[PATCH] align_value attribute in Clang
Hal Finkel
hfinkel at anl.gov
Wed Oct 1 17:10:21 PDT 2014
----- Original Message -----
> From: "Alexey Bataev" <a.bataev at hotmail.com>
> To: "Richard Smith" <richard at metafoo.co.uk>
> Cc: "cfe commits" <cfe-commits at cs.uiuc.edu>, "aaron ballman" <aaron.ballman at gmail.com>, "Michael Spencer"
> <bigcheesegs at gmail.com>, reviews+d4635+public+b7e82bdc1d8c324a at reviews.llvm.org, "Hal Finkel" <hfinkel at anl.gov>
> Sent: Tuesday, September 30, 2014 5:03:00 AM
> Subject: Re: [PATCH] align_value attribute in Clang
>
> Hi Richard,
> your guess is correct, it is applied to the outer pointer.
Okay, in that case if we keep it out of the type system, then I suppose the current patch is essentially okay. The optimizer will deduce the alignment information when it can (especially after inlining), but if it is not a separate type, then we actually can't have it follow template deduction otherwise because it can't affect the mangling. Sound right?
-Hal
>
> Best regards,
> Alexey Bataev
> =============
> Software Engineer
> Intel Compiler Team
>
> 30.09.2014 4:50, Richard Smith пишет:
> > On Mon, Sep 29, 2014 at 11:36 AM, Hal Finkel <hfinkel at anl.gov
> > <mailto:hfinkel at anl.gov>> wrote:
> >
> > Hi Richard,
> >
> > What does all of this suggest to you about the right design?
> >
> >
> > One more question. Given:
> >
> > typedef int * __attribute__((align_value(32))) *Y;
> >
> > ... does the attribute apply to the outer pointer or the inner
> > pointer? (My guess: it surprisingly applies to the outer pointer.)
> > If
> > that's the case, then I think align_value should live entirely
> > outside
> > the type system, in the same way that __attribute__((aligned))
> > does.
> > Note that, given:
> >
> > typedef int __attribute__((aligned(64))) *p
> > __attribute__((aligned(8)))
> >
> > p is a 64-byte-aligned pointer to int, not an 8-byte-aligned
> > pointer
> > to a 64-byte-aligned int.
> >
> > If that's the case, I suggest you model it exactly like AlignedAttr
> > --
> > as an attribute on the TypedefDecl -- and look for it explicitly in
> > the cases where you want to use it by walking the type sugar and
> > looking for a TypedefDecl.
> >
> > FWIW, maybe creating some kind of synthetic typedef type for
> > the
> > return value from DeduceTemplateArgumentsByTypeMatch would
> > work?
> >
> > Thanks again,
> > Hal
> >
> > ----- Original Message -----
> > > From: "Alexey Bataev" <a.bataev at hotmail.com
> > <mailto:a.bataev at hotmail.com>>
> > > To: "Hal Finkel" <hfinkel at anl.gov <mailto:hfinkel at anl.gov>>,
> > reviews+D4635+public+b7e82bdc1d8c324a at reviews.llvm.org
> > <mailto:reviews%2BD4635%2Bpublic%2Bb7e82bdc1d8c324a at reviews.llvm.org>
> > > Cc: cfe-commits at cs.uiuc.edu <mailto:cfe-commits at cs.uiuc.edu>,
> > "aaron ballman" <aaron.ballman at gmail.com
> > <mailto:aaron.ballman at gmail.com>>, bigcheesegs at gmail.com
> > <mailto:bigcheesegs at gmail.com>, richard at metafoo.co.uk
> > <mailto:richard at metafoo.co.uk>
> > > Sent: Friday, September 26, 2014 12:26:29 AM
> > > Subject: Re: [PATCH] align_value attribute in Clang
> > >
> > > 1. ICC does not create a new type for type with this
> > > attribute. All
> > > types are still the same.
> > > 2.
> > >
> > > int n __attribute__((aligned(32)));
> > > decltype(&n) - int*.
> > >
> > >
> > > Best regards,
> > > Alexey Bataev
> > > =============
> > > Software Engineer
> > > Intel Compiler Team
> > >
> > > 26.09.2014 4:28, Hal Finkel пишет:
> > > > ----- Original Message -----
> > > >> From: "Richard Smith" <richard at metafoo.co.uk
> > <mailto:richard at metafoo.co.uk>>
> > > >> To: hfinkel at anl.gov <mailto:hfinkel at anl.gov>, "aaron
> > > >> ballman"
> > <aaron.ballman at gmail.com <mailto:aaron.ballman at gmail.com>>,
> > > >> bigcheesegs at gmail.com <mailto:bigcheesegs at gmail.com>,
> > richard at metafoo.co.uk <mailto:richard at metafoo.co.uk>, "a
> > > >> bataev" <a.bataev at hotmail.com
> > > >> <mailto:a.bataev at hotmail.com>>
> > > >> Cc: cfe-commits at cs.uiuc.edu
> > > >> <mailto:cfe-commits at cs.uiuc.edu>
> > > >> Sent: Thursday, September 25, 2014 7:06:33 PM
> > > >> Subject: Re: [PATCH] align_value attribute in Clang
> > > >>
> > > >> I'd like some more information about how ICC treats this
> > > >> attribute.
> > > >> Does it produce a new type? For instance:
> > > > I did a quick experiment with icc (ICC) 14.0.1 20131008
> > > > (with
> > -Wall
> > > > -O3 -pedantic):
> > > >
> > > >> typedef int *I32 __attribute__((align_value(32)));
> > > >> typedef int *I64 __attribute__((align_value(64)));
> > > >>
> > > >> typedef I32 X; typedef I64 X; // ill-formed?
> > > > No diagnostic.
> > > >
> > > >> template<typename> struct Y;
> > > >> typedef Y<I32> Z; typedef Y<I64> Z; // ill-formed?
> > > > No diagnostic.
> > > >
> > > >> extern I32 i32;
> > > >> I64 i64 = i32; // ill-formed? Is an explicit cast
> > > >> required?
> > > > No diagnostic.
> > > >
> > > >> I32 i32 = i64; // ill-formed? Is there an implicit
> > > >> conversion
> > > >> that
> > > >> discards alignment?
> > > > No diagnostic.
> > > >
> > > > Alexey, can you please comment on the specifics?
> > > >
> > > > -Hal
> > > >
> > > >> ... and if there's an implicit conversion that discards
> > alignment,
> > > >> what is its conversion rank?
> > > >>
> > > >> If I have:
> > > >>
> > > >> int n __attribute__((aligned(32)));
> > > >>
> > > >> ... then what is `decltype(&n)`? Is it `int*`, or `int*
> > > >> __attribute__((align_value(32)))`?
> > > >>
> > > >> http://reviews.llvm.org/D4635
> > > >>
> > > >>
> > > >>
> > >
> > >
> >
> > --
> > Hal Finkel
> > Assistant Computational Scientist
> > Leadership Computing Facility
> > Argonne National Laboratory
> >
> >
>
>
--
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory
More information about the cfe-commits
mailing list