long double and i386 optimization

[Jan Hubicka]

Hi
I have noticed quite strange think: Main mset calculation loop should
be (IMO) exactly same with long double as with double, since whole
calculation is done in register and no memory is accesed and i386
uses "long double" in FPU registers. So I wonder why mset loop compiled
with -O3 -fomit-frame-pointer -mpentium -funroll-loops 
is slower in long double than in double, since one extra fxch (or fld 
from register I am not sure) is added. Also intructions are scheduled
in a bit different order so result is quite significant(10%) slowdown.
I may send sources too if someone is interested.

Honza
-- 
------------------------------------------------------------------------------
                   Have you browsed my www pages? Look at:
                       http://www.paru.cas.cz/~hubicka
      Koules-the game for Svgalib,X11 and OS/2,  Xonix-the game for X11
      czech documentation for linux index, original 2D computer art and
              funny 100 years old photos and articles are there!