This is the mail archive of the gcc-patches@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]

Re: Alias changes


>>>>> "Jeffrey" == Jeffrey A Law <law@hurl.cygnus.com> writes:

    Jeffrey> I'd prefer to have this option off by default for the
    Jeffrey> short term; just call me Mr. Conservative.  I'll

Dear Mr. Conservative :-)

    Jeffrey> personally be trying it, but I'm a little uncomfortable
    Jeffrey> having it be on by default for the upcoming release.

Fine.

    Jeffrey> Somewhere in the rtl/tree changes you should note that
    Jeffrey> the alias set is just an integer and does not directly
    Jeffrey> correspond to a particular type, storage class, memory
    Jeffrey> location, etc.  That's a pretty fundamental concept that
    Jeffrey> needs a mention somewhere (if I missed it, then ignore my
    Jeffrey> comment).

OK.

    Jeffrey> I think we need to save/restore the current alias set #
    Jeffrey> as we change function contexts so that the current alias
    Jeffrey> set # isn't scrogged when we compile a nested function.
    Jeffrey> Even if this isn't an issue right now we need to go ahead
    Jeffrey> and save/restore the set #.  It's simple to do, so I
    Jeffrey> don't think you need to resubmit after making this
    Jeffrey> change.

I don't quite get this.  With my patch, the C front-end assigns alias
sets to types lazily, as it creates MEMs who need alias sets.  It
doesn't make sense to have different types in the same alias set, so I
don't see why we'd ever want to restore the alias set counter to a
previous value.  It should monotonically increase, I think.

    Jeffrey> We've already discussed the long vs int stuff.

Yup, got it.

    Jeffrey> I think you should install the patch after making the
    Jeffrey> changes noted above.

Thanks.  I'm going to go ahead and do that, but *without* the
save/restore across function contexts.  I'll be happy to go ahead and
add that if you convince me it's needed.

    Jeffrey> jeff

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com


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