Date: Wed, 21 Aug 1996 10:32:51 +0200 (MET DST) From: Christoph Kukulies <kuku@gilberto.physik.rwth-aachen.de> To: joerg_wunsch@uriah.heep.sax.de Cc: freebsd-hackers@freebsd.org, kuku@gilberto.physik.rwth-aachen.de Subject: Re: max math performance - how? Message-ID: <199608210832.KAA08275@gilberto.physik.rwth-aachen.de> In-Reply-To: <199608202032.WAA17791@uriah.heep.sax.de> from J Wunsch at "Aug 20, 96 10:32:27 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
> As Bruce Evans wrote: > > > 386BSD used libm, which is slow. -current uses msun, which is > > slower, except possibly if it is compiled with option HAVE_FPU. > > Why don't we compile it with this option? I thought the emulator will > serve those folks who don't have an FPU? We could also ship the > non-FPU lib in a separate package in releases, but i think a large > number of machines now come with an FPU as an integral part of their > system, so still defaulting to non-FPU math libs seems a little silly. Another idea comes into mind: (don't know 386 architecture so well though): Assumed you have an FPU inlined libm. Would it be possible to run the emulator through a trap (emulator trap instruction) or intercepting the illegal instruction trap, checking the stack for a valid FPU instruction and pass instruction execution to the emulator in that case. > > -- > cheers, J"org > > joerg_wunsch@uriah.heep.sax.de -- http://www.sax.de/~joerg/ -- NIC: JW11-RIPE > Never trust an operating system you don't have sources for. ;-) > --Chris Christoph P. U. Kukulies kuku@gil.physik.rwth-aachen.de
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199608210832.KAA08275>