Re: IEEE rounding modes: Do people actually change on the fly?

In article <pan.2007.08.24.01.25.50.471638@xxxxxxxxxxxxxxxxxxxxx>,
andrew-newspost@xxxxxxxxxxxxxxxxxxxxx (Andrew Reilly) wrote:

I'm not sure what you mean by "megabytes of variation".

That's not a good term, for something that I vaguely understand but
don't know how to put into words.

Say eight channels/box at 100kHz at four bytes/sample

A channel gives you a single four-byte number? These numbers have to be
watched for going out of range; presumably the valid ranges for some
depend on others? Do you deal with each sample independently, watching
current values with no examination of history, or do you have to deal
with trends and rates of change? If so, over what period?

There's little or no conditional execution in the code, and no
storage, so coverage is close to a no-brainer, and the
haracteristics of the data itself are quite tightly constrained.

None of those helpful characteristics apply to the kind of modelling I'm
working on.

It's not some common setting that I'm missing, is it? I should go and
read the manuals again...

Not as far as I know.

--
John Dallman, jgd@xxxxxxxxx, HTML mail is treated as probable spam.
.