This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] powerpc: Fix logbl on power7 [BZ# 21280]
- From: Adhemerval Zanella <adhemerval dot zanella at linaro dot org>
- To: munroesj at linux dot vnet dot ibm dot com
- Cc: Tulio Magno Quites Machado Filho <tuliom at linux dot vnet dot ibm dot com>, libc-alpha at sourceware dot org
- Date: Fri, 28 Apr 2017 16:48:50 -0300
- Subject: Re: [PATCH] powerpc: Fix logbl on power7 [BZ# 21280]
- Authentication-results: sourceware.org; auth=none
- References: <1490043778-30228-1-git-send-email-tuliom@linux.vnet.ibm.com> <50b87cce-2f3a-d84b-bc38-d81fa3623738@linaro.org> <87wpbi4ck0.fsf@totoro.br.ibm.com> <b22f1e69-55c3-88ae-a057-3983a9e89d52@linaro.org> <1493403960.9395.3.camel@oc7878010663>
On 28/04/2017 15:26, Steven Munroe wrote:
> On Tue, 2017-03-21 at 17:18 -0300, Adhemerval Zanella wrote:
>>
>> On 21/03/2017 17:09, Tulio Magno Quites Machado Filho wrote:
>>> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes:
>>>
>>>> On 20/03/2017 18:02, Tulio Magno Quites Machado Filho wrote:
>>>>> 1. Fix the results for negative subnormals by ignoring the signal when
>>>>> normalizing the value.
>>>>> 2. Fix the output when the high part is a power of 2 and the low part
>>>>> is a nonzero number with opposite sign. This fix is based on commit
>>>>> 380bd0fd2418f8988217de950f8b8ff18af0cb2b.
>>>>
>>>> It is not a blocker, but based on the complexity of the change and the
>>>> optimization real case usage, I wonder if it would be better to just remove
>>>> it and use the default implementation instead.
>>>
>>> I tried to do that before fixing these, but the performance difference is
>>> still around 2x.
>>
>> I would expect it, but the question is do we really care about the performance
>> of logbl-ibm128 to maintain a separated implementation for power7+ only?
>>
> Yes we still care. The power7 optimization avoids LHS by performing the
> mask and convert in VSX. and applies to power8, power9 for IBM long
> double..
Yeah I am aware of it, I was the one that code it ;)
>
> It will be some time before _float128 is fully deployed and until then
> customers will notice, if this IBM long double optimization went
> missing.
>
It is not a blocker, just a realization if it worth the maintenance of
keep a separate implementation for powerpc for this specific symbol.
I usually prefer to simplify the code and try to use the default
implementation where possible, however if the powerpc arch maintainer
sees that it is still worth the trouble I am ok with it.