Hi,
On Tue, 4 Oct 2011, Jakub Jelinek wrote:
> If it causes miscompilations, then GCC is buggy and needs to be fixed.
That's not easy without effectively making the whole restrict support
ineffective.
> > Keep in mind that the whole current restrict-tag support was designed to
> > support fortrans specific needs where we can rely on the compiler
> > generating code we most probably (!) won't miscompile later. Further we
> > interpret some of the part where ISO-C says nothing in our favor.
> > Together this means that e.g. (intentionally _not_ ISO-C, but sort of
> > middle-end speak):
> >
> > foo (struct s)
> > { restrict int *p1 = s.p;
> > restrict int *p2 = s.p;
> > *p1 = 0; *p2 = 1; return *p1;
> > }
>
> Not sure what you mean here. restrict int doesn't make sense,
I said that it wasn't C. Read all my examples as "int * restrict pX", ...
> int *restrict p1 would make sense, but then it would be invalid C.
... and invalidity in C doesn't necessarily enter the picture. The above
is middle-end instructions and we need to say what we wish to happen with
the above code (which _is_ generated by valid fortran use for instance,
and could conceivably be generated by other middle-end transformation,
e.g. code duplication.
> Loads don't generate new tags.
See Richis (upcoming) answer, that's one of the various hacks^Wdesign
decisions made to at least get something useful out of the restrict tag
things for fortran without breaking too much.
> > You introduce even more freedom to this mud. Effectively:
> >
> > foo (struct s)
> > { restrict int *p1 = s.p;
> > restrict int *p2 = s.q; // same restrict tag as s.p
> > }
>
> Again, in your above syntax it is unclear what is TYPE_RESTRICT, if it is
> TYPE_RESTRICT (TREE_TYPE (p1)) or something else. If the user writes the
> above in C and then accesses both through p1 and p2 the same object, it is
> invalid.
Hmm? I thought the point of your patch was precisely to make accesses
through _M_start and _M_final have the same restrict tag, so that users
can access the same object through both pointers, but still enjoy
disambiguation with e.g. _M_start from different objects. Possibly I've
misunderstood.
Ciao,
Michael.
