Re: [PATCH, middle end]: Fix PR 37908, thinko with atomic NAND operation

[David Daney <ddaney at caviumnetworks dot com>]

Lu, Hongjiu wrote:
> Hi,
>
> According to Dan, __sync_fetch_and_nand intrinsic should be implemented as ~(target & val). Uros's patch is correct.

At a minimum, it is incomplete.  If one were to change the semantics of 
__sync_fetch_and_nand, they probably should do it universally rather 
than just for a subset of GCC's targets.

The MIPS port would have to be changed, and IIRC sparc as well.

David Daney


> Thanks.
>
> H.J.
> hongjiu.lu@intel.com
>
> > -----Original Message-----
> > From: Andrew Haley [mailto:aph@redhat.com]
> > Sent: Friday, October 24, 2008 6:01 AM
> > To: Jakub Jelinek
> > Cc: Uros Bizjak; GCC Patches; Lu, Hongjiu
> > Subject: Re: [PATCH, middle end]: Fix PR 37908, thinko with atomic NAND
> > operation
> >
> > Jakub Jelinek wrote:
> > > On Fri, Oct 24, 2008 at 01:29:55PM +0100, Andrew Haley wrote:
> > > > I don't think it's that exactly, but that what the builtins do is
> > > > wrong, or at best very misleading.  NAND is a well-defined term of
> > > > the art; it isn't defined by the gcc documentation.
> > > The name of the builtin may be misleading, but changing semantics of a
> > > builtin after it has been in use for 9 years is even worse.  The origin
> > > of the builtin is in ia64intrin.h and therefore icc, I guess, so
> > > it matters what icc is implementing.
> > Indeed.  It's a wonder no-one commented at the time.  At least, I can't
> > remember anyone doing so.
> >
> > FAOD, I'm certainly not suggesting that we should change anything except
> > perhaps to add a line or two to the documentation.
> >
> > Andrew.