AMD K6 support

[Kurt Garloff]

Hi everybody,

I found my AMD K6 perform best with -O3 -mpentiumpro -malign-functions=4 
 -fschedule-insns2 -ffast-math on the numerical calculations I make.

As I don't like specify -mpentiumpro -malign-functions=4 all the time, I
decided to create a -mamdk6 to basically select the correct values for
scheduling etc.
I looked into the K6 optimization docs, played a little bit with the
params in config/i386/i386.{md,c,h} and created a patch. I get some better
performance than with -mppro -ma-f=4.

The patch is appended. Just enjoy, if you got a AMD K6. 
Maybe some more experiments with non-numerical code could help to get even
better results, but I wasn't able to find params to produce faster code, so
far.

Jeff: I'd really be lucky to see this in egcs.
I hereby declare, that this little patch was entirely written by me and I
hereby release it to the public under the terms of the GNU GPL.

Regards,
-- 
+--------------------------------------------------------------+
|  cand.phys. Kurt Garloff        C1-O4-101 (Chemiegeb"aude)   |
|Lehrstuhl Hochfrequenztechnik  (Fakult"at f"ur Elektrotechnik)|
|   Universit"at Dortmund             D - 44221 Dortmund       |
|   Tel.: (0231) 755-3947            Fax: (0231) 755-4631      |
|            garloff@hft.e-technik.uni-dortmund.de             |
|                 Public PGP key available !                   | 
+--------------------------------------------------------------+

P.S.:
Just for your information:
I also handcoded some routines when experimenting. In contrast to what
AMD docs say, you can use at least one of the new AMD-3DNow! instructions
without getting a penalty for the switch between FPU and MMX/3D: prefetch
It greatly improved (factor 1.4) my double(!) Matrix Vector multiplication by 
prefetching the next cache line before using it.