Codebase Status: Synced with repository as of 2026-02-04.

Comprehensive performance benchmarks for the ABI framework, measuring throughput, latency, and resource utilization across all major subsystems.

# Run all benchmark suites
zig build benchmarks

# Run all benchmark suites (including competitive)
zig build bench-all

# Run specific suite
zig build benchmarks -- --suite=simd

# Quick mode (reduced iterations)
zig build benchmarks -- --quick

# Verbose output
zig build benchmarks -- --verbose

Tests vectorized operations using SIMD intrinsics:

• Dot product (single/batch)
• Matrix multiplication
• L2 norm computation
• Cosine similarity
• Distance calculations (Euclidean, Manhattan)

zig build benchmarks -- --suite=simd

Measures allocator performance:

• General purpose allocator throughput
• Arena allocator patterns
• Pool allocator efficiency
• Fragmentation under stress
• Memory pressure handling

zig build benchmarks -- --suite=memory

Tests lock-free data structures:

• Lock-free queue throughput
• Work-stealing deque performance
• Atomic counter operations
• MPMC queue contention
• Thread pool scaling

zig build benchmarks -- --suite=concurrency

WDBX vector database benchmarks:

• Vector insertion (single/batch)
• Linear search performance
• HNSW approximate search
• Concurrent search operations
• Cache-aligned memory access
• Memory prefetching effectiveness

zig build benchmarks -- --suite=database

Network protocol benchmarks:

• HTTP header parsing
• JSON encoding/decoding
• WebSocket frame processing
• Request routing overhead

zig build benchmarks -- --suite=network

Cryptographic operation benchmarks:

• SHA-256/SHA-512 hashing
• AES-256 encryption
• HMAC computation
• Key derivation (PBKDF2, Argon2)
• Random number generation

zig build benchmarks -- --suite=crypto

Machine learning operation benchmarks:

• GEMM (General Matrix Multiply)
• Attention mechanism
• Activation functions (ReLU, GELU, SiLU)
• Softmax computation
• Layer normalization

zig build benchmarks -- --suite=ai

GPU kernel benchmarks:

• Matmul, vector ops, reductions
• Backend comparisons
• GPU vs CPU comparisons

zig build benchmarks -- --suite=gpu

Compare ABI performance against industry-standard implementations:

# Run competitive benchmarks
zig build bench-competitive

# With custom dataset size
zig build bench-competitive -- --vectors=100000 --dims=768

Results are output as JSON for easy integration with CI/CD pipelines.

zig build benchmarks -- [OPTIONS]

OPTIONS:
  --suite=<name>    Run specific suite (simd, memory, concurrency, database, network, crypto, ai, gpu)
  --quick           Run with reduced iterations
>>>>>>> origin/cursor/ai-module-source-organization-0282
  --verbose         Show detailed output
  --json            Output results as JSON to stdout
  --output=<file>   Write JSON report to a file

# All suites with verbose output
zig build benchmarks -- --verbose

# Database benchmarks only
zig build benchmarks -- --suite=database

# Quick verification run
zig build benchmarks -- --quick

# JSON output for CI integration
zig build benchmarks -- --output=benchmark_results.json

• ops/sec: Operations per second (higher is better)
• MB/sec or GB/sec: Data throughput (higher is better)
• GFLOPS: Billion floating-point operations per second

• μs (microseconds): 1/1,000,000 second
• ns (nanoseconds): 1/1,000,000,000 second
• p50/p99: Percentile latencies

• RSS: Resident Set Size (actual memory usage)
• fragmentation %: Wasted memory due to allocation patterns
• allocs/sec: Allocation rate

Baseline reports are stored under benchmarks/baselines/ (see benchmarks/baselines/README.md). After significant changes, generate a fresh JSON report and store it under the appropriate branch or release directory:

# Generate a new baseline report
zig build benchmarks -- --output=benchmarks/baselines/branches/my_branch.json

New benchmarks should follow this pattern:

const BenchmarkSuite = @import("mod.zig").BenchmarkSuite;

pub fn run(allocator: std.mem.Allocator) !void {
    var suite = BenchmarkSuite.init(allocator, "My Suite");
    defer suite.deinit();

    suite.benchmark("operation_name", struct {
        fn bench() void {
            // Operation to benchmark
        }
    }.bench, .{});

    suite.report();
}

• Disable CPU frequency scaling: sudo cpupower frequency-set -g performance
• Close background applications
• Run multiple iterations and average
• Use --quick for initial verification

• Increase iteration count with --iterations=N
• Check for thermal throttling
• Ensure consistent memory pressure

# Ensure all dependencies are available
zig build benchmarks -Denable-database=true -Denable-gpu=true

• benchmarks/baselines/README.md - Baseline format and CI flow
• docs/content/gpu.html - GPU benchmarking guide
• CLAUDE.md - Development guidelines