This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 01/05] Math optimizations for POWER8
- From: Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>
- To: libc-alpha at sourceware dot org
- Date: Thu, 27 Feb 2014 16:06:00 -0300
- Subject: Re: [PATCH 01/05] Math optimizations for POWER8
- Authentication-results: sourceware.org; auth=none
- References: <5303722A dot 2090104 at linux dot vnet dot ibm dot com> <20140219093916 dot GU22181 at 0c3 dot net> <53049C10 dot 9050703 at linux dot vnet dot ibm dot com> <530F57E6 dot 7010606 at linux dot vnet dot ibm dot com>
On 27-02-2014 12:21, Adhemerval Zanella wrote:
> On 19-02-2014 08:57, Adhemerval Zanella wrote:
>> On 19-02-2014 06:39, Adam Conrad wrote:
>>> On Tue, Feb 18, 2014 at 11:46:02AM -0300, Adhemerval Zanella wrote:
>>>> Tested on powerpc64 and powerpc64le.
>>> I assume this means you've tested ppc64/ppc64le on POWER8 to make sure
>>> the new bits behave as you expect. Have you also tested on something
>>> oldish (POWER4/5, 970MP, etc) to make sure you got all the IFUNC bits
>>> right and didn't accidentally regress any other platforms?
>>>
>>> ... Adam
>>>
>> Fair enough, I have tested on a POWER5 and test-double/float/ldouble shows
>> the same output (double/float have some ULPs from newer tests that I haven, t
>> updated it) of POWER8 box.
>>
> If no one opposes, I intend to commit it in a couple of hours with the mentioned fixes
> from the reviews.
>
I pushed a slight modified path to address a issue when building with older binutils that
does not support POWER8 instruction.