GeezerVison

Hm, it doesn't sound right to override profile-guided optimization, especially since test_decimal (the only current caller of _PyType_GetModuleByDef2) is in the PGO test set.
Does this PR have a significant performance impact?

It is mentioned on the faster-cpython repo that the telco test has slowed down a lot.

According to MSVC, the module state access counts were:

Function / breakdown     entry-cnt  alternative access
-----------------------  ---------  ---------------------------
PyType_GetModuleByDef()    6852643
    convert_op             2971848  via context object
    PyDecType_New          1651188  via context object (partial)
    dec_addstatus          1651188  via context object (partial)
    current_context        1486221  via context object (partial)
    dec_mpd_qquantize       247731  METH_METHOD
    ctx_mpd_qquantize       165000  METH_METHOD
    ...

_PyType_GetModuleByDef2()  1073193
    nm_mpd_qadd             660462
    nm_mpd_qmul             412731

Tested with the /Ob3 option, switching the inlining specifier: f740a5d. My Release/PGO builds on Windows get slower using TLS version of PyThreadState_Get(), which is also observed at #103324 (comment). If *nix OSes are in good health with TLS, I guess I also need to run the telco with a good condition (without TLS):

This patch would need to be applied if we wanted as much speed as the global state access on Windows, which has little effect alone (1%) for some reason.

Windows PGO:
- telco: 9.21 ms +- 0.12 ms -> 9.02 ms +- 0.09 ms: 1.02x faster

Is it acceptable that test_decimal.py has a test case like below, instead of touching the C code?

@requires_cdecimal
class CArithmeticOperatorsTest(ArithmeticOperatorsTest, unittest.TestCase):
    ...
    @unittest.skipIf(not test.support.PGO, 'PGO training only')
    def test_excecise_binop(self):
        Decimal = self.decimal.Decimal
        d = Decimal('11.1')
        for i in range(500000):
            1 + d   # at least 300000 times

I'll try PyType_GetBaseByToken() version.

Closing in favor of proposing the PyType_GetBaseByToken() version, which can supersede _PyType_GetModuleByDef2() on PGO and Relase(/Ob3) builds.