Mail Archives: djgpp/1998/09/02/06:23:23
On Wed, 2 Sep 1998, Neil Townsend wrote:
> I want to ensure that only physical memory is used and that the
> memory management is as fast as possible. Therefore, from what I've managed
> to garner form the docs I've seen with djgpp and this group, it struck me
> that CWSDPR0 (or perhaps PMODETSR?) would be ideal because there would be no
> overhead with potential (but unused) disk swap decisions.
I would imagine that the overhead of the (unused) disk swap decisions is
minimal. I don't think that it should have any real effect on the speed
of your program. In any case, I'd suggest to test this before you
decide. Since CWSDPR0 is tested (by DJGPP users) much less than CWSDPMI,
chances are that it has more bugs or surprising features.
It might be more important to understand how does your data-gathering
program gets the data. For example, if it uses hardware interrupts, then
CWSDPR0 might be better because it has less overhead in handling
interrupts. But even then I'd suggest to try CWSDPMI first, and if it
suits you, stick with it, that was the recommendation of Charles
Sandmann, the author of both CWSDPMI and CWSDPR0, last time I checked.
> Does that mean that under CWSDPR0 the allegro timer routines, for example,
> can cause problems? Or have I misunderstood you here - would it be more to
> do with interrupt routines that I've registered using the dpmi interrupt
> registration routines?
No, I meant hardware interrupts, like the timer tick, and signals, like
SIGINT. I don't know whether Allegro timers will have trouble, but they
certainly do work with CWSDPMI.
So, bottom line is: CWSDPR0 is less tested, and I suggest to switch to it
only if you actually get insufficient performance with CWSDPMI.
- Raw text -