This is the mail archive of the
mailing list for the glibc project.
Re: [RFC PATCH] aarch64: improve memset
- From: Andrew Pinski <pinskia at gmail dot com>
- To: Marcus Shawcroft <marcus dot shawcroft at gmail dot com>
- Cc: Wilco Dijkstra <wdijkstr at arm dot com>, Richard Henderson <rth at twiddle dot net>, Will Newton <will dot newton at linaro dot org>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Tue, 11 Nov 2014 09:22:36 -0800
- Subject: Re: [RFC PATCH] aarch64: improve memset
- Authentication-results: sourceware.org; auth=none
- References: <002701cffaa0$77623570$6626a050$ at com> <002801cffaa5$eb2852f0$c178f8d0$ at com> <CAFqB+Pw4oEhmORJGSjBNtaTn9ZOgWS6-25p=4AYFwGuv72jddg at mail dot gmail dot com>
On Tue, Nov 11, 2014 at 5:05 AM, Marcus Shawcroft
> On 7 November 2014 16:14, Wilco Dijkstra <email@example.com> wrote:
>>> Richard Henderson wrote:
>> I've got a few comments on this patch:
>> * Do we really need variants for cache line sizes that are never going to be used?
>> I'd say just support 64 and 128, and default higher sizes to no_zva.
> We shouldn't be removing support for the other sizes already supported
> by the existing implementation. If the other sizes were deprecated
> from the architecture then fair game, but that is not the case. From
> offline conversation with Wilco I gather part of the motivation to
> remove is that the none 64 cases cannot be readily tested on HW.
> That particular issue was solved in the original implementation using
> a hacked qemu.
I will have the ability to test on hardware which uses 128 byte case
soon. I already testing using a simulator which sets it to 128 byte
(though I use it for performance analysis though).
>> * Why special case line size=64 only? Unrolling might not help for 128 but should not
>> harm either, and the alignment overhead only increases with larger line sizes, so you
>> want to bypass the zva code in all cases if N < 3-4x line size.
>> * Is the no-ifunc variant still required/used? We're now having at least 4 different
>> variants which all need to be tested and maintained...
>> * Finally, which version is used when linking statically? I presume there is some
>> makefile magic that causes the no-zva version to be used, however that might not be
>> optimal for all targets.