RE: [ng-spice-devel] save all
> > ..save ic(q*) vdsat(m*) i(d1-d11)
>
> ACS does that.
I haven't got any further trying to compile ACS in Windows.
You use namespaces, which the latest Borland compiler
understands, but it obviously handles it differently from
Microsoft.
I've never used namespaces, so I'm not quite sure what the
problem is. And my Watcon compiler (10.6) seems to be too
old to understand them as well, so I'm stuck with the
Borland one.
> Another issue is that perhaps not all data is available at every time
> step. Every output would be a [time, probename, value] triplet.
> They come in arbitrary order, not necessarily sorted by anything.
That would be nice, but that could take up a lot more disk space.
Are you saying that the number of nodes/currents/parameters which
don't change between timepoints would outway the extra data needed
for every point ?
I suppose the probename could be (for example) a 16 bit index into
a table of fully qualified parameter names, (i.e. the list in the
rawfile header), and something similar for the time. Actually, you
could have simply [probename, value] pairs, and arrange them in
bunches with the first one having probename=reference variable,
and all the rest up to the next probename=reference variable one
being associated with that timepoint. With double precision data,
the worst case would only be 50% bigger than a standard file.
The problem with this would be the randomness of the intervals
between the appearance of any single variable. As I hinted before,
I find that the largest price for large data files is paid when
you try to read a curve back out of the file. If the postprocessor
has to scan through the whole file to read back every single curve,
then things will be a lot slower than they need to be.
Therefore, the data would need to be clumped together in some way
to reduce the amount of scanning necessary to read it back. I think
that could be done, though. It's a very interesting idea.
Cheers,
Alan
Partial thread listing: