gh-90213: Speed up right shifts of negative integers

[mdickinson]

Taking some ideas from the work of @xuxinhang in #30044, this PR makes right shift of negative integers more efficient, by avoiding the need for temporary intermediate results.

Closes#90213

https://bugs.python.org/issue46055

[mdickinson]

One example timing: evaluating -7**100 >> 123 is approximately twice as fast on this branch as on main:

lovelace:cpython mdickinson$ git rev-parse main
3581c7abbe15bad6ae08fc38887e5948f8f39e08
lovelace:cpython mdickinson$ git rev-parse faster-rshift-of-negative
9bf9dcdd3342e9e770de23439065d46e07401a5c
lovelace:cpython mdickinson$ git checkout -q faster-rshift-of-negative && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -7**100; b=123" "a>>b"
5000000 loops, best of 5: 44.8 nsec per loop
lovelace:cpython mdickinson$ git checkout -q main && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -7**100; b=123" "a>>b"
5000000 loops, best of 5: 95 nsec per loop

[mdickinson]

Smaller and larger examples also benefit on my machine: -1234567890 >> 8 is approximately 1.9 times faster; -7**100000 >> 12345 is approximately 2.8 times faster:

lovelace:cpython mdickinson$ git checkout -q main && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -1234567890; b=8" "a>>b"
5000000 loops, best of 5: 81.5 nsec per loop
lovelace:cpython mdickinson$ git checkout -q faster-rshift-of-negative && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -1234567890; b=8" "a>>b"
5000000 loops, best of 5: 42.6 nsec per loop
lovelace:cpython mdickinson$ git checkout -q main && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -7**100000; b=12345" "a>>b"
20000 loops, best of 5: 19.3 usec per loop
lovelace:cpython mdickinson$ git checkout -q faster-rshift-of-negative && make >/dev/null 2>&1 && ./python.exe -m timeit -s "a = -7**100000; b=12345" "a>>b"
50000 loops, best of 5: 6.82 usec per loop

[mdickinson]

I've reworked to remove the duplication between the positive and negative branches, and to move the shift==0 check earlier.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @mdickinson for commit b3a1a8f 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @mdickinson for commit fd4ce81 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[markshannon]

I'm not sure that I understand 100%, but I think it is correct and it does pass the tests.

A few of the variables could have their scope narrowed.

[mdickinson]

@markshannon Thanks for the review! I've addressed your comments, and I think this should be ready for re-review.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @mdickinson for commit c5e3f81 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[markshannon]

AFAICT this looks good, and all the buildbots are happy.