Skip to content

Isurwars/Correlation / .agents/skills/benchmarking

benchmarking

Verified

Enforce data-oriented design (SoA vs AoS), SIMD vectorization, Google Benchmark microbenchmarks, and cache-line alignment. Use when profiling or optimizing throughput and latency.

Microbenchmarking & Performance Optimization

This skill governs high-performance C++ execution, memory-locality optimizations, SIMD vectorization contracts, and microbenchmarking protocols across Correlation.

1. Memory Locality & Data-Oriented Design (DOD)

  1. Contiguous Memory First: Prefer std::vector, std::array, or contiguous std::span buffers over pointer-chasing structures (std::list, raw pointer arrays, node-based trees).
  2. SoA vs AoS Layout Selection:
    • Array of Structures (AoS): Suitable for scalar access or non-vectorized operations (struct Atom { Vector3D pos; Vector3D vel; double mass; };).
    • Structure of Arrays (SoA): Required for high-throughput SIMD vector loops (struct AtomBuffer { std::vector<float> x, y, z; };) to ensure stride-1 sequential memory access and alignment.
  3. Cache Line Alignment (alignas(64)):
    • Align thread-local accumulator structures to 64 bytes (alignas(64)) to eliminate false sharing in multi-threaded OpenMP/TBB reduction loops.
// GOOD: Cache-line aligned thread-local accumulator
struct alignas(64) ThreadLocalHistogram {
    std::vector<uint64_t> bins;
};

2. SIMD Vectorization Contracts

  1. Vector-Friendly Loops: Keep inner distance/histogram computation loops free of conditional branching (if/else), virtual function calls, or complex control flow.
  2. Contiguous Access: Ensure pointer reads inside inner loops process contiguous memory blocks without strided pointer indirection.
  3. Use SIMD Intrinsics / Wrappers: Prefer vectorized SIMD wrappers (Vector3SIMD, dot_block) or standard C++23 std::span over custom assembly instructions unless profiling explicitly demonstrates auto-vectorization failure.

3. Microbenchmarking Protocol (Google Benchmark)

  1. Dead-Code Elimination Guard: Wrap benchmarked output variables in benchmark::DoNotOptimize(...) and call benchmark::ClobberMemory() to prevent compiler optimization passes from discarding calculated values.
#include <benchmark/benchmark.h>

static void BM_RadialDistribution(benchmark::State& state) {
    auto trajectory = create_test_trajectory();
    RDFCalculator calculator;
    
    for (auto _ : state) {
        auto result = calculator.calculate(trajectory);
        benchmark::DoNotOptimize(result);
        benchmark::ClobberMemory();
    }
}
BENCHMARK(BM_RadialDistribution);
  1. Isolation & Statistical Rigor:
    • Run benchmarks under Release builds with target architecture flags (-O3 -march=native).
    • Ensure benchmark targets are reproducible and isolated from external I/O operations.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 319kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 196kUpdated Claude CodeCodexVerifiedInstall