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]

Re: egcs-1.1 release schedule


> Jeff> Note that some non-ABI breaking changes to align doubles and
> Jeff> other values have gone into the x86 compiler.  In particular we
> Jeff> should be properly aligning all data in the static store.
>
>Does that make -malign-double redundant?

See my just-emailed long answer, but the short answer is "no".

With -malign-double, g77 0.5.23, egcs 1.0, and the 19980615 snapshot
align automatic and COMMON doubles.  Without it, they don't.  And,
without it, g77 0.5.23 and egcs 1.0 don't even align static doubles!

So the improvement in egcs 1.1-in-development is that -malign-double
is not needed to align the static doubles, but there's still the
automatic and, for okay-to-break-ABI cases, COMMON doubles to align,
not to mention all those stack-based doubles (IMO probably the most
important of the lot, but maybe you or others know better).

However, none of these newer g77's align stack-based doubles *regardless*
of -malign-double.  That, *at least*, should be fixed in egcs 1.1,
as it represents a serious regression even for performance-hungry
g77 users who use -malign-double against g77 0.5.21 and 0.5.22.
(Though with 0.5.2[12] they really have to use -O as well, though
of course they would be anyway.  I don't know why -O makes a
difference, offhand.)

        tq vm, (burley)


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