Signed divide by 2 pessimization

[Geoff Keating]

Marek Michalkiewicz <marekm@linux.org.pl> writes:

> I submitted a small patch for expmed.c that adds a new
> macro to give the target some control over this test, but
> it was rejected.  While I agree that ideally, rtx costs
> should be used instead of adding new macros, that hasn't
> happened, and doing it would require a lot of testing on
> many different machines to make sure it doesn't make things
> worse.  (I tried once, and it made x86 code slower, because
> its BRANCH_COST seems to be defined lower than it usually is.
> I can't tell what happens on other machines.)

This is a sign that x86 should be fixed.  It's bad to work around
problems with a backend in the frontend.  Why is the x86 BRANCH_COST
set too low?

-- 
- Geoffrey Keating <geoffk@cygnus.com>