GH-87849: Simplify stack effect of SEND and specialize it for generators. by markshannon · Pull Request #101788 · python/cpython
This avoids most the overhead when transferring control from caller to the generator or coroutine in a yield from or await.
The benchmarks results are in the noise, except for the "generators" benchmark which is 2.5x faster.
Mean speedup is about 1%, although that might not be meaningful if one benchmark is 150% faster.