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: Failure in math/test-float-vlen8-avx2-log.out


On 05/19/2018 04:04 PM, Paul Pluzhnikov wrote:
> Greetings,
> 
> I am seeing failure in math/test-float-vlen8-avx2-log.
> Is that a real failure, or should ulps be updated?
> 
> Thanks,
> 
> $ cat math/test-float-vlen8-avx2-log.out
> testing float (vector length 8)
> Failure: Test: log_vlen8_avx2 (0xb.0d5dfp-4)
> Result:
>   is:         -3.69957983e-01  -0x1.7ad644p-2
>   should be:  -3.69957894e-01  -0x1.7ad63ep-2
>   difference:  8.94069672e-08   0x1.800000p-24
>   ulp       :  3.0000
>   max.ulp   :  2.0000
> Maximal error of `log_vlen8_avx2'
>   is      : 3 ulp
>   accepted: 2 ulp
> 
> Test suite completed:
>    36 test cases plus 0 tests for exception flags and
>      0 tests for errno executed.
>    2 errors occurred.
> 
> 
> $ cat /proc/cpuinfo
> processor : 0
> vendor_id : AuthenticAMD

I haven't seen this failure, but I also haven't tried
the CPU you're using.

Given that this is an AMD CPU, a difference of 1 ULP
from 2 to 3 for log is OK. Real errors generally show
up as thousands of ULPs error.

I would submit an ULPs update patch for review and state
the hardware you used.

-- 
Cheers,
Carlos.


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