Add FlashMLA flashmla benchmark perf gate

tools/benchmark_perf_gate.py

@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+"""Compare baseline and current performance JSON and fail on regressions."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+METRIC = "throughput"
+TOLERANCE = 0.05
+
+
+def load(path: Path) -> dict[str, float]:
+    data = json.loads(path.read_text(encoding="utf-8"))
+    if isinstance(data, list):
+        return {str(item["name"]): float(item[METRIC]) for item in data}
+    return {
+        str(k): float(v[METRIC] if isinstance(v, dict) else v)
+        for k, v in data.items()
+    }
+
+
+def compare(baseline: dict[str, float], current: dict[str, float]) -> dict[str, object]:
+    rows: list[dict[str, object]] = []
+    failed = False
+    for name, old in sorted(baseline.items()):
+        if name not in current:
+            rows.append({"name": name, "status": "missing-current"})
+            failed = True
+            continue
+        new = current[name]
+        ratio = (new - old) / old if old else 0.0
+        status = "regression" if ratio < -TOLERANCE else "ok"
+        failed = failed or status != "ok"
+        rows.append(
+            {
+                "name": name,
+                "baseline": old,
+                "current": new,
+                "delta_ratio": ratio,
+                "status": status,
+            }
+        )
+    return {"ok": not failed, "metric": METRIC, "rows": rows}
+
+
+def self_test() -> None:
+    data_ok = compare({"case": 100.0}, {"case": 99.0})
+    data_regression = compare({"case": 100.0}, {"case": 90.0})
+    data_missing = compare({"case": 100.0}, {})
+    if not data_ok["ok"] or data_regression["ok"] or data_missing["ok"]:
+        raise RuntimeError(
+            {
+                "ok_case": data_ok,
+                "regression_case": data_regression,
+                "missing_case": data_missing,
+            }
+        )
+    print(json.dumps({"ok": True, "rows": len(data_ok["rows"])}, ensure_ascii=False))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("baseline", nargs="?")
+    parser.add_argument("current", nargs="?")
+    parser.add_argument("--self-test", action="store_true")
+    args = parser.parse_args()
+    if args.self_test:
+        self_test()
+        return 0
+    if not args.baseline or not args.current:
+        parser.error("baseline and current are required unless --self-test is used")
+    result = compare(load(Path(args.baseline)), load(Path(args.current)))
+    print(json.dumps(result, ensure_ascii=False, indent=2))
+    return 0 if result["ok"] else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())