[PATCH] Have HasSideEffects() return false for __attribute__((const)) functions

[Hal Finkel]

----- Original Message -----
> From: "Aaron Ballman" <aaron.ballman at gmail.com>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: "Michael M Kuperstein" <michael.m.kuperstein at intel.com>, "Richard Smith" <richard at metafoo.co.uk>, "llvm cfe"
> <cfe-commits at cs.uiuc.edu>, reviews+D8548+public+9c0a033e02b7b4b5 at reviews.llvm.org
> Sent: Monday, March 23, 2015 12:00:39 PM
> Subject: Re: [PATCH] Have HasSideEffects() return false for __attribute__((const)) functions
> 
> On Mon, Mar 23, 2015 at 12:44 PM, Hal Finkel <hfinkel at anl.gov> wrote:
> > ----- Original Message -----
> >> From: "Aaron Ballman" <aaron.ballman at gmail.com>
> >> To: reviews+D8548+public+9c0a033e02b7b4b5 at reviews.llvm.org
> >> Cc: "Michael M Kuperstein" <michael.m.kuperstein at intel.com>,
> >> "Richard Smith" <richard at metafoo.co.uk>, "Hal Finkel"
> >> <hfinkel at anl.gov>, "llvm cfe" <cfe-commits at cs.uiuc.edu>
> >> Sent: Monday, March 23, 2015 11:38:02 AM
> >> Subject: Re: [PATCH] Have HasSideEffects() return false for
> >> __attribute__((const)) functions
> >>
> >> On Mon, Mar 23, 2015 at 12:33 PM, hfinkel at anl.gov
> >> <hfinkel at anl.gov>
> >> wrote:
> >> > In http://reviews.llvm.org/D8548#145253, @mkuper wrote:
> >> >
> >> >> I think PureAttr isn't strong enough.
> >> >>
> >> >> char foo(char *a)
> >> >>  {
> >> >>
> >> >>   return *a;
> >> >>
> >> >> }
> >> >>
> >> >> is pure, but isn't side-effect free.
> >> >
> >> >
> >> > Loads are not considered side effects currently, however. This
> >> > works fine:
> >> >
> >> >   void bar(int *i) {
> >> >     __builtin_assume(*i > 0);
> >> >   }
> >> >
> >> > and, thus, my recommendation ;)
> >>
> >> Some loads are side-effecting though, such as volatile loads:
> >>
> >> int foo(volatile int *a) {
> >>   return *a;
> >> }
> >>
> >> Can that be marked __attribute__((pure))? Or can pure access a
> >> volatile global variable?
> >
> > I think that the answer is no, but only by indirect inference. The
> > GCC docs say:
> >
> > [from GCC docs]
> > Many functions have no effects except the return value and their
> > return value depends only on the parameters and/or global
> > variables. Such a function can be subject to common subexpression
> > elimination and loop optimization just as an arithmetic operator
> > would be.
> > [end from GCC docs]
> >
> > and since volatile loads often cannot be CSE'd, I suspect they're
> > meant to be excluded. What do you think?
> 
> I'm not overly familiar with __attribute__((pure)), but what you say
> seems logical. Looking at the GCC docs a bit more closely, it also
> says: "Interesting non-pure functions are functions with infinite
> loops or those depending on volatile memory or other system resource,
> that may change between two consecutive calls (such as feof in a
> multithreading environment). "

Indeed; good point.

> 
> So I think it might be safe to do this for __attribute__((pure)) as
> well.

Cool; let's do it!

 -Hal

> 
> ~Aaron
> 
> >
> >  -Hal
> >
> >>
> >> ~Aaron
> >>
> >> >
> >> >
> >> > http://reviews.llvm.org/D8548
> >> >
> >> > EMAIL PREFERENCES
> >> >   http://reviews.llvm.org/settings/panel/emailpreferences/
> >> >
> >> >
> >>
> >
> > --
> > Hal Finkel
> > Assistant Computational Scientist
> > Leadership Computing Facility
> > Argonne National Laboratory
> 

-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory