Jonathan Polley wrote:
>
> My C training goes back to circa 1985, at which time all floats were
> passed as doubles.
Yes, they were.
> In fact, the modern C/C++ compilers to which I have
> access still do this (I don't use gcc, except to rebuild FlightGear).
Oh? Which compilers?
> All
> floating-point co-processors do their work in extended precision (usually
> 80-96 bits) and "shorten" the numbers back to the range desired by the
> user,
Certainly some do. I'm pretty sure that not all do, at least not for functions that
can be done significantly faster at lower precision.
> and the single-precision math libraries I have used just cast the
> results of the double-precision routines to be single-precision (it saves
> on duplicated code).
At work we use an Intermetrics C compiler ("Intertools" for NEC V20 series processors)
that is supplied with a standard software floating-point library that includes
separate single and double-precision versions of most of the routines. There is also
an alternate version of the library which uses the same data sizes (4 and 8 bytes) but
does not calculate to the full accuracy, and is considerably faster. You can choose
the most appropriate library for your application.
> I must be older than anyone else here, but my collegiate C training drove
> home the fact that mixing integer and floating-point can give you
> unpredictable results. Granted, this was pre-ANSI C.
Maybe it was unpredictable before standardisation, but I feel that's unlikely; I don't
have the original (K&R) reference book to check. Another possibility is that the
teacher couldn't predict the results because he/she didn't know the language well
enough ... that's quite common, unfortunately.
- Julian
_______________________________________________
Flightgear-devel mailing list
[EMAIL PROTECTED]
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
