[Embench] Issues with cubic benchmark

Jeremy Bennett jeremy.bennett at embecosm.com
Sun Nov 3 16:58:36 CET 2019


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 22/10/19 01:19, Anton Blanchard wrote:
> Hi,
> 
>> Cubic uses long double which is not standardized. On x86-64 it 
>> provides only 80 bits of precision, but on other architectures
>> it provides either 64 bits or 128 bits of precision.
>> 
>> This not only means that we aren't computing the same results,
>> but we are also unfairly penalising architectures that use a 128
>> bit long double. Those architectures often have a software
>> emulation library for 128 bit floating point arithmetic. It was
>> also pointed out to me that the benchmark isn't consistently
>> using long double - it's using double versions of the
>> transcendental functions.
>> 
>> I would suggest using double everywhere so that all architectures
>> are doing identical computations.
> 
> A potential fix:
> 
> https://github.com/embench/embench-iot/pull/18

Hi Anton,

Thanks for this and sorry for the slow reply while I've been away.

I've commented on this in the pull request and posted a separate email
to the mailing list, since I think there are arguments for and
against. We'll see if anyone has any comments to make. If not, I'll
merge your suggestion.

Thanks for contributing.

Best wishes,


Jeremy

> 
> Thanks, Anton
> 


- -- 
Tel: +44 (1590) 610184
Cell: +44 (7970) 676050
SkypeID: jeremybennett
Twitter: @jeremypbennett
Email: jeremy.bennett at embecosm.com
Web: www.embecosm.com
PGP key: 1024D/BEF58172FB4754E1 2009-03-20
-----BEGIN PGP SIGNATURE-----

iEYEARECAAYFAl2++SsACgkQvvWBcvtHVOH7RgCfQPQJAyreEV6wQ1svQwlWY+0g
G2kAmgIScAZxDKvc9BY0QWcTN7/r/KMt
=Fx+w
-----END PGP SIGNATURE-----



More information about the Embench mailing list