This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Use generic lowlevellock-futex.h in x86_64 lowlevellock.h.
- From: Torvald Riegel <triegel at redhat dot com>
- To: Roland McGrath <roland at hack dot frob dot com>
- Cc: GLIBC Devel <libc-alpha at sourceware dot org>
- Date: Thu, 18 Dec 2014 00:54:28 +0100
- Subject: Re: [PATCH] Use generic lowlevellock-futex.h in x86_64 lowlevellock.h.
- Authentication-results: sourceware.org; auth=none
- References: <1418858157 dot 20194 dot 30 dot camel at triegel dot csb> <20141217233744 dot 01E612C3ABF at topped-with-meat dot com>
On Wed, 2014-12-17 at 15:37 -0800, Roland McGrath wrote:
> You didn't describe testing you did. For this, it seems like you should
> not have too much trouble examining the generated code just by eyeball and
> characterizing the differences to give us confidence.
I tested on x86_64 and there are no regressions. If I hadn't done any
testing on the only arch this affects, I would have mentioned this in
the description of the patch (or not submitted the patch if I actually
encountered regressions).
I did not inspect the generated code because on x86_64, the futex calls
used from C code are on the slow paths. Looking now at pthread_once.o,
the generated code seems reasonable -- although I don't claim to have
checked whether it's optimal.