Re: Use assert()s for optimization?

[Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>]

On Thu, 6 Nov 2003, Falk Hueffner wrote:

> Richard Guenther <rguenth@tat.physik.uni-tuebingen.de> writes:
>
> > Is there a way to turn an assert() into optimization hints for gcc?
> > Like in
> >
> > double mysqrt(double x)
> > {
> >   assert(x >= 0.0);
> >   return sqrt(x);
> > }
>
> Interestingly, Toon Moene just asked for the same thing on gcc-patches :)
> I don't think it's currently possible. Maybe we could have
> something similar to Microsoft's __assume(), which would then feed the
> nonnegative_p etc. functions and the VRP to come. I'm not sure how to
> represent it best in the intermediary representation, though.

It should probably turn right into REG_EQUAL notes and the like by
cse/gcse. So like we do with the redundant null pointer check for

  if (!foo)
    abort();
  if (foo)
    bar();

should work for __assume(), too - just not creating code for the first
test (or possibly do with a special flag).  So its __builtin_assert() with
the compiler taking advantage out of it in optimized builds.

Richard.