git: af366d353b84 - main - amd64: implement strlen in assembly

Tue Feb 9 03:53:00 UTC 2021

Warner,

My intent was not to attack Jessica just as I do not believe her
intent was to pick on Mateuz but from different perspectives
frustrations can build up to the point that the message is not well
received.

What I intended was the opposite:  to encourage tact in this kind of exchange.

If we can start over in the future, the best way to do this is
something like:  "Mateuz, this is interesting work.  I would like to
have a chance to review your amd64 assembly changes before you commit
them... [because $reasons might be nice to help someone internally
process and change their behavior, but without justification sure it's
just implied they care about the quality of the code and that is
enough]"..

I stand behind my position that commanding someone to do something
without offer or without having an existing relationship such that
they will receive the request in the way it was intended is a good way
to burn people out.  There are two contributors here.  Siding with one
is a negative outcome.  I am trying to propose an actionable criteria
that Mateuz can act on and have good faith in because I know him well
enough to know he will do it.

Regards,
Kevin

On Mon, Feb 8, 2021 at 8:20 PM Warner Losh <imp at bsdimp.com> wrote:
>
> Kevin,
>
> I'm sure that you think you are being reasonable. But you sure are coming off as attacking Jessica and I for a polite request to adhere to documented project norms. It's not unreasonable to make a request. You are proposing a crazy and unreasonable standard by attacking Jessica and I with the "would you have reviewed it?" line. I personally look at every single src review. I don't comment on them all. I don't know if I can until I take a look, but I at least read the summaries.
>
> It's cool that mjg has gone ahead and rewritten strlen. However, that's not a blank check to ignore project norms, nor for others to attack those that suggest it. The code is only about 50 lines of assembler, well within the range of code that often gets a good review. Suggesting that it would have benefitted from a review is not picking on mjg.
>
> Anyway, this is the last thing I'll post here... These threads get way out of hand, so if you want to post a rebuttal, I'll give you the last word.
>
> Warner
>
> On Mon, Feb 8, 2021 at 5:18 PM Kevin Bowling <kevin.bowling at kev009.com> wrote:
>>
>> I understand your position and Warner’s from the documentation.  The problem which is not described is that frustration is asymptotically higher in the other direction without volunteering to do work.  As another example I could reply and ask for unit tests for any change (tests are obviously helpful too) but unless I am willing to help it is just a suggestion and should not be sent as a command.  If you are willing to do such reviews timely, or have command of someone who will, I will coordinate with mjg and person.  Otherwise it’s volunteering other people’s time and reduces the willpower to fix these performance areas.
>>
>> The head model supports occasional break and revert.  I think this is important given the resources FreeBSD has available, since a lot of the less glamorous work is unpaid and underpaid.
>>
>> Picking on mjg for this is suspicious given how frequently breakage happens by anyone right now juxtaposed to his track record of improvements and quickness to address issues or revert where issue arose.
>>
>> Regards,
>> Kevin
>>
>> On Mon, Feb 8, 2021 at 4:19 PM Jessica Clarke <jrtc27 at freebsd.org> wrote:
>>>
>>> > On 8 Feb 2021, at 23:13, Kevin Bowling <kevin.bowling at kev009.com> wrote:
>>> >
>>> > FreeBSD does not require pre-commit approval unless called out
>>> > specifically.  Are you volunteering to review the changes, and if so
>>> > where is your guidance?  These messages are otherwise unhelpful.
>>>
>>> It is not a hard requirement, but it is strongly encouraged. Section 7
>>> of the committer's guide says:
>>>
>>>         • All non-trivial changes should be reviewed before they are
>>>         committed to the repository.
>>>
>>> This was a non-trivial change. I was particularly frustrated to see
>>> this commit go in without review having previously called out mjg@ for
>>> not getting any reviews for his (now reverted) previous strlen change.
>>>
>>> Jess
>>>
>>> > On Mon, Feb 8, 2021 at 12:37 PM Jessica Clarke <jrtc27 at freebsd.org> wrote:
>>> >>
>>> >> On 8 Feb 2021, at 19:15, Mateusz Guzik <mjg at FreeBSD.org> wrote:
>>> >>>
>>> >>> The branch main has been updated by mjg:
>>> >>>
>>> >>> URL: https://cgit.FreeBSD.org/src/commit/?id=af366d353b84bdc4e730f0fc563853abc338271c
>>> >>>
>>> >>> commit af366d353b84bdc4e730f0fc563853abc338271c
>>> >>> Author:     Mateusz Guzik <mjg at FreeBSD.org>
>>> >>> AuthorDate: 2021-02-08 17:01:48 +0000
>>> >>> Commit:     Mateusz Guzik <mjg at FreeBSD.org>
>>> >>> CommitDate: 2021-02-08 19:15:21 +0000
>>> >>>
>>> >>>   amd64: implement strlen in assembly
>>> >>>
>>> >>>   The C variant in libkern performs excessive branching to find the
>>> >>>   non-zero byte instead of using the bsfq instruction. The same code
>>> >>>   patched to use it is still slower than the routine implemented here
>>> >>>   as the compiler keeps neglecting to perform certain optimizations
>>> >>>   (like using leaq).
>>> >>>
>>> >>>   On top of that the routine can is a starting point for copyinstr
>>> >>>   which operates on words instead of bytes.
>>> >>>
>>> >>>   Tested with glibc test suite.
>>> >>>
>>> >>>   Sample results (calls/s):
>>> >>>
>>> >>>   Haswell:
>>> >>>   $(perl -e "print 'A' x 3"):
>>> >>>   stock:  211198039
>>> >>>   patched:338626619
>>> >>>   asm:    465609618
>>> >>>
>>> >>>   $(perl -e "print 'A' x 100"):
>>> >>>   stock:   83151997
>>> >>>   patched: 98285919
>>> >>>   asm:    120719888
>>> >>>
>>> >>>   AMD EPYC 7R32:
>>> >>>   $(perl -e "print 'A' x 3"):
>>> >>>   stock:  282523617
>>> >>>   asm:    491498172
>>> >>>
>>> >>>   $(perl -e "print 'A' x 100"):
>>> >>>   stock:  114857172
>>> >>>   asm:    112082057
>>> >>
>>> >> No Reviewed by? More than one pair of eyes on non-trivial assembly is
>>> >> almost always a good idea.
>>> >>
>>> >> Jess
>>> >>
>>> >> _______________________________________________
>>> >> dev-commits-src-main at freebsd.org mailing list
>>> >> https://lists.freebsd.org/mailman/listinfo/dev-commits-src-main
>>> >> To unsubscribe, send any mail to "dev-commits-src-main-unsubscribe at freebsd.org"
>>>