Add a standard key strategy for maps without prefixe

Remove the stupid "next" field
This commit is contained in:
Linus Björnstam 2026-04-16 19:49:31 +02:00
parent 7bea233edc
commit 4d87e30b40
3 changed files with 48 additions and 18 deletions

View file

@ -13,14 +13,45 @@ public interface IKeyStrategy<K>
{ {
int Compare(K x, K y); int Compare(K x, K y);
long GetPrefix(K key); long GetPrefix(K key);
bool UsesPrefixes => true; bool UsesPrefixes => true;
// //
bool IsLossless => false; bool IsLossless => false;
} }
/// <summary>
/// A universal key strategy for any type that relies on standard comparisons
/// (IComparable, IComparer, or custom StringComparers) without SIMD prefixes.
/// </summary>
public readonly struct StandardStrategy<K> : IKeyStrategy<K>
{
private readonly IComparer<K> _comparer;
// If no comparer is provided, it defaults to Comparer<K>.Default
// which automatically uses IComparable<K> if the type implements it.
public StandardStrategy(IComparer<K>? comparer = null)
{
_comparer = comparer ?? Comparer<K>.Default;
}
// Tell the B-Tree to skip SIMD routing and just use LinearSearch
public bool UsesPrefixes => false;
// This will never be called because UsesPrefixes is false,
// but we must satisfy the interface.
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public long GetPrefix(K key) => 0;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public int Compare(K x, K y)
{
return _comparer.Compare(x, y);
}
}
public struct UnicodeStrategy : IKeyStrategy<string> public struct UnicodeStrategy : IKeyStrategy<string>
{ {
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
@ -33,17 +64,17 @@ public struct UnicodeStrategy : IKeyStrategy<string>
// 1. Prepare Buffer (8 bytes) // 1. Prepare Buffer (8 bytes)
// stackalloc is virtually free (pointer bump) // stackalloc is virtually free (pointer bump)
Span<byte> utf8Bytes = stackalloc byte[8]; Span<byte> utf8Bytes = stackalloc byte[8];
// 2. Transcode (The "Safe" Magic) // 2. Transcode (The "Safe" Magic)
// This intrinsic handles ASCII efficiently and converts Surrogates/Chinese // This intrinsic handles ASCII efficiently and converts Surrogates/Chinese
// into bytes that maintain the correct "Magnitude" (Sort Order). // into bytes that maintain the correct "Magnitude" (Sort Order).
// Invalid surrogates become 0xEF (Replacement Char), which sorts > ASCII. // Invalid surrogates become 0xEF (Replacement Char), which sorts > ASCII.
System.Text.Unicode.Utf8.FromUtf16( System.Text.Unicode.Utf8.FromUtf16(
key.AsSpan(0, Math.Min(key.Length, 8)), key.AsSpan(0, Math.Min(key.Length, 8)),
utf8Bytes, utf8Bytes,
out _, out _,
out _, out _,
replaceInvalidSequences: true); // True ensures we get 0xEF for broken chars replaceInvalidSequences: true); // True ensures we get 0xEF for broken chars
// 3. Load as Big Endian Long // 3. Load as Big Endian Long
@ -63,7 +94,7 @@ public struct IntStrategy : IKeyStrategy<int>
[MethodImpl(MethodImplOptions.AggressiveInlining)] [MethodImpl(MethodImplOptions.AggressiveInlining)]
public long GetPrefix(int key) public long GetPrefix(int key)
{ {
// Pack the 32-bit int into the high 32-bits of the long. // Pack the 32-bit int into the high 32-bits of the long.
// This preserves sorting order when scanning the long array. // This preserves sorting order when scanning the long array.
// Cast to uint first to prevent sign extension confusion during the shift, // Cast to uint first to prevent sign extension confusion during the shift,
@ -92,13 +123,13 @@ public struct DoubleStrategy : IKeyStrategy<double>
// -Negative Max -> 0 // -Negative Max -> 0
// -0 -> Midpoint // -0 -> Midpoint
// +Negative Max -> Max // +Negative Max -> Max
long mask = (bits >> 63); // 0 for positive, -1 (All 1s) for negative long mask = (bits >> 63); // 0 for positive, -1 (All 1s) for negative
// If negative: bits ^ -1 = ~bits (Flip All) // If negative: bits ^ -1 = ~bits (Flip All)
// If positive: bits ^ 0 = bits (Flip None) // If positive: bits ^ 0 = bits (Flip None)
// Then we toggle the sign bit (0x8000...) to shift the range to signed long. // Then we toggle the sign bit (0x8000...) to shift the range to signed long.
return (bits ^ (mask & 0x7FFFFFFFFFFFFFFF)) ^ unchecked((long)0x8000000000000000); return (bits ^ (mask & 0x7FFFFFFFFFFFFFFF)) ^ unchecked((long)0x8000000000000000);
} }
} }
@ -120,7 +151,7 @@ public static class PrefixScanner
//if (targetPrefix == long.MinValue) //if (targetPrefix == long.MinValue)
//{ //{
// return 0; // return 0;
//} //}
// Fallback for short arrays or unsupported hardware // Fallback for short arrays or unsupported hardware
if (!Avx2.IsSupported || prefixes.Length < 4) if (!Avx2.IsSupported || prefixes.Length < 4)

View file

@ -91,7 +91,6 @@ public sealed class LeafNode<K, V> : Node<K>
public K[]? Keys; public K[]? Keys;
public V[] Values; public V[] Values;
public LeafNode<K, V>? Next;
internal long[]? _prefixes; internal long[]? _prefixes;
@ -111,9 +110,7 @@ public sealed class LeafNode<K, V> : Node<K>
{ {
Keys = new K[Capacity]; Keys = new K[Capacity];
Values = new V[Capacity]; Values = new V[Capacity];
Header.Count = original.Header.Count; Header.Count = original.Header.Count; _prefixes = new long[Capacity];
Next = original.Next;
_prefixes = new long[Capacity];
// Copy data // Copy data
Array.Copy(original.Keys, Keys, original.Header.Count); Array.Copy(original.Keys, Keys, original.Header.Count);
@ -311,4 +308,4 @@ public readonly struct OwnerId(uint id, ushort gen) : IEquatable<OwnerId>
{ {
return !left.Equals(right); return !left.Equals(right);
} }
} }

View file

@ -1,4 +1,4 @@
* NiceBtree (PersistentMap) * PersistentMap
A high-performance, persistent (Copy-on-Write) B+ Tree implemented in C#. A high-performance, persistent (Copy-on-Write) B+ Tree implemented in C#.
@ -14,6 +14,8 @@ It is designed for zero-overhead reads, SIMD-accelerated key routing, and alloca
** When should I use this? ** When should I use this?
Never, probably. This was just a fun little project. If you want a really fast immutable sorted map you should consider it. Despite this map being faster than LanguageExt.HashMap for some key types, you should definitely use that if you don't need a sorted collection. It is well tested and does not have any problems key collisions, which will slow this map down by a lot. Never, probably. This was just a fun little project. If you want a really fast immutable sorted map you should consider it. Despite this map being faster than LanguageExt.HashMap for some key types, you should definitely use that if you don't need a sorted collection. It is well tested and does not have any problems key collisions, which will slow this map down by a lot.
The general version of this, using =StandardStrategy<K>= does not benefit from the prefix optimization.
** Quick Start ** Quick Start
*** 1. Basic Immutable Usage *** 1. Basic Immutable Usage