gh-106529: Split FOR_ITER_RANGE into uops by gvanrossum · Pull Request #106638 · python/cpython
Ignore the commit history, this is a pile of different experiments.
Benchmark says 1% slower. Now, that's for the version where the "is it exhausted?" check pushes a NULL onto the stack if it is, and the next value if it isn't, which isn't the latest design. So I'm not sure whether I care. If Mark likes the latest version I'll re-run the benchmarks.
(Benchmarking says nothing about the Tier 2 interpreter speed, since that's off except during a few tests.)