This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] x86-64: Add sinf with FMA


On Fri, Dec 8, 2017 at 8:07 AM, Arjan van de Ven <arjan@linux.intel.com> wrote:
> On 12/8/2017 8:02 AM, Nick Alcock wrote:
>>
>> On 4 Dec 2017, H. J. Lu uttered the following:
>>
>>> On Skylake, bench-sinf reports performance improvement:
>>>
>>>              Before        After         Improvement
>>> max        153.996       100.094           54%
>>> min        8.546         6.852             25%
>>> mean       18.1223       14.4616           25%
>>>
>>> Any comments?
>>
>>
>> Do we have any benchmark runs on older processors? They're not remotely
>> obsolete: Intel is still selling Broadwell server parts, and the vast
>> majority of SSE2-capable parts out there at present are not as new as
>> Skylake.
>
>
> (oh and Haswell & Broadwell also support FMA already)

On Haswell, without FMA:

  "sinf": {
   "": {
    "duration": 3.4905e+10,
    "iterations": 1.91281e+09,
    "max": 450.098,
    "min": 8.091,
    "mean": 18.2481
   }
  }

With FMA:

  "sinf": {
   "": {
    "duration": 3.49046e+10,
    "iterations": 2.44188e+09,
    "max": 173.855,
    "min": 7.253,
    "mean": 14.2942
   }
  }

FMA is faster on HSW.  I expect Broadwell should also be faster with FMA.


-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]