Date: Sun, 13 Jan 2013 12:33:20 -0800 From: John-Mark Gurney <jmg@funkthat.com> To: Konstantin Belousov <kostikbel@gmail.com> Cc: Adrian Chadd <adrian@freebsd.org>, freebsd-arch@freebsd.org Subject: Re: how long to keep support for gcc on x86? Message-ID: <20130113203320.GP1410@funkthat.com> In-Reply-To: <20130113132402.GR2561@kib.kiev.ua> References: <20130112233147.GK1410@funkthat.com> <20130113014242.GA61609@troutmask.apl.washington.edu> <CAJ-VmomrSFXcZg%2BKj6C2ARhpmjB9hxZATYJyRZB7-eRrcBLprg@mail.gmail.com> <20130113053725.GL1410@funkthat.com> <CAJ-VmomGKayr-1VucfwgodhXEHrXxx8r=9crHZJf74iVKZyTmQ@mail.gmail.com> <CAGE5yCoT4NZ2ULS60oZTXhQGgTbLRMZRvHmzioS7ToK9L8aZ_A@mail.gmail.com> <20130113132402.GR2561@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
Konstantin Belousov wrote this message on Sun, Jan 13, 2013 at 15:24 +0200:
> On Sun, Jan 13, 2013 at 12:09:09AM -0800, Peter Wemm wrote:
> > On Sat, Jan 12, 2013 at 11:44 PM, Adrian Chadd <adrian@freebsd.org> wrote:
> >
> > > Thus I think adding clang-only code to the system right now is very,
> > > very premature. There still seem to be reasons to run systems on GCC
> > > instead of clang.
> >
> > I don't have a problem with it so long as the system isn't *broken* if
> > you're not using clang. ie: if the status-quo is maintained for gcc
> > systems and g-faster bits are enabled with clang. It's fine to
> > provide incentives to try clang, but it is not ok to regress the gcc
> > case.
> Absolutely agree.
>
> Please note that in the AES-NI case, gcc 'support' is only partially
> gcc issue, if gcc at all. Our 2.17 gas does not know about AES-NI
> mnemonics and cannot assemble them.
gcc support would be better than gas support belive it or not..
> AFAIR the patch uses C built-in for AES-NI and SSE3 or 4, which I think
> could be implemented manually in the amount needed for the patch, for
> old gcc.
It was actually using assembly functions since gcc doesn't have the
necessary intrinsics... If it did, I could make things go even faster
for both i386 and amd64... As it stands, the i386 implementation will
be a bit slower since I'll have to pass in more of the blocks via stack
instead of registers... amd64 allows 8 128bit args passed in via reg,
while i386 only allows three...
--
John-Mark Gurney Voice: +1 415 225 5579
"All that I will do, has been done, All that I have, has not."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130113203320.GP1410>