Re: [Bug-gsl] GSL 2.3 built with Intel 17.0.0 does not pass linalg test suite

[Patrick Alken]

Hi,

  Thanks for the report. Most developers are using gcc so we set the
test tolerances accordingly and sometimes other compilers don't compute
things as accurately, etc. But you can see from the output that its not
a big problem. The QR code is computing the right value to about 12
decimal places which is pretty good. I think that particular test
tolerance is set to around 13 decimal places. I think I'll leave the
tolerance unchanged since its been set to that value for many years and
has worked on many systems and compilers. Maybe this issue is only with
the intel 17 compiler. I've successfully passed make check using older
intel compilers on my systems.

Patrick

On 12/21/2016 12:15 PM, Stewart, Adam James wrote:
> I was just able to successfully pass `make check` with GCC 6.1.0, so I think 
> the problem is either transient or restricted to the Intel build. Note that I 
> was using `make -j16 check` for both tests.
>
> Adam J. Stewart
> Assistant Systems Administrator
> Laboratory Computing Resource Center
> Computing, Environment and Life Sciences
> Argonne National Laboratory
> ________________________________
> From: Stewart, Adam James
> Sent: Monday, December 19, 2016 10:32 AM
> To: address@hidden
> Subject: GSL 2.3 built with Intel 17.0.0 does not pass linalg test suite
>
> Hi,
>
> I'm trying to run `make check` on GSL 2.3 built with Intel 17.0.0  on CentOS 
> 6.7 and am seeing the following test-suite.log:
>
> ====================================
>    gsl 2.3: linalg/test-suite.log
> ====================================
>
> # TOTAL: 1
> # PASS:  0
> # SKIP:  0
> # XFAIL: 0
> # FAIL:  1
> # XPASS: 0
> # ERROR: 0
>
> .. contents:: :depth: 2
>
> FAIL: test
> ==========
>
>   4[0]:   -63.9999999999623128                      -64
> FAIL:   QR_solve hilbert(4) [37853]
> FAIL: QR Solve [37859]
>   4[0]:   -63.9999999999622204                      -64
> FAIL:   QR_QRsolve hilbert(4) [37943]
> FAIL: QR QR Solve [37949]
> (  4,  4)[0]:   -63.9999999999623128                      -64
> FAIL:   QR_lssolve hilbert(4) [37953]
> FAIL: QR LS Solve [37959]
> (caught [svd.c:148: SVD decomposition failed to converge (11)])
> returned error code 11 = exceeded max number of iterations
>
> ...
>
> FAIL:   SV_decomp bigsparse [38076]
> FAIL: Singular Value Decomposition [104750]
>   4[0]:   -63.9999999999570335                      -64
>   4[1]:    899.999999999545594                      900
> FAIL:   SV_solve hilbert(4) [271653]
> FAIL: SVD Solve [271659]
>
>
> Adam J. Stewart
> Assistant Systems Administrator
> Laboratory Computing Resource Center
> Computing, Environment and Life Sciences
> Argonne National Laboratory