Messages in this thread | | | Date | Tue, 29 Jul 1997 10:22:51 -0400 (EDT) | From | Robert L Krawitz <> | Subject | Re: Interesting pentium-memcpy results |
| |
Date: Tue, 29 Jul 1997 16:19:31 +0200 (MET DST) From: Ingo Molnar <mingo@pc7537.hil.siemens.at>
the 'soft' routine is not complete, eg. it does not save FPU flags .. but it saves the 'main bulk', the 80 bit FPU registers. Maybe there are other things too to save? Everything that might be affected by those fildl and fistl instructions should be saved.
There's also the issue of making certain that the FPU is in a sane state. If the FPU's in a weird state, it's possible that fildq/fistpq might not do the right thing, although I don't know that for certain.
also, to reduce the size of the saved context, it's not necessary to use all FPU registers. I think 4 of them would be enough. This halves context saving cost. Probably there are enough delay slots to do the looping 'for free'.
The performance advantage from using all FPU registers is significant, although I don't remember offhand what it was on my system.
-- Robert Krawitz <rlk@tiac.net> http://www.tiac.net/users/rlk/
Tall Clubs International -- http://www.tall.org/ or 1-800-521-2512 Member of the League for Programming Freedom -- mail lpf@uunet.uu.net
| |