+ Stacker: fragmented clusters read, layout measured against dmsdos

Compression.Core/Dictionary/StackerSd4/StackerSd4BuildingBlock.cs

@@ -0,0 +1,39 @@
+using System.Buffers.Binary;
+using Compression.Core.FileSystems.Cvf;
+using Compression.Registry;
+
+namespace Compression.Core.Dictionary.StackerSd4;
+
+/// <summary>
+/// Stacker 4's SD-4 codec as a raw primitive: LZ77 over a 2720-byte window, Huffman-coded with a
+/// run-length coded code table (<c>docs/CVF-CODECS-ON-DISK.md</c>). The building-block wrapper stores
+/// the uncompressed size (32-bit little-endian) before one SD-4 stream covering the whole input.
+/// </summary>
+public sealed class StackerSd4BuildingBlock : IBuildingBlock {
+  /// <inheritdoc/>
+  public string Id => "BB_StackerSd4";
+  /// <inheritdoc/>
+  public string DisplayName => "Stacker SD-4";
+  /// <inheritdoc/>
+  public string Description => "Stac Stacker 4 cluster codec: LZ77 (2720-byte window) with canonical Huffman codes and a run-length coded table";
+  /// <inheritdoc/>
+  public AlgorithmFamily Family => AlgorithmFamily.Dictionary;
+
+  /// <inheritdoc/>
+  public byte[] Compress(ReadOnlySpan<byte> data) {
+    var stream = Sd4Codec.Encode(data, level: 3);
+    var result = new byte[4 + stream.Length];
+    BinaryPrimitives.WriteInt32LittleEndian(result, data.Length);
+    stream.CopyTo(result, 4);
+    return result;
+  }
+
+  /// <inheritdoc/>
+  public byte[] Decompress(ReadOnlySpan<byte> data) {
+    if (data.Length < 4) throw new InvalidDataException("Stacker SD-4: missing size header.");
+    var size = BinaryPrimitives.ReadInt32LittleEndian(data);
+    if (size < 0) throw new InvalidDataException("Stacker SD-4: negative size.");
+    var stream = data[4..].ToArray();
+    return Sd4Codec.Decode(stream, stream.Length, size);
+  }
+}

Compression.Core/FileSystems/Cvf/ClusterMatchFinder.cs

@@ -0,0 +1,81 @@
+namespace Compression.Core.FileSystems.Cvf;
+
+/// <summary>
+/// Hash-chain LZ77 match finder over one cluster, shared by the DS/JM and SD-4 encoders. Every match
+/// it reports lies wholly inside the cluster: no CVF codec carries a dictionary from one cluster to
+/// the next.
+/// </summary>
+internal sealed class ClusterMatchFinder {
+  private const int HashBits = 13;
+  private const int MinLength = 3;
+
+  private readonly byte[] _data;
+  private readonly int[] _head = new int[1 << HashBits];
+  private readonly int[] _previous;
+  private readonly int _maxDistance;
+  private readonly int _maxLength;
+  private readonly int _maxChain;
+  private readonly int _niceLength;
+  private int _inserted;
+
+  /// <param name="data">The cluster.</param>
+  /// <param name="maxDistance">The farthest distance the format can express.</param>
+  /// <param name="maxLength">The longest match the format can express.</param>
+  /// <param name="level">Search effort, 1 (fastest) and up: how many earlier positions with the same
+  /// hash are tried, and how long a match must be to end the search early.</param>
+  public ClusterMatchFinder(byte[] data, int maxDistance, int maxLength, int level) {
+    this._data = data;
+    this._previous = new int[data.Length];
+    this._maxDistance = maxDistance;
+    this._maxLength = maxLength;
+    (this._maxChain, this._niceLength) = level switch {
+      <= 1 => (16, 64),
+      2 => (64, 258),
+      3 => (256, 1024),
+      _ => (1024, int.MaxValue),
+    };
+    Array.Fill(this._head, -1);
+  }
+
+  /// <summary>
+  /// The longest match for the bytes at <paramref name="position"/>, preferring the nearest on ties;
+  /// <c>(0, 0)</c> when there is none of at least three bytes. Positions before
+  /// <paramref name="position"/> are indexed first, so calls must come in ascending order.
+  /// </summary>
+  /// <param name="position">Where the match starts.</param>
+  /// <param name="end">Exclusive bound the match may not run past; the cluster end when omitted.</param>
+  public (int Length, int Distance) Find(int position, int end = int.MaxValue) {
+    this.IndexUpTo(position);
+    var data = this._data;
+    var limit = Math.Min(this._maxLength, Math.Min(data.Length, end) - position);
+    if (limit < MinLength) return (0, 0);
+
+    int bestLength = 0, bestDistance = 0;
+    var candidate = this._head[Hash(data, position)];
+    for (var chain = this._maxChain; candidate >= 0 && chain > 0; --chain, candidate = this._previous[candidate]) {
+      var distance = position - candidate;
+      if (distance > this._maxDistance) break;
+      if (data[candidate + bestLength] != data[position + bestLength]) continue;
+
+      var length = 0;
+      while (length < limit && data[candidate + length] == data[position + length]) ++length;
+      if (length <= bestLength) continue;
+
+      (bestLength, bestDistance) = (length, distance);
+      if (length == limit || length >= this._niceLength) break;
+    }
+    return bestLength >= MinLength ? (bestLength, bestDistance) : (0, 0);
+  }
+
+  private void IndexUpTo(int position) {
+    var last = Math.Min(position, this._data.Length - MinLength + 1);
+    for (; this._inserted < last; ++this._inserted) {
+      var h = Hash(this._data, this._inserted);
+      this._previous[this._inserted] = this._head[h];
+      this._head[h] = this._inserted;
+    }
+  }
+
+  private static int Hash(byte[] data, int position)
+    => (int)(((uint)data[position] << 16 | (uint)data[position + 1] << 8 | data[position + 2]) * 2654435761u >> (32 - HashBits));
+}