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

["siddhesh at redhat dot com" <sourceware-bugzilla at sourceware dot org>]

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

Siddhesh Poyarekar <siddhesh at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #18 from Siddhesh Poyarekar <siddhesh at redhat dot com> 2013-04-26 10:00:09 UTC ---
(In reply to comment #16)
> (In reply to comment #13)
> > This should have been fixed with the patch for bug 14496.  Markus, can you
> > confirm that?
> 
> It now takes ~40 seconds to render the test-page on trunk.
> Better than 60 seconds but still 40% slower than s_sincos.S.

That's expected.  The assembly implementation used the fsincos instruction and
the current implementation calls the sin() and cos() functions.  The rounding
mode changes have nothing to do with this part of the performance regression. 
Maybe we could implement a sincos_finite that calls the fsincos instruction and
actually gets used with -ffinite-math.

(In reply to comment #17)
> Perhaps we should add a predefined macro for -frounding-math in gcc, and the
> precise variant of libm could be selected using that macro, if the correct
> rounding precise variants would be suffixed differently (like the *_finite
> entrypoints are for -ffast-math), then all the 3 libm variants could live in
> the same libm.so.6 and people would just choose what they want using
> -ffast-math vs. default vs. -frounding-math.

I like this idea.

-- 
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.