> -----Original Message-----
> From: Dan Michaels [SMTP:oricom@LYNX.SNI.NET]
> Sent: Monday, August 21, 2000 5:18 PM
> To:   PICLIST@MITVMA.MIT.EDU
> Subject:      Re: [PIC]: Sine table
>
> AndyK wrote:
> >>Ahh, so. "Deterministic timing" is whatever the compiler **gives** you,
> >>whether it gives 10 cycles to code a loop, or 25 - whether a C
> instruction
> >>gets coded into 1 cycle or 8. Slowly it's coming across.
> >
> >
> >Maybe the definition of "determinism" is the problem.  It simply means
> that I
> >know ahead of time how long something takes, whether 5 or 50 cycles (or
> 6/51 in
> >certain cases).
> >
> >It means nothing about the identical timing of some routine(s).
> >
> >If I need to make something ALWAYS take the same amount of time, I adjust
> the
> >generated code so it does.  That is a special case of determinism.
> >
> >Deterministic behavior means it will always come to a (correct) answer.
> It can
> >also have time limitations added to it if you like.
> >
>
>
> Andy, I think you painted yourself into a corner. With C, if you
> change just one instruction, the timing may be 10s of cycles different.
> This is not deterministic. Fiddling with the C, counting the #cycles
> in the assembler output, and iterating endlessly just doesn't seem
> especially elegant or efficient.
>
I hope you don't see this as nit-picking, but the whole point is that you
can't change one *instruction* in C.  You can change an operation, a
variable or a literal and depending what you have done, then yes, the
compiled code may have increased (or decreased) by several cycles.  But if
you needed to do the same thing in ASM, chances are that you would need to
alter or add more than one instruction also.

> Best to know both C and assembler, and use the latter when you
> need an exact number of cycles.
>
Agreed.

> I think it's time to retract your initial statement that C can produce
> deterministic code - don't you think? [come'on , give up already - I
> need to do some work today - even if it is monday].
>
It's almost definately *possible*.  It's almost certainly not *practical*

Mike

--
http://www.piclist.com hint: To leave the PICList
mailto:piclist-unsubscribe-request@mitvma.mit.edu