Logged In: YES 
user_id=80475

FWIW, I think the patch ought to also apply to BINARY_ADD. 
There is tons of code that writes a=a+b rather than a+=b.  
It's important that there not be a substantial performance
difference between the two; otherwise, we're introducing
another bit of arcana.

Attached is a revised patch that combines BINARY_ADD with
INPLACE_ADD for a net savings of instructions in ceval.c. 
The timings improve on all my benchmarks (PyBench for
example runs 4% faster).