Re: libatomic, 32-bit object code vs. 64-bit object, GCC 8.3.0

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, 20 Jun 2019 at 21:26, Kacvinsky, Tom <Tom.Kacvinsky@xxxxxxxxxx> wrote:
>
> Hi,
>
> > -----Original Message-----
> > From: Jonathan Wakely <jwakely.gcc@xxxxxxxxx>
> > Sent: Thursday, June 20, 2019 11:54 AM
> > To: Kacvinsky, Tom <Tom.Kacvinsky@xxxxxxxxxx>
> > Cc: gcc-help@xxxxxxxxxxx
> > Subject: Re: libatomic, 32-bit object code vs. 64-bit object, GCC 8.3.0
> >
> > On Thu, 20 Jun 2019 at 16:19, Kacvinsky, Tom <Tom.Kacvinsky@xxxxxxxxxx>
> > wrote:
> > >
> > > Hi,
> > >
> > > I have built a 32-bit GCC 8.3.0 (no multi-lib support) on RHEL 5 and a
> > > 64-bit GCC 8.3.0 (again, no multi-lib support) on CentOS 5.11.  What I
> > > find interesting is that when I go to compile a program that will need
> > > atomic locks (a Boost template class pulled in via a header file), the
> > > 32-bit object code requires the library -latomic, but 64-bit object code does
> > not require that.
> > >
> > > Looking over "info gcc" I see a multitude of options for atomic
> > > operations.  The take away for me sis that for architectures that have
> > > an instruction set that has support for atomic operations, libatomic is not
> > necessary, but for other architectures, it is.
> >
> > Right. Sort of. It depends what kind of variables you're using the atomic
> > operations with.
> >
> > libatomic can still be needed on x86_64, but only for types larger than 64-bits.
> >
>
> OK, for our purposes we are using integral types of 64-bits wide or less.
>
> > On i386 you need libatomic even for 32-bit integers, as it has no atomic
> > operations. i586 has 64-bit atomics though.
> >
> > So if you configured your 32-bit compiler for i386, not i586 or later, then by
> > default it will want to use libatomic for all atomic ops.
>
> If  by configuring GCC for i386 you mean having i386 in the build triplet (used in the
> the --build option), then yes indeed I compiled my 32-bit GCC for i386.  Not sure if
> we should target i586 (or higher).

Right. You can either configure --with-arch=i586 to set the default
-march value, or equivalently use i586 in the triplet.

Either way, that only sets the default, you can still change it
per-compilation using -march=i586 or -march=whatever, but if you
originally built for i386 then the C++ runtime library (libstdc++)
will be built for i386 and so will assume no atomics, and so probably
require libatomic anyway.


> All of that said, it's not a problem to use libatomic, I just wanted to understand why
> our 32-bit product needs libatomic but our 64-bit product does not.
>
> Thanks for the help.

You're welcome.



[Index of Archives]     [Linux C Programming]     [Linux Kernel]     [eCos]     [Fedora Development]     [Fedora Announce]     [Autoconf]     [The DWARVES Debugging Tools]     [Yosemite Campsites]     [Yosemite News]     [Linux GCC]

  Powered by Linux