This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] powerpc: strcasestr optimization
- From: Roland McGrath <roland at hack dot frob dot com>
- To: munroesj at linux dot vnet dot ibm dot com
- Cc: OndÅej BÃlka <neleai at seznam dot cz>, Rajalakshmi Srinivasaraghavan <raji at linux dot vnet dot ibm dot com>, GNU C Library <libc-alpha at sourceware dot org>, Steve Munroe <sjmunroe at us dot ibm dot com>
- Date: Tue, 9 Jun 2015 14:42:02 -0700 (PDT)
- Subject: Re: [PATCH] powerpc: strcasestr optimization
- Authentication-results: sourceware.org; auth=none
- References: <55687597 dot 1060101 at linux dot vnet dot ibm dot com> <556C36D8 dot 2070208 at linux dot vnet dot ibm dot com> <20150601122830 dot GA14649 at domone> <1433169407 dot 10235 dot 5 dot camel at sjmunroe-ThinkPad-W500> <20150601162215 dot GA8955 at domone> <1433270823 dot 14958 dot 15 dot camel at sjmunroe-ThinkPad-W500> <20150602210000 dot GB20004 at domone> <1433882432 dot 21101 dot 71 dot camel at sjmunroe-ThinkPad-W500> <20150609210555 dot 3A7CA2C3BDC at topped-with-meat dot com> <1433885184 dot 21101 dot 83 dot camel at sjmunroe-ThinkPad-W500>
> If Ond ej Bílka believes there is a problem that only he understands
> then the onus is on him to provide the appropriate benchmark.
Agreed. It's reasonable to hold up the change based on announced intent to
quickly supply a benchmark that changes the available objective analysis of
the change's performance effect. But if the objector doesn't commit to
delivering that benchmark soon, then he doesn't have grounds to object to
the change.
OTOH, if it's the case that the only benchmarks we have for a particular
function are very poor representatives and the consensus agrees on that
assessment, then it's reasonable to demand that better benchmarks be
achieved before changes claimed to improve performance of that function can
be entertained. (I don't know if that's the situation here; I'm just
talking about the general policy of the project.)
In short, if you intend either to propose a lot of performance-improvement
changes or to be skeptical of such proposals by others, then you should
expect to spend a lot of effort on writing better benchmarks and reaching
consensus on their quality.
Thanks,
Roland