>reasons for these failure cases are actually known by the egcs
>developers, and they realize how insurmountable it is to fix "right
>this moment".
Actually, the problems aren't as insurmountable as it would seem.
It's just that gcc is messing up on the housekeeping of which registers
to use. This is something which doesn't only happen with regparm, it
can happen without as well (it's just a lot less likely).
>Richard Kenner made a huge mistake by letting regparm on the intel
>into the gcc code in the first place, because of this crucial
>fundamental flaw which prevents them from working reliably.
Actually, in practice, the only cases which were still causing problems
were the indirect function calls. They have the potential to test
the compiler's register allocation routines to the max.
All other cases had been resolved (I'm not current on any new problems
that ecgs might have created).
>So don't blame the egcs people for trying to deal with the mess Kenner
>created.
Well, to be more precise, I created a working regparm=5 gcc, Kenner insisted
on the patches being *completely* generic, I worked with him most
of the way, but couldn't resolve this last issue (took me four months
of arguing to get my proposed fixes to be denied).
>regard, cut them slack for once for crying out loud. They've chosen
>to disable a feature instead of allowing it to generate incorrect
>code. I think thats a good engineering decision.
Yes, and no. If you turn it off, there is less incentive to fix it.
In order to fix it, someone needs to dig into it again. If that person
is not available (I know I don't have the time right now), it's a problem.
OTOH, the case where regparm doesn't work used to be rather well-defined.
>I think that with how much you knew yourself about it's brokenness,
>relying on it for the few cases where your version of gcc happens to
>get it right is playing with fire.
Not really, the cases where it doesn't work are few (unless egcs has
introduced new problems).
>listen to everything I say and leave on all the features I want for
>the kernel, even if they are broken." The people who sweat night and
>day on egcs take offense to this, and now in this instance myself
>included.
People who've sweated night and day to debate with Richard Kenner and generate
yet another variation to get regparm patches approved have feelings too,
though :-).
--
Sincerely, srb@cuci.nl
Stephen R. van den Berg (AKA BuGless).
Skiing beyond this point may result in death and/or loss of skiing privileges.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html