gh-136681: make argsbuf static to speedup processing positional arguments with AC

[skirpichev]

• Issue: Slow functions calls for C-coded functions, accepting positional-or-kwarg argument vs positional-only #136681

[skirpichev]

benchmark: #136681 (comment)

[ZeroIntensity]

Won't this break when called concurrently?

[skirpichev]

Hmm, @ZeroIntensity, test seems fixed with using combined static and _Thread_local. May this work or this is too naive approach?

[ZeroIntensity]

Hm, you could try it and see what the test suite says. To my knowledge, thread local lookups are generally slower, so we might not see a speedup.

[skirpichev]

Well, CI tests pass, but that might be just an accident.

thread local lookups are generally slower, so we might not see a speedup.

Here my quick measurements for default configure arguments on Linux box. (Free-threading build might change the picture.) Micro-benchmarks are for math.fmin(): when only positional arguments are allowed (as in the main) vs positional-or-keyword allowed.

In the main:

Benchmark	posonly-ref	posorkw-ref
fmin(1.0, 2.0)	169 ns	188 ns: 1.11x slower
fmin(1.0, 2.0) x 2 times	969 ns	1.00 us: 1.03x slower
fmin(1.0, 2.0) x 10 times	2.09 us	2.27 us: 1.08x slower
fmin(1.0, 2.0) x 100 times	14.8 us	16.5 us: 1.12x slower
Geometric mean	(ref)	1.09x slower

With the patch:

Benchmark	posonly-patch	posorkw-patch
fmin(1.0, 2.0)	170 ns	173 ns: 1.02x slower
fmin(1.0, 2.0) x 10 times	2.06 us	2.08 us: 1.01x slower
fmin(1.0, 2.0) x 100 times	14.8 us	15.0 us: 1.01x slower
Geometric mean	(ref)	1.01x slower

Benchmark hidden because not significant (1): fmin(1.0, 2.0) x 2 times

Details

importpyperffrommathimportfmindeff(n): for_inrange(n): fmin(1.0, 2.0) runner=pyperf.Runner() runner.bench_func('fmin(1.0, 2.0)', fmin, 1.0, 2.0) fornin [2, 10, 100]: s=f'fmin(1.0, 2.0) x {n:3} times'runner.bench_func(s, f, n)