Date: Fri, 19 Aug 94 09:56:14 -0400
From: dj AT ctron DOT com (DJ Delorie)
To: olly AT mantis DOT co DOT uk
Cc: mcastle AT umr DOT edu, haltmaye AT ibm22m DOT RZ DOT Uni-Augsburg DOT DE,
        djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: Speed tuning programs 

> >From the recursive make discussion, I'd gathered that Windows DPMI is
> better than QDPMI in the amount of conventional memory it needs.  I'd
> vaguely assumed this meant that it was accidentally well-written.

Windows requests a much smaller chunk of dos memory for per-program
storage than QDPMI does, about 6K vs 100K.  That doesn't mean they're
better, just that they at least store their data in extended memory.

> >Actually, I think the idea is to not be using any dpmi at all.
> 
> So is VCPI faster than DPMI?  Or am I missing some point here?

Some functions of go32 just don't work under DPMI, and some work
different.