[rc5] About 486 and 6x86 code

Remi Guyomarch rguyom at mail.dotcom.fr
Sat Aug 9 19:49:08 EDT 1997


I've seen some posts about the 6x86 code being faster than the 486 code
on a 486. Let's me explain what's going on :

- I think the code autodetection is based on a short benchmark of all
different routines. Perhaps this benchmark is a bit too short. (I
haven't seen the 2.004 code yet).

- There isn't much difference between the 6x86 code and the 486 code. In
fact it's just instruction reordering and the choice of two adds vs an
lea instruction. In theory there should be no difference between the two
routines on a 486. The 486 code is slightly smaller than the 6x86 one,
and should do better on a loaded machine.
On my DX4/100 under Linux I get 92-93 kkeys/s whenever I choose the 486
optimized code or the 6x86 code.

I suggest you to "-benchmark" the two codes, and to choose the faster.

-- 
Rémi
----
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.



More information about the rc5 mailing list