This is the mail archive of the gcc@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: [lto][RFC] Do not emit hybrid object files


On Sat, Oct 18, 2008 at 8:47 AM, Diego Novillo <dnovillo@google.com> wrote:
> On Sat, Oct 18, 2008 at 08:33, Kenneth Zadeck <zadeck@naturalbridge.com> wrote:
>
>> I actually think that the hybrid files should be the default.  If you
>> are willing to make invasive changes to your build environment to
>> support two files, then you should be willing to add extra options to
>> support that.
>
> Sorry, I don't follow, what two files are you referring to?  Adding
> -flto to your Makefiles hardly qualifies as invasive.
>
> The only scenario where hybrid files are useful is when building a
> library that will be deployed to a third party that may not be using
> the same compiler version that was used to build the library.
>
> In this scenario, I agree that the final user (the user of the
> library) should not be forced to change their build process.  They
> would only change it if they are willing to take advantage of the
> additional optimization opportunities.
>
> The only ones who should request generation of hybrid files are the
> creators of the library.  And they are already changing their
> Makefiles to generate LTO information, the extra flag won't matter to
> them.

IBM XLC whole program IPA mode defaults to hybrid files.  One explicitly
specified -qipa=noobject to avoid emitting standard object files.  Hybrid by
default follows the rule of least surprise.

David


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