gh-145866: Eliminate redundant refcounting from _CALL_INTRINSIC_2

[KevinH15291]

• Issue: Convert more uops to leave their inputs on the stack to be cleaned up by _POP_TOP #145866

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[KevinH15291]

This is my first CPython PR and I'm not very experienced. This PR largely just copies the style of the other PRs in #145866 and #134584, mainly #145964.

I'm not sure if

self.assertLessEqual(count_ops(ex, "_POP_TOP"), 4)

in my test is completely correct, as well as my testfunc

def testfunc(n):
    x = 0
    for _ in range(n):
        def test_testfunc[T](n):
            pass
    return x

I'm also not 100% sure if regenerated the opcode metadata stuff correctly. Otherwise I believe the PR should be good.

[Fidget-Spinner]

Nice, this looks really good, thanks for doing this!

[Fidget-Spinner]

GreaterEqual might be better. Also did you count how many POP_TOP_NOPs this produces? You can print the trace logs by compiling with the tier 2 interpreter (--enable-experimental-jit=interpreter) and passing the env var PYTHON_LLTRACE=2.

[KevinH15291]

Well honestly, I got lucky; I doubled the numbers used in the test for _CALL_INTRINSIC_1. Later I did verify this, by uh, modifying the end of the test function to print uops, and looking at the printed output, as I was not aware of PYTHON_LLTRACE=2 yet.

[Fidget-Spinner]

Ok cool, thanks for verifying. We want to use GreaterEqual here in case we introduce more optimizations in the future, and we don't want to cause this to break.