gh-149180: avoid double checking tp_as_number, tp_as_sequence and tp_as_mapping

[NekoAsakura]

Benchmark results

	main	branch	%
map_check_int	8.51 ns	7.95 ns	-6.5%
map_check_obj	8.50 ns	8.00 ns	-6.0%
seq_check_obj	8.32 ns	7.92 ns	-4.8%
seq_check_int	8.32 ns	7.94 ns	-4.5%
seq_check_tuple	8.21 ns	7.94 ns	-3.3%
seq_check_list	8.22 ns	8.01 ns	-2.5%
map_check_dict	8.12 ns	7.93 ns	-2.4%
map_check_tuple	8.09 ns	7.93 ns	-1.9%
getitem_dict	12.92 ns	12.68 ns	-1.9%
mul_str_int	14.74 ns	14.53 ns	-1.4%
mul_int_str	14.70 ns	14.51 ns	-1.3%
map_check_list	8.06 ns	7.96 ns	-1.2%
mul_int_list	16.11 ns	15.98 ns	-0.8%
mul_list_int	15.87 ns	15.76 ns	-0.7%
getitem_list	3.55 ns	3.53 ns	-0.4%
getitem_tuple	3.45 ns	3.44 ns	-0.3%
getitem_str	3.51 ns	3.50 ns	-0.1%
bool_none	6.78 ns	6.77 ns	-0.1%
bool_dict	7.49 ns	7.50 ns	+0.2%
bool_list	7.51 ns	7.54 ns	+0.4%
seq_check_dict	8.32 ns	8.39 ns	+0.9%
bool_obj	7.03 ns	7.19 ns	+2.3%
bool_int	7.66 ns	7.96 ns	+3.8%
Geomean			−1.0%

Benchmark scripts

Build flags: ./configure --enable-experimental-jit --enable-optimizations --with-lto

import pyperf

r = pyperf.Runner()
I = "from _testlimitedcapi import "

for n, s in [("list", "v=[1]"), ("dict", "v={1:2}"), ("int", "v=5"),
             ("none", "v=None"), ("obj", "v=object()")]:
    r.timeit(f"bool_{n}", "bool(v)", s)

for n, s in [("list", "v=[1]"), ("tuple", "v=(1,)"), ("int", "v=5"),
             ("dict", "v={1:2}"), ("obj", "v=object()")]:
    r.timeit(f"seq_check_{n}", "f(v)", I + "sequence_check as f;" + s)
    r.timeit(f"map_check_{n}", "f(v)", I + "mapping_check as f;" + s)

for n, s in [("list", "v=[1,2,3]"), ("tuple", "v=(1,2,3)"),
             ("dict", "v={0:'a',1:'b'}"), ("str", "v='abc'")]:
    r.timeit(f"getitem_{n}", "v[1]", s)

for n, s in [("int_list", "a=2;b=[1]"), ("list_int", "a=[1];b=2"),
             ("int_str", "a=2;b='ab'"), ("str_int", "a='ab';b=2")]:
    r.timeit(f"mul_{n}", "a*b", s)

P.S.
Different build flags might affect the benchmarks, but I haven't had the time to test them yet.
I did see a regression on PyMapping_Check matching @picnixz's concern(#149317 (comment)) w/ --enable-experimental-jit (no PGO/LTO) build only, but I suspect my testing at the time wasn't sufficiently rigorous. Will take a closer look later.

• Issue: Avoid double checking tp_as_number, tp_as_sequence and tp_as_mapping #149180

[read-the-docs-community]

Documentation build overview

📚 cpython-previews | 🛠️ Build #32576493 | 📁 Comparing 8cefc49 against main (b2582a6)

  🔍 Preview build  

40 files changed · ± 39 modified · - 1 deleted

± Modified

• c-api/concrete.html
• c-api/exceptions.html
• c-api/extension-modules.html
• c-api/index.html
• c-api/interp-lifecycle.html
• c-api/module.html
• c-api/stable.html
• c-api/threads.html
• c-api/type.html
• c-api/typeobj.html
• and 29 more...

- Deleted

• c-api/slots.html