[MPlayer-dev-eng] dsputil_mmx.c: -O2

[Uoti Urpala]

I did manage to reproduce a noticeable advantage (over 5%) with -O2.
-finline_functions is the switch that has a negative effect with -O3 and
above.

I did some profiling with OProfile and believe that the difference is
mainly caused by code size. With inlining most functions seemed to be
slower and some functions which should contain no inlining themselves
were noticeably slower. With -fno-inline-functions for dsputil_mmx only
the stripped mplayer binary has size 6116064; without it 6774528.

I tried using OProfile with the ICACHE_MISSES event (haven't checked
exactly what level of cache misses that measures). That showed about
240k events/frame for the inline version and 132k events/frame for the
noinline version.