Re: What is faster Ada or C?

["Jeff Creem" <jeff@thecreems.com>]

Note that if you are really interested in the performance difference of
the two discussed square roots you may want to look for a few minutes
at the instruction set of your computer and your problem domain since
there is sometimes an instruction that calculates the 1/sqrt(x) which can
then be used in places where you had code like  Y := B / SQRT(X) which
can now be  Y := B * my_sqrt_recip(X).   If your processor has and
instruction like
this it can save a few cycles.   I use this as an example but there may be
other cases where
things like this can be found.

A small package machine_code coded special purpose math library can
sometimes really
help in tight loops. (And of course it can be a fun waste of time in not so
tight loops :)





"Jacob Sparre Andersen" <sparre@nbi.dk> wrote in message
news:3C14E817.8A2E798E@nbi.dk...
> Robert Dewar wrote:
>
> > Jacob Sparre Andersen <sparre@nbi.dk> wrote in message
news:<3C10F77E.6BDD9A3B@nbi.dk>...
> > > When I look at the corresponding assembler code (gnatgcc
> > > -S/g++ -S) I can see that GNU C++ inlines the square root
> > > function, while GNU Ada doesn't.
> >
> > You seem to be assuming that the function called square root in C++
> > and the function called square root in Ada
> > are the same function,
>
> Correct.
>
> > presumably because their names
> > are similar, and both are usable in your context, but they
> > are not at all the same routine. The Ada routine has additional
> > semantics not present in the C++ routine (e.g.
> > the handling of -0.0 which is required to return -0.0).
>
> I had forgotten (or had not noticed) this.
>
> > Those extra tests take significant time. Now one can sit
> > back and say "Ada is a junk language, it insists on testing
> > for errors at runtime, and in giving accurate IEEE results,
> > and I don't need either",
>
> I happen to need accurate results, and I appreciate being
> able to specify things like the number of significant digits
> in my floating point types. But I am (as a physicist) not
> used to -0.0 being different from +0.0, nor to being unable
> to calculate the square root of -0.0, as your explanations
> seem to indicate that IEEE floating point processors are.
>
> > but that's not a very reasonable
> > position given that if you like you can easily call the
> > C routine directly, since Ada, unlike other languages,
> > carefully defines the syntax and semantics of calling routines written
> > in other languages :-)
>
> Since I - in the example program - know that the values will
> never get close to zero, I suppose that I can use the C
> function "sqrt".
>
> Jacob
> --
> I teorien lyver folk ikke i retten.