[Bug math/14412] Removal of sysdeps/x86_64/fpu/s_sincos.S causes regressions

["wbrana at gmail dot com" <sourceware-bugzilla at sourceware dot org>]

http://sourceware.org/bugzilla/show_bug.cgi?id=14412

--- Comment #24 from wbrana at gmail dot com 2013-04-26 15:20:54 UTC ---
(In reply to comment #10)
> If we fix sinf to be fast again it will be imprecise which is a critical bug
> for another class of users. There is no implementation that can satisfy all
> requirements from all users. At the very least the precise implementation is at
> least correct if slow. Therefore this is an enhancement request to make the
> implementation fast.
> 
> We are working on the possibility of having 3 libm variants, fast, default, and
> precise. At that point -ffast-math will select the fast variant and provide a
> fast but imprecise implementation.
> 

http://sourceware.org/bugzilla/show_bug.cgi?id=13658

impreciseness of fast sincos can't be critical bug because it took 15 years
(1997-2012) 
to report this bug which means s_sincos.S is precise enough for 99% of apps.
s_sincos.S should be restored because slow sincos can break many apps 
because most open source apps aren't properly tested and app devs aren't aware 
that sincos is 50% slower since 2.16.
Only users will notice than e.g. new Ubuntu is slow but users usually can't
submit proper bug report like this one to app devs.
App devs won't know that they should add -ffast-math to restore performance.
Very few apps are using -ffast-math in present.
-ffast-math can add new bugs to apps which aren't broken by fast sincos.

New function named e.g. sincos_precise should be added to glibc for 1% apps
which are affected by impreciseness of fast sincos.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.