Re: [Patch, Fortran] PR 57160: short-circuit IF only with -ffrontend-optimize

[Janus Weil <janus at gcc dot gnu dot org>]

2018-07-24 17:41 GMT+02:00 Janne Blomqvist <blomqvist.janne@gmail.com>:
> Optimization bugs that pop up at different optimization levels are hard
> enough for users to figure out

Right, and they're impossible to detect if there is no way to disable
the optimization, which is what this PR is about.


> without the frontend generating different
> code to begin with depending on the optimization level.

In the end it doesn't make much of a difference whether the
optimizations are done in the front or the middle end. The user knows
nothing about this, and he doesn't need to.

The problematic point here is that short-circuiting is an optimization
that is enabled already at -O0.


> Also, with a
> separate option it would be easy to check how it affects performance at
> different optimization levels.

For the case at hand, the short-circuiting is an absolutely valid
optimization. There is no reason why you wouldn't wanna do it (with -O
flags).


> What about making it a -fcheck=short-circuit-logicals (or however you want
> to spell it?) option, that also would be enabled with -fcheck=all?

What would such a flag even do? The actually invalid operation in the
test case is a null-pointer access, which could be caught by
-fcheck=pointer if we disable the optimization that removes it (i.e.
short-circuiting).

However, it also seems like -fcheck=pointer could use some
enhancement, since it does not even seem to catch simple cases like
this:

integer, pointer :: p => null()
print *,p
end

Cheers,
Janus