Re: (IBM) SHARE - LNGC requirements

From: Mark Yudkin (myudkinATcompuserveDOTcom_at_nospam.org)
Date: 04/11/04 

Date: Sun, 11 Apr 2004 09:58:08 +0200

Unfortunately, FWAIT is still around. You ask the compiler to generate FWAIT
instructions with the NOIMPRECISE option. Unfortunately, parts of the
runtime library don't issue FWAIT, thus leading to imprecise interrupts and
subsequent 8094 errors. I have a PMR open against ERFC for such an issue.

"glen herrmannsfeldt" <gah@ugcs.caltech.edu> wrote in message
news:abvac.145400$Cb.1522768@attbi_s51...
> Mark Yudkin wrote:
>
> (snip regarding overflow detection)
>
> > You forgot the cost of the FWAIT (or you end up with the dreaded
imprecise
> > interrupt). Since this flushes the floating point queue, the cost of
which
> > is by no means negligable. Needless to say, I do pay that cost, and
> > everything is compiled with noimprecise by default.
>
> I haven't done this for a while now, but I thought the need
> for FWAIT went away a few generations ago. The 8086 had the
> external 8087, and it actually had to wait for it to finish
> an operation. The coupling was slightly closer on the 80286,
> with the 80287, and by the 80486 the FPU was internal.
>
> I thought that with the 486 no FWAIT was required.
>
> Then again, this is a strange newsgroup to ask.
>
> -- glen
>

Relevant Pages

  • Re: Faster general Trunc/Int functions
    ... > The cost of FWAIT is very very little.In the complex number ... > challenges it is less than out measurement error. ... At least on Athlon is assimilated to NOP, ...
    (borland.public.delphi.language.basm)
  • Re: (IBM) SHARE - LNGC requirements
    ... Since this flushes the floating point queue, the cost of which ... I thought that with the 486 no FWAIT was required. ...
    (comp.lang.cobol)