This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 2/7] Don't disable SSE in x86-64 ld.so
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Roland McGrath <roland at hack dot frob dot com>
- Cc: GNU C Library <libc-alpha at sourceware dot org>
- Date: Tue, 25 Aug 2015 14:55:30 -0700
- Subject: Re: [PATCH 2/7] Don't disable SSE in x86-64 ld.so
- Authentication-results: sourceware.org; auth=none
- References: <20150825122223 dot GA1588 at gmail dot com> <20150825183521 dot 2898C2C3A73 at topped-with-meat dot com> <CAMe9rOrGNoUMu9=yhp3xrXTcy2CqG-BgVR26qw4O0pry_azM7g at mail dot gmail dot com> <20150825210854 dot AF8382C3A94 at topped-with-meat dot com> <CAMe9rOoLbnkEJiKQDcNPu=3CRsoraLHbinPnNAd5rXQqXafEWQ at mail dot gmail dot com> <20150825213151 dot 029E82C3A94 at topped-with-meat dot com> <CAMe9rOpNOzCWTfH6V+5WU3OM2VG+trywzf_Zogm0GCHZOPGz0g at mail dot gmail dot com> <20150825214835 dot 2E6A02C3A94 at topped-with-meat dot com>
On Tue, Aug 25, 2015 at 2:48 PM, Roland McGrath <roland@hack.frob.com> wrote:
>> We use -mno-mmx on i386 since mm registers are passed to pass
>> __m64 arguments. On x86-64, I can't think of a good reason to use
>> MMX in ld.so.
>
> The patch is OK now since -mno-mmx was the status quo ante for x86_64.
>
> But the rationale for -mno-mmx is being stated in a backwards fashion.
> We should add special options only if there is a good reason to have
> them. If you just don't think it's optimal to use MMX, then that is not
> a good reason to use -mno-mmx; it's a good reason to make some compiler
> change to improve optimization. If there is a specific, known,
> pathological optimization issue, then there should be a GCC bug filed
> for that and the comments in the makefile should point to that bug. If
> there is no such reason to specifically use -mno-mmx, then we should
> drop it.
Here is a partial list of MMX bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=48397
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47759
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=40983
--
H.J.