home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!yale!mintaka.lcs.mit.edu!ai-lab!mobot!bleck
- From: bleck@mobot.ai.mit.edu (Olaf Bleck)
- Newsgroups: comp.sys.m68k
- Subject: Re: MC68332 TCR1/TCR2 Problem
- Date: 21 Jan 1993 05:45:53 GMT
- Organization: MIT Artificial Intelligence Laboratory
- Lines: 67
- Sender: bleck@mobot (Olaf Bleck)
- Distribution: world
- Message-ID: <1jldahINNigk@life.ai.mit.edu>
- References: <1j7ua0INNsci@life.ai.mit.edu> <1993Jan19.220045.28984@newsgate.sps.mot.com>
- Reply-To: bleck@ai.mit.edu
- NNTP-Posting-Host: mobot.ai.mit.edu
-
- In article <1993Jan19.220045.28984@newsgate.sps.mot.com>, Jeff Loeliger <jeffl@darth.sps.mot.com> writes:
- |> This is not a stupid feature. It provides the best possible real time
- |> response.
- |> With the greater than or equal comparator, if an event gets scheduled in
- |> the "past"
- |> it will happen immediately. This is useful if the TPU misses an event do
- |> to heavy
- |> loading. If the event was scheduled for $9000, but the TCR was $9002 by
- |> the
- |> time the TPU serviced the request the match will happen immediatly. If
- |> the TPU
- |> had only a equal comparator the match would happen $FFFD counts late!
-
- Well, there is some merit to this claim, but if you're scheduling things so
- close to the time when they're supposed to happen, then I'd want to rethink
- the code. Additionally, if you count on this to execute late scheduled
- events, and you're depending on knowing when the event actually happened so
- that you can schedule the next even relative to that time, then it's
- important to realize that the TPU channel in fact saved the *actual time*,
- i.e. the $9002, instead of the $9000. Thus this little error in time will
- propagate through and grow until and can resync everything at some other known
- time.
-
-
-
- |> Events can be scheduled $8000 from the current time without a problem. If
- |> you
- |> want to scheule a event $8001 from the current TCR, you should shedule an
- |> event $1 and then $8000. The event scheduled $1 from the current time
- |> will
- |> be serviced late, but the $8000 will be service at the correct time.
- |> Using this
- |> method you should always schedule the >$8000 part of a event first, and
- |> then
- |> schedule events $8000 away or repeatedly schedule events $8000 away. This
- |> will provide the mininal amount of loading on the TPU.
-
-
- That's an interesting twist--wish I'd thought of it! I will consider it
- (especially given that mod $8000 is pretty easy...)
-
-
-
- |> >"draft" copy, but who knows what that means. Basically it was a manual
- |> for
- |> >the TPU assembler, which is *not* on the BBS. Then, you'll have to get a
- |> >copy of all the stuff on the BBS, including the TPU assembler, and take
- |> it
- |> >from there. One of these days I will be tackling this task too...
- |> > -stuff deleted-
- |>
- |> The TPU assembler is on the BBS, I checked today (Jan. 19). The assembler
- |> on the BBS
- |> was meant to be used to assemble new "mixes" of functions on the BBS.
-
-
-
- Ah, what I meant was that the TPU Assembler *Manual* is NOT on the BBS. It's
- the only place where I've actually found all the syntax of the TPU assembler
- described.
-
- --
- Tear Here:
- -----------------------------------------------------------------
- _________________________________________________________________
- Olaf Bleck bleck@ai.mit.edu
-
-