[Bug middle-end/118012] [avr][13/14/15/16/17 Regression] Expensive code (bit extract + extend + neg + and) instead of bit test

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118012

--- Comment #28 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
...though the mult part is from Drea in r14-1655 and r14-1654.

The previous negation was already bad, but these turned bad to mad.

So what speaks against costing it, or reverting?