Cljr Performance Optimization

This skill guides performance optimization for the Cljr Clojure-to-.NET compiler, targeting .NET 10 with C# 14 preview features.

When to use this skill

Optimizing emitter-generated C# code
Improving runtime collection performance
Reducing memory allocations
Leveraging .NET 10 / C# 14 performance features
Writing or analyzing benchmarks
Optimizing REPL evaluation performance

Project Performance Profile

Target Framework

Runtime: .NET 10 (LTS)
Language: C# 14 preview (Runtime, CLI, REPL) / C# 12 (Compiler, SourceGenerator)

Existing Optimizations

| Optimization | Status | Location | |--------------|--------|----------| | Span<T> / ReadOnlySpan<T> | ✓ | PersistentVector | | FrozenDictionary | ✓ | Protocol, MethodImplCache | | [AggressiveInlining] | ✓ | Hot paths throughout | | Symbol/Keyword interning | ✓ | Symbol.cs, Keyword.cs | | HAMT with SIMD PopCount | ✓ | PersistentHashMap | | 32-way trie | ✓ | PersistentVector | | Transient collections | ✓ | Vector, HashMap batch ops | | Method dispatch caching | ✓ | MultiFn, Protocol | | BenchmarkDotNet suite | ✓ | tests/Cljr.Benchmarks/ |

Opportunities for Further Optimization

| Optimization | Status | Potential Use | |--------------|--------|---------------| | stackalloc | ✗ | Small temp arrays in hot paths | | ArrayPool<T> | ✗ | Reusable buffer arrays | | SearchValues<T> | ✗ | Character/byte searching | | CompositeFormat | ✗ | String formatting | | C# 14 implicit span conversions | ✗ | Cleaner span APIs | | AVX10.2 intrinsics | ✗ | SIMD operations |

Performance Optimization Workflow

Step 1: Measure First

Never optimize without benchmarks. Run existing benchmarks:

cd tests/Cljr.Benchmarks
dotnet run -c Release -- --filter "*"

Or specific benchmarks:

dotnet run -c Release -- --filter "VectorBenchmarks"
dotnet run -c Release -- --filter "*Conj*"

Step 2: Identify Hot Paths

Use profiling or benchmark results to identify:

High-frequency operations
Allocation-heavy code paths
Cache miss patterns

Step 3: Apply Optimizations

See reference files for specific patterns:

Step 4: Verify Improvement

Re-run benchmarks and compare:

dotnet run -c Release -- --filter "YourBenchmark" --runtimes net10.0

Quick Reference: Key Optimizations

AggressiveInlining

For small, frequently-called methods:

[MethodImpl(MethodImplOptions.AggressiveInlining)]
public T GetValue() => _value;

Span<T> for Zero-Copy

// Instead of creating arrays
public ReadOnlySpan<object?> ArrayFor(int i) => _tail.AsSpan();

// Span-based construction (already in PersistentVector)
public static PersistentVector Create(ReadOnlySpan<object?> items)

FrozenDictionary for Read-Heavy Lookups

using System.Collections.Frozen;

// Mutable during construction
private readonly ConcurrentDictionary<Type, Cache> _impls = new();
// Frozen for fast reads
private volatile FrozenDictionary<Type, Cache>? _frozen;

public void Freeze() => _frozen = _impls.ToFrozenDictionary();

Symbol/Keyword Interning

Already implemented - ensures reference equality:

// Good: reference equality after interning
if (ReferenceEquals(sym1, sym2)) { ... }

// Avoid: value equality is slower
if (sym1.Equals(sym2)) { ... }

Transients for Batch Operations

// Slow: O(n²) allocations
var vec = PersistentVector.Empty;
foreach (var item in items)
    vec = vec.Conj(item);

// Fast: O(n) with transient
var transient = PersistentVector.Empty.AsTransient();
foreach (var item in items)
    transient.ConjBang(item);
return transient.Persistent();

Key Files

| Area | Files | |------|-------| | Runtime Collections | src/Cljr.Runtime/Collections/PersistentVector.cs, PersistentHashMap.cs | | Protocol Dispatch | src/Cljr.Runtime/Protocol.cs, MultiFn.cs | | Symbol Interning | src/Cljr.Runtime/Symbol.cs, Keyword.cs | | Emitter | src/Cljr.Compiler/Emitter/CSharpEmitter.cs | | Benchmarks | tests/Cljr.Benchmarks/CollectionBenchmarks.cs |

References

.NET 10 / C# 14 Features - Latest platform features
Emitter Patterns - Optimized C# code generation
Benchmarking Guide - How to measure and compare

cljr-performance