[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: mk.conf



On Tue, Aug 26, 2003 at 04:33:26PM +0000, Brad Brad wrote:

> "COPTS+= -march=pentium -mcpu=pentium4" safe for a nice simple/stable 
> optimization?

  maybe you could consider the cost/benefit ratio to be worthwhile on
  a P54C; but a pentium_4_ ??

  call it a worthless offhand guess, but if things are too slow for you
  running on a P4, maybe you could get a better HD setup or better RAM; 
  something in the peripheral area...

  i used to run -march=i586 and -O3 on a p200/mmx without issues; however
  absolutely *nobody* onlist going to be interested in hearing any problems
  you encounter while running that setup, unless they themselves are 
  also doing the same thing, which is not likely to be common.  plus, 
  the first thing i would think about when i did encounter odd goofiness
  was that the above optimizations were the cause of the trouble... but,
  to answer your question, see the first sentence of this paragraph. 

  i would think it might be worthwhile, if you're actually going to do 
  a kernel and make build with the optimizations, to run some decent
  benchmarking on the system both before and after.  if it is going to 
  be a "production" server, try to do as much benchmarking of this
  type without it in a production role first, to eliminate varied
  network load from the equation...

  run netperf/netstrain on localhost; do 'openssl speed', etcetc; 
  see if the optimizations have *really* created a benefit...

  someone mentioned a while back ( last time this came up ) questioning
  what actual optimizations are being done differently between 
  -march=i{3,4,5,6}86 ( i think it was when the intel c compiler
  topic was brought up ), and that perhaps not much difference is
  there to be had.
 
  anyways, test it.  if it crashes and reboots a lot, you have your
  answer.
  
  jared

-- 

[ openbsd 3.4-beta GENERIC ( aug 24 ) // i386 ]