] == _maximum) {
final_value = _values[N-1];
}
r.values[1] = initial_value;
r.values[N+2] = final_value;

// perform boundary checks!
// the maximum size of the sparse array is 2^32
// we need to check that every element in `self.keys` is less than 2^32
// because `self.keys` is sorted, we can simply validate that
// sorted_keys.sorted[0] < 2^32
// sorted_keys.sorted[N-1] < maximum
sorted_keys.sorted[0].assert_max_bit_size(32);
_maximum.assert_max_bit_size(32);
(_maximum - sorted_keys.sorted[N - 1]).assert_max_bit_size(32);
r
}

/**
* @brief determine whether `target` is present in `self.keys`
* @details if `found == false`, `self.keys[found_index] < target < self.keys[found_index + 1]`
**/
unconstrained fn search_for_key(self, target: Field) -> (Field, Field) {
let mut found = false;
let mut found_index = 0;
let mut previous_less_than_or_equal_to_target = false;
for i in 0..N + 2 {
// if target = 0xffffffff we need to be able to add 1 here, so use u64
let current_less_than_or_equal_to_target = self.keys[i] as u64 <= target as u64;
if (self.keys[i] == target) {
found = true;
found_index = i as Field;
break;
}
if (previous_less_than_or_equal_to_target & !current_less_than_or_equal_to_target) {
found_index = i as Field - 1;
break;
}
previous_less_than_or_equal_to_target = current_less_than_or_equal_to_target;
}
(found as Field, found_index)
}

/**
* @brief return element `idx` from the sparse array
* @details cost is 14.5 gates per lookup
**/
fn get(self, idx: Field) -> T {
let (found, found_index) = unsafe {
self.search_for_key(idx)
};
// bool check. 0.25 gates cheaper than a raw `bool` type. need to fix at some point
assert(found * found == found);

// OK! So we have the following cases to check
// 1. if `found` then `self.keys[found_index] == idx`
// 2. if `!found` then `self.keys[found_index] < idx < self.keys[found_index + 1]
// how do we simplify these checks?
// case 1 can be converted to `self.keys[found_index] <= idx <= self.keys[found_index]
// case 2 can be modified to `self.keys[found_index] + 1 <= idx <= self.keys[found_index + 1] - 1
// combine the two into the following single statement:
// `self.keys[found_index] + 1 - found <= idx <= self.keys[found_index + 1 - found] - 1 + found
let lhs = self.keys[found_index];
let rhs = self.keys[found_index + 1 - found];
let lhs_condition = idx - lhs - 1 + found;
let rhs_condition = rhs - 1 + found - idx;
lhs_condition.assert_max_bit_size(32);
rhs_condition.assert_max_bit_size(32);

// self.keys[i] maps to self.values[i+1]
// however...if we did not find a non-sparse entry, we want to return self.values[0] (the default value)
let value_index = (found_index + 1) * found;
self.values[value_index]
}
}

mod test {

use crate::SparseArray;
#[test]
fn test_sparse_lookup() {
let example = SparseArray::create([1, 99, 7, 5], [123, 101112, 789, 456], 100);

assert(example.get(1) == 123);
assert(example.get(5) == 456);
assert(example.get(7) == 789);
assert(example.get(99) == 101112);

for i in 0..100 {
if ((i != 1) & (i != 5) & (i != 7) & (i != 99)) {
assert(example.get(i as Field) == 0);
}
}
}

#[test]
fn test_sparse_lookup_boundary_cases() {
// what about when keys[0] = 0 and keys[N-1] = 2^32 - 1?
let example = SparseArray::create(
[0, 99999, 7, 0xffffffff],
[123, 101112, 789, 456],
0x100000000
);

assert(example.get(0) == 123);
assert(example.get(99999) == 101112);
assert(example.get(7) == 789);
assert(example.get(0xffffffff) == 456);
assert(example.get(0xfffffffe) == 0);
}

#[test(should_fail_with = "call to assert_max_bit_size")]
fn test_sparse_lookup_overflow() {
let example = SparseArray::create([1, 5, 7, 99999], [123, 456, 789, 101112], 100000);

assert(example.get(100000) == 0);
}

#[test(should_fail_with = "call to assert_max_bit_size")]
fn test_sparse_lookup_boundary_case_overflow() {
let example = SparseArray::create([0, 5, 7, 0xffffffff], [123, 456, 789, 101112], 0x100000000);

assert(example.get(0x100000000) == 0);
}

#[test(should_fail_with = "call to assert_max_bit_size")]
fn test_sparse_lookup_key_exceeds_maximum() {
let example = SparseArray::create([0, 5, 7, 0xffffffff], [123, 456, 789, 101112], 0xffffffff);
assert(example.maximum == 0xffffffff);
}
#[test]
fn test_sparse_lookup_u32() {
let example = SparseArray::create(
[1, 99, 7, 5],
[123 as u32, 101112 as u32, 789 as u32, 456 as u32],
100
);

assert(example.get(1) == 123);
assert(example.get(5) == 456);
assert(example.get(7) == 789);
assert(example.get(99) == 101112);

for i in 0..100 {
if ((i != 1) & (i != 5) & (i != 7) & (i != 99)) {
assert(example.get(i as Field) == 0);
}
}
}

struct F {
foo: [Field; 3]
}
impl std::cmp::Eq for F {
fn eq(self, other: Self) -> bool {
self.foo == other.foo
}
}

impl std::default::Default for F {
fn default() -> Self {
F { foo: [0; 3] }
}
}

#[test]
fn test_sparse_lookup_struct() {
let values = [F { foo: [1, 2, 3] }, F { foo: [4, 5, 6] }, F { foo: [7, 8, 9] }, F { foo: [10, 11, 12] }];
let example = SparseArray::create([1, 99, 7, 5], values, 100000);

assert(example.get(1) == values[0]);
assert(example.get(5) == values[3]);
assert(example.get(7) == values[2]);
assert(example.get(99) == values[1]);
for i in 0..100 {
if ((i != 1) & (i != 5) & (i != 7) & (i != 99)) {
assert(example.get(i as Field) == F::default());
}
}
}
}
Loading
Loading