René's URL Explorer Experiment

Title: gh-146558: optimize dict access with objects with known hash by kumaraditya303 · Pull Request #146559 · python/cpython · GitHub

Open Graph Title: gh-146558: optimize dict access with objects with known hash by kumaraditya303 · Pull Request #146559 · python/cpython

X Title: gh-146558: optimize dict access with objects with known hash by kumaraditya303 · Pull Request #146559 · python/cpython

Description: I asked Claude to create a benchmark for this, it created the following script which shows a real significant speedup. Benchmark No JIT JIT Speedup dict_get[str] 55.7M/s 70.1M/s +26% dict_get[int] 55.5M/s 75.4M/s +36% dict_get[float] 50.9M/s 74.7M/s +47% dict_get[complex] 44.9M/s 74.9M/s +67% dict_store[str] 74.9M/s 85.7M/s +14% dict_store[int] 76.0M/s 94.1M/s +24% dict_store[float] 66.6M/s 97.0M/s +46% dict_store[complex] 55.9M/s 98.0M/s +75% dict_get[multi] 23.4M/s 35.1M/s +50% dict_get[obj] 52.9M/s 68.8M/s +30% dict_store[obj] 75.4M/s 92.5M/s +23% Details """Benchmark dict get/store with constant keys to measure _BINARY_OP_SUBSCR_DICT_KNOWN_HASH and _STORE_SUBSCR_DICT_KNOWN_HASH speedup. Usage: PYTHON_JIT=0 ./python.exe bench_dict_known_hash.py PYTHON_JIT=1 ./python.exe bench_dict_known_hash.py """ import time import os import sys N = 50_000_000 def bench_get_str(n): d = {'a': 1, 'b': 2, 'c': 3} x = 0 for _ in range(n): x += d['a'] return x def bench_get_int(n): d = {1: 10, 2: 20, 3: 30} x = 0 for _ in range(n): x += d[1] return x def bench_store_str(n): d = {} for _ in range(n): d['a'] = 1 return d def bench_store_int(n): d = {} for _ in range(n): d[1] = 1 return d def bench_get_float(n): d = {1.5: 1, 2.5: 2, 3.5: 3} x = 0 for _ in range(n): x += d[1.5] return x def bench_store_float(n): d = {} for _ in range(n): d[1.5] = 1 return d def bench_get_complex(n): d = {1+2j: 1, 3+4j: 2} x = 0 for _ in range(n): x += d[1+2j] return x def bench_store_complex(n): d = {} for _ in range(n): d[1+2j] = 1 return d def bench_get_multi(n): d = {'a': 1, 1: 2, b'x': 3, (1, 2): 4} x = 0 for _ in range(n): x += d['a'] + d[1] + d[b'x'] + d[(1, 2)] return x class _Key: pass _KEY = _Key() def bench_get_obj(n): d = {_KEY: 1} x = 0 for _ in range(n): x += d[_KEY] return x def bench_store_obj(n): d = {} for _ in range(n): d[_KEY] = 1 return d def run_avg(name, func, n, runs=3): func(1000) # warmup times = [] for _ in range(runs): t0 = time.perf_counter() func(n) times.append(time.perf_counter() - t0) avg = sum(times) / len(times) rate = n / avg / 1e6 print(f" {name:25s} {avg:.3f}s ({rate:.1f}M iter/s)") return avg if __name__ == "__main__": jit = "JIT" if os.environ.get("PYTHON_JIT", "1") == "1" else "no JIT" print(f"Python {sys.version.split()[0]} ({jit}), avg of 3 runs\n") run_avg("dict_get[str]", bench_get_str, N) run_avg("dict_get[int]", bench_get_int, N) run_avg("dict_get[float]", bench_get_float, N) run_avg("dict_get[complex]", bench_get_complex, N) run_avg("dict_store[str]", bench_store_str, N) run_avg("dict_store[int]", bench_store_int, N) run_avg("dict_store[float]", bench_store_float, N) run_avg("dict_store[complex]",bench_store_complex,N) run_avg("dict_get[multi]", bench_get_multi, N // 4) run_avg("dict_get[obj]", bench_get_obj, N) run_avg("dict_store[obj]", bench_store_obj, N) Issue: gh-146558

Open Graph Description: I asked Claude to create a benchmark for this, it created the following script which shows a real significant speedup. Benchmark No JIT JIT Speedup dict_get[str] 55.7M/s 70.1M/s +26% dict_...

X Description: I asked Claude to create a benchmark for this, it created the following script which shows a real significant speedup. Benchmark No JIT JIT Speedup dict_get[str] 55.7M/s 70.1M/s +26% dict_...

Opengraph URL: https://github.com/python/cpython/pull/146559

X: @github

direct link

Domain: github.com

Links:

Viewport: width=device-width