acdream/src/AcDream.Content/DecodedTextureCache.cs
Erik 7b456b49d6 perf(diag): per-frame history export + checkpoint LOH/cache counters + soak capped mode (2026-07-24 audit review)
An adversarial performance review found our own instruments cannot
measure the project's own performance gates:

- FrameProfiler aggregated CPU/GPU/alloc/stage samples into ~5-second
  windows and reset the ring buffers after each report, so route-wide
  p50/p95/p99 distributions across a whole soak could not be
  reconstructed after the fact. ACDREAM_FRAME_HISTORY=<path> now opts
  into a separate per-frame history (one record per frame, ~72
  bytes/record, accumulated in memory with zero frame-thread I/O) that
  a shutdown-only Dispose() writes as CSV. The aggregated [frame-prof]
  report format and its existing metrics are unchanged.

- The canonical checkpoint JSON tracked cache residency (entry/byte
  counts) but never LOH size/fragmentation, process-wide allocated
  bytes, or cache hit/miss/eviction traffic — a committed audit JSON
  showed 65% LOH fragmentation that no tracked instrument recorded,
  and "does a revisit portal hit or miss the caches" was unanswerable
  from an artifact alone. WorldLifecycleResourceSnapshot now carries
  loh_size_bytes/loh_fragmentation_bytes (GCMemoryInfo.GenerationInfo
  index 3), process_total_allocated_bytes (GC.GetTotalAllocatedBytes),
  and Interlocked hit/miss/eviction counters for the CPU mesh cache,
  decoded-texture cache, and the four bounded DAT-object caches
  (portal/cell/highRes/language, aggregated).

- run-connected-r6-soak.ps1 unconditionally forced
  ACDREAM_UNCAPPED_RENDER=1 with no capped mode, while its sibling
  lifecycle-gate script correctly gated it behind a switch. Added
  -Uncapped (default capped, matching the sibling script's pattern),
  fixed the stationary dwell (12s -> 26s, past the 25s
  LiveEntityLivenessController deadline the adjacent comment already
  cited), and now write an env-disclosure.json into the automation
  artifact directory before every launch listing every ACDREAM_* var
  the script sets plus -Uncapped, since the prior audit could only see
  ACDREAM_DUMP_MOVE_TRUTH and nothing else was ever recorded anywhere.

Cache counters are wired via the existing composition path
(ObjectMeshManager already owns the CPU mesh cache and the mesh
extractor directly; content.Dats is threaded into
WorldLifecycleResourceSnapshotSource the same way every other
composition consumer receives it). The DAT-object cache lives behind
IDatReaderWriter, a third-party interface from the DatReaderWriter
package that cannot be extended; RuntimeDatCollection (the one
production implementation) exposes the aggregate stats directly and a
pattern match reads them, degrading to zero for any test double —
no new static registry was introduced (GpuMemoryTracker remains the
one precedented process-wide static).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(cherry picked from commit 1da2c33c875b41fa383dd79694ee2765f0e21896)
2026-07-24 12:00:30 +02:00

161 lines
5.4 KiB
C#

using System;
using System.Collections.Generic;
using System.Collections.Concurrent;
using System.Threading;
namespace AcDream.Content;
/// <summary>
/// Thread-safe, byte-bounded residency for canonical decoded texture pixels.
/// Mesh extraction may run on several workers, so cached arrays are immutable
/// after admission and callers clone them before applying surface-local alpha.
/// </summary>
internal sealed class DecodedTextureCache {
private sealed record Entry(DecodedTextureKey Key, byte[] Pixels);
private readonly object _gate = new();
private readonly Dictionary<DecodedTextureKey, LinkedListNode<Entry>> _entries = new();
private readonly LinkedList<Entry> _lru = new();
private readonly ConcurrentDictionary<DecodedTextureKey, Lazy<byte[]>> _inflight = new();
private readonly long _maximumBytes;
private readonly int _maximumEntries;
private long _residentBytes;
// Hit/miss/eviction traffic counters (2026-07-24 measurement-tooling
// review). Read via Stats without the _gate lock, so plain fields with
// Interlocked increments rather than lock-protected longs.
private long _hits;
private long _misses;
private long _evictions;
/// <summary>Cumulative hit/miss/eviction counts since construction.</summary>
public CacheStats Stats => new(
Interlocked.Read(ref _hits),
Interlocked.Read(ref _misses),
Interlocked.Read(ref _evictions));
public DecodedTextureCache(long maximumBytes, int maximumEntries) {
ArgumentOutOfRangeException.ThrowIfNegative(maximumBytes);
ArgumentOutOfRangeException.ThrowIfNegative(maximumEntries);
_maximumBytes = maximumBytes;
_maximumEntries = maximumEntries;
}
public int Count {
get {
lock (_gate) {
return _entries.Count;
}
}
}
public long ResidentBytes {
get {
lock (_gate) {
return _residentBytes;
}
}
}
public bool TryGet(DecodedTextureKey key, out byte[] pixels) {
lock (_gate) {
if (!_entries.TryGetValue(key, out var node)) {
Interlocked.Increment(ref _misses);
pixels = null!;
return false;
}
Interlocked.Increment(ref _hits);
Touch(node);
pixels = node.Value.Pixels;
return true;
}
}
/// <summary>
/// Returns the canonical pixels for <paramref name="key"/>. If another
/// worker admitted the same key first, its array wins. <paramref name="isCached"/>
/// reports whether the returned array is retained and therefore immutable.
/// </summary>
public byte[] RetainOrUse(DecodedTextureKey key, byte[] pixels, out bool isCached) {
ArgumentNullException.ThrowIfNull(pixels);
lock (_gate) {
if (_entries.TryGetValue(key, out var existing)) {
Touch(existing);
isCached = true;
return existing.Value.Pixels;
}
if (_maximumEntries == 0 || pixels.LongLength > _maximumBytes) {
isCached = false;
return pixels;
}
while (_lru.First is { } oldest
&& (_entries.Count >= _maximumEntries
|| _residentBytes + pixels.LongLength > _maximumBytes)) {
Remove(oldest);
}
var node = _lru.AddLast(new Entry(key, pixels));
_entries.Add(key, node);
_residentBytes += pixels.LongLength;
isCached = true;
return pixels;
}
}
/// <summary>
/// Decode a missing key once across concurrent mesh workers. The expensive
/// factory runs outside the LRU lock and unrelated keys proceed in parallel.
/// A failed or oversized decode is not stranded as permanent in-flight state.
/// </summary>
public byte[] GetOrCreate(
DecodedTextureKey key,
Func<byte[]> factory,
out bool isCached) {
ArgumentNullException.ThrowIfNull(factory);
if (TryGet(key, out byte[] cached)) {
isCached = true;
return cached;
}
var candidate = new Lazy<byte[]>(
factory,
LazyThreadSafetyMode.ExecutionAndPublication);
Lazy<byte[]> shared = _inflight.GetOrAdd(key, candidate);
try {
byte[] decoded = shared.Value;
return RetainOrUse(key, decoded, out isCached);
}
finally {
_inflight.TryRemove(
new KeyValuePair<DecodedTextureKey, Lazy<byte[]>>(key, shared));
}
}
private void Touch(LinkedListNode<Entry> node) {
if (node != _lru.Last) {
_lru.Remove(node);
_lru.AddLast(node);
}
}
private void Remove(LinkedListNode<Entry> node) {
_lru.Remove(node);
_entries.Remove(node.Value.Key);
_residentBytes -= node.Value.Pixels.LongLength;
Interlocked.Increment(ref _evictions);
}
}
/// <summary>
/// Every input that can change decoded RGBA belongs in the cache identity.
/// INDEX16/P8 clip-map conversion and A8 additive conversion are surface
/// dependent even when they reference the same RenderSurface DID.
/// </summary>
internal readonly record struct DecodedTextureKey(
uint RenderSurfaceId,
bool IsClipMap,
bool IsAdditive);