add profiling infra: bench profile, profile feature, inline(never) attrs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: PeaBrane

CLAUDE.md

@@ -14,3 +14,8 @@
 | metropolis + PT | 0.081 |
 
 Run: `.venv/bin/python benchmarks/sweep_modes.py`
+
+## Performance notes
+
+- Bottleneck is cache pressure + dependent load chains, not compute — see `refs/cache-optimization.md`
+- Sweep ordering options (checkerboard, typewriter, random, etc.) — see `refs/sweep-orderings.md`

Cargo.lock

@@ -17,3 +17,7 @@ pyo3 = { version = "0.24", features = ["extension-module"] }
 numpy = "0.24"
 indicatif = "0.17"
 ctrlc = "3"
+
+[profile.bench]
+inherits = "release"
+debug = true

Cargo.toml

@@ -9,6 +9,9 @@ repository = "https://github.com/PeaBrane/peapods"
 keywords = ["ising", "monte-carlo", "spin-glass", "parallel-tempering", "cluster-algorithm"]
 categories = ["science", "simulation"]
 
+[features]
+profile = []
+
 [dependencies]
 rand = "0.8"
 rand_xoshiro = "0.6"

spin-sim/Cargo.toml

@@ -8,6 +8,7 @@ use rand_xoshiro::Xoshiro256StarStar;
 ///
 /// `energies`: per-replica average energy (energy per spin)
 /// `n_spins`: total number of spins (for converting to total energy)
+#[cfg_attr(feature = "profile", inline(never))]
 pub fn parallel_tempering(
     energies: &[f32],
     temperatures: &[f32],