Date: Wed, 10 Jul 2013 07:55:53 -0600 From: Warner Losh <imp@bsdimp.com> To: Konstantin Belousov <kostikbel@gmail.com> Cc: Baptiste Daroussin <bapt@freebsd.org>, Adrian Chadd <adrian@freebsd.org>, Dimitry Andric <dim@FreeBSD.org>, Andrew Turner <andrew@fubar.geek.nz>, freebsd-arch@freebsd.org Subject: Re: Adding a MACHINE_ARCH note Message-ID: <C5CB098F-5FFA-46B5-BBBD-49869C902286@bsdimp.com> In-Reply-To: <20130710070319.GX91021@kib.kiev.ua> References: <20130709090744.0e497e7e@bender.Home> <32F979BD-FB5C-4111-9586-4C5E7C6DFA71@bsdimp.com> <20130709234837.559e3769@bender.Home> <CAJ-Vmo=iV8BsGriFRgNuP-ZJdQhpmBLhjAkz-nSVRS0HPKSyOQ@mail.gmail.com> <CAGE5yCpJmRDvnaYtozj4bCqNoQXH=1e96HPJAqwJuRdn4H9BZA@mail.gmail.com> <CAJ-Vmo=tmGDW3Ubw9nr5rb30bXr1dcJUkKLOU7L=_bx29zvEhw@mail.gmail.com> <CAGE5yCq9gQERDkbi4wu=6tNUap24ZR7sL7aF%2BzmEO0eT6nxPsA@mail.gmail.com> <F79E2F76-A234-499A-ABB7-1ABA62283E9D@FreeBSD.org> <20130710070319.GX91021@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On Jul 10, 2013, at 1:03 AM, Konstantin Belousov wrote: > On Wed, Jul 10, 2013 at 08:54:05AM +0200, Dimitry Andric wrote: >> On Jul 10, 2013, at 03:08, Peter Wemm <peter@wemm.org> wrote: >>> On Tue, Jul 9, 2013 at 5:56 PM, Adrian Chadd <adrian@freebsd.org> = wrote: >>>> ... boy I'd like to see this particular x86 hiccup fixed before = this >>>> stuff is mainstream. >>>=20 >>> I'm not entirely sure how much support there is behind "x32". I = don't >>> know if its much more than an academic curiosity or if there's real >>> demand for it. >>=20 >> It seems to be driven by Intel and Google. The idea is that for some >> applications (or maybe even most :), an ILP32 model will perform = better. >> Quoting from one of the presentations: >>=20 >> On Core i7 2600K 3.40GHz: >> - Improved SPEC CPU 2K/2006 INT geomean by 7-10% over ia32 and 5-8% = over >> Intel64. >> - Improved SPEC CPU 2K/2006 FP geomean by 5-11% over ia32. >> - Very little changes in SPEC CPU 2K/2006 FP geomean, comparing = against >> Intel64. >> - Comparing against ia32 PIC, x32 PIC: >> - Improved SPEC CPU 2K INT by another 10%. >> - Improved SPEC CPU 2K FP by another 3%. >> - Improved SPEC CPU 2006 INT by another 6% >> - Improved SPEC CPU 2006 FP by another 2%. >>=20 >> As to how often it is actually used in practice, I am not sure. >>=20 >>=20 >>> gcc-4.8 and clang have it, or have patches for it. >>=20 >> You also need a fairly recent binutils. And kernel + libc support... >> It is probably not a trivial task. :-) >=20 > You definitely need a support from libc, libthr and rtld. > I am not convinced that the kernel modifications are needed, > except for the image activator to recognize new ELF ids. In > other words, I believe it is better to put shims into libc in > the long run. For MIPS N32, you also need to make sure that the system calls grok the = ABI. You're going to give up some of the performance gains by shimming, = plus you run the risk of mixing x32 with amd64 code, which would be bad. = Attractive as it may sound to try to put it all in libc, I'm not = convinced that would work out in the end. Warner=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C5CB098F-5FFA-46B5-BBBD-49869C902286>