返回 Skill 列表
extension
分类: 开发与工程无需 API Key

profile-engine

对Rust模拟引擎进行性能分析,以识别性能瓶颈。在优化引擎性能或调查慢代码路径时使用。

person作者: jakexiaohubgithub

Profile Engine

Profile the Rust simulation engine using profile.py to identify hot paths.

Quick Start

cd ./crates/engine
./profile.py 5000          # Profile with 5000 iterations
./profile.py 10000 --top 30  # More iterations, show top 30 functions
./profile.py --json        # Output as JSON for parsing

Prerequisites

cargo install samply
brew install binutils  # Optional: better symbol resolution

What profile.py Does

  1. Builds the engine with debug symbols
  2. Runs samply profiler on the benchmark binary
  3. Parses the profile and resolves symbols
  4. Shows hot modules and functions by self time

Reading the Output

HOT MODULES (by self time)
─────────────────────────────────────────────────────────────────
12.5% █████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ engine::sim::engine

HOT FUNCTIONS (by self time, excluding std/core)
─────────────────────────────────────────────────────────────────
 Self  Total  Function                                           Location
───── ─────  ──────────────────────────────────────────────────  ────────────
 8.2%  45.3%  run_simulation                                     engine.rs:123
 4.1%  12.5%  evaluate_rotation                                  engine.rs:456
  • Self time: Time spent in the function itself (optimization target)
  • Total time: Time including all called functions
  • High self time = hot path to optimize

Current Hot Paths

  1. run_simulation - main simulation loop
  2. handle_gcd_ready - GCD event processing
  3. evaluate_rotation - spell availability checks
  4. cast_spell - spell execution
  5. EventQueue::push/pop - priority queue operations

Baseline Performance

  • 300s fights, 5000 iterations
  • ~0.26M sims/sec (optimized build)
  • ~0.22M sims/sec (profiling build with debug symbols)

Advanced Usage

Manual samply (with browser UI)

cd ./crates/engine
RUSTFLAGS="-C force-frame-pointers=yes" cargo build --release
samply record --rate 4000 ./target/release/bench 5000

Opens Firefox Profiler with interactive flame graph.

Production Build Settings

After profiling, ensure production settings in Cargo.toml:

[profile.release]
lto = true
codegen-units = 1
panic = "abort"

Sources