This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Repair TYPE_REF_CAN_ALIAS_ALL machinery


> > It turns out that I need more than the call-clobbered variables, I also
> > need the call-clobbered SMTs so NMTs won't do (see for example p.adb).
>
> Uh, why do you think this is true?

Because there was a bug in Gigi whereby

<bb 0>:
  D.386_1 = __gnat_malloc (4);

yielded

 D.386_1 = &ANYTHING

and I thought I needed to workaround it.  In fact, fixing this bug is 
sufficient to let the first testcase pass :-D.

> You need only an NMT with the same may-aliases as the SMT's have.  You
> shouldn't need to alias the bare SMT's, as they represent variables
> outside of the current scope anyway, unless you can also alias these.

Yes, I think this should work.  Thanks for the tip.

-- 
Eric Botcazou


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]