GCC 2.95 x Kernel 2.2.11-ac2 Opt. 'Problems' with RAID ???

[Leonardo Marques de Souza]

Hi,


	I have some doubts about optimizations with gcc-2.95.
After install the new gcc over RedHat6.0(very clean install) and
upgraded to kernel 2.2.11-ac2, i tryed to compile the new kernel with
more 'optimizations' (modifying the arch/i386/Makefile):
first test (without any touch in CFLAGS arqs..)

my Machine is K6-2 333MHz...

DEFAULT in Makefile ( -m486 -O2)
i got this (the new raid patch apeer in Alan coxs patchs - ac)

Aug 14 05:46:07 carol kernel:    pII_mmx   :   734.949 MB/sec 
Aug 14 05:46:07 carol kernel:    p5_mmx    :   711.708 MB/sec 
Aug 14 05:46:07 carol kernel:    8regs     :   496.824 MB/sec 
Aug 14 05:46:07 carol kernel:    32regs    :   309.753 MB/sec 
Aug 14 05:46:07 carol kernel: using fastest function: pII_mmx (734.949
MB/sec)

After modify to -O9 -march=k6 i got 

Aug 14 06:07:10 carol kernel: hda: QUANTUM FIREBALL EL5.1A, 4892MB
w/418kB Cache, CHS=623/255/63 
Aug 14 06:07:10 carol kernel:    pII_mmx   :   717.423 MB/sec 
Aug 14 06:07:10 carol kernel:    p5_mmx    :   684.657 MB/sec 
Aug 14 06:07:10 carol kernel:    8regs     :   470.916 MB/sec 
Aug 14 06:07:10 carol kernel:    32regs    :   332.613 MB/sec 

After change it to -O9 -mpentium i got:

Aug 15 00:23:08 carol kernel: hda: QUANTUM FIREBALL EL5.1A, 4892MB
w/418kB Cache, CHS=623/255/63 
Aug 15 00:23:08 carol kernel:    pII_mmx   :   711.708 MB/sec 
Aug 15 00:23:08 carol kernel:    p5_mmx    :   685.038 MB/sec 
Aug 15 00:23:08 carol kernel:    8regs     :   470.916 MB/sec 
Aug 15 00:23:08 carol kernel:    32regs    :   322.326 MB/sec 

Why its happen?? the GCC compiler make the Raid analiser some 'confused'
and optimizations really happen
or its a really strange optimizations with RAID causing to 'downgrade'
with -m486 and -march=k6 or -mpentium ....

I would like to understand that...(if exists any solutions.. are very
wellcome!!! but i would like to understand only...)

HOoo yes!!  my kernel are running very well, without _anY_ problem or
fault... thats grate!!! Why the Kernel guys do not make new
Optimizations Flags in kernel source??? if is to discover new bugs in
kernel/gcc, the must thing are change this flags to make kernel more
fast!! (but , of course, if happen any problem... do a any options to do
'default' flags to worste case... like '-O -g -m386 -fno-softmath.. and
so on...)


Thanks in advanced...

T+
Leo