This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Remove statisticaly unsound benchmarks
- From: David Miller <davem at davemloft dot net>
- To: siddhesh at redhat dot com
- Cc: neleai at seznam dot cz, libc-alpha at sourceware dot org
- Date: Tue, 23 Apr 2013 13:24:37 -0400 (EDT)
- Subject: Re: [PATCH] Remove statisticaly unsound benchmarks
- References: <20130419124602 dot GA16032 at domone dot kolej dot mff dot cuni dot cz> <20130423124436 dot GU16329 at spoyarek dot pnq dot redhat dot com>
From: Siddhesh Poyarekar <siddhesh@redhat.com>
Date: Tue, 23 Apr 2013 18:14:36 +0530
> On Fri, Apr 19, 2013 at 02:46:02PM +0200, Ondřej Bílka wrote:
>> Given these flaws any results that use them are unscientific and we
>> should remove them.
>>
>> * debug/test-strcpy_chk.c (do_one_test, do_test): Remove benchmark.
>> * string/test-memccpy.c (do_one_test, do_test): Likewise.
>> * string/test-memccpy.c (do_one_test, do_test): Likewise.
>> * string/test-memchr.c (do_one_test, do_test): Likewise.
>> * string/test-memcmp.c (do_one_test, do_test): Likewise.
>> * string/test-memcpy.c (do_one_test, do_test): Likewise.
>> * string/test-memmem.c (do_one_test, do_test): Likewise.
>> * string/test-memmove.c (do_one_test, do_test): Likewise.
>> * string/test-memset.c (do_one_test, do_test): Likewise.
>> * string/test-rawmemchr.c (do_one_test, do_test): Likewise.
>> * string/test-strcasecmp.c (do_one_test, do_test): Likewise.
>> * string/test-strcasestr.c (do_one_test, do_test): Likewise.
>> * string/test-strcat.c (do_one_test, do_test): Likewise.
>> * string/test-strchr.c (do_one_test, do_test): Likewise.
>> * string/test-strcmp.c (do_one_test, do_test): Likewise.
>> * string/test-strcpy.c (do_one_test, do_test): Likewise.
>> * string/test-strlen.c (do_one_test, do_test): Likewise.
>> * string/test-strncasecmp.c (do_one_test, do_test): Likewise.
>> * string/test-strncat.c (do_one_test, do_test): Likewise.
>> * string/test-strncmp.c (do_one_test, do_test): Likewise.
>> * string/test-strncpy.c (do_one_test, do_test): Likewise.
>> * string/test-strnlen.c (do_one_test, do_test): Likewise.
>> * string/test-strpbrk.c (do_one_test, do_test): Likewise.
>> * string/test-strrchr.c (do_one_test, do_test): Likewise.
>> * string/test-strspn.c (do_one_test, do_test): Likewise.
>> * string/test-strstr.c (do_one_test, do_test): Likewise.
>> * string/test-string.h (do_one_test, do_test): Remove HP_TIMING_BEST.
>
> I do not agree with removing them without putting in place an
> equivalent or better set of performance tests. If you're interested
> in adding such tests then please do so first and then propose a patch
> to remove these tests. If you want, I can hold off my patch to move
> these tests into benchtests for a while.
>
> I'd really appreciate if you could commit to contributing such tests
> within the 2.18 timeframe though, because I don't want to hold off the
> change without reason. If you cannot make it within 2.18 due to any
> other constraints/commitments then let me do the move and then you can
> put your changes on top of them.
I also strongly disagree with removing these tests without a
replacement, even with their flaws they are extremely useful and I do
use them myself.