search

SPECFEM / specfem2d

SPECFEM2D simulates forward and adjoint seismic wave propagation in two-dimensional acoustic, (an)elastic, poroelastic or coupled acoustic-(an)elastic-poroelastic media, with Convolution PML absorbing conditions.
GNU General Public License v3.0
194 stars 146 forks source link

specfem2d nightly build has failed #353

Closed buildbot-princeton closed 9 years ago

buildbot-princeton commented 9 years ago

See build 88.

komatits commented 9 years ago

I will have a look on Friday when I come back.

Dimitri.

On 05/20/2015 04:00 AM, buildbot-princeton wrote:

See build 88 http://specfem2d-buildbot.princeton.edu/builders/specfem2d_nightly/builds/88.

— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem2d/issues/353.

Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr

luet commented 9 years ago

It was a problem on my side. The reference seismograms were generated with a code compiled with a different compiler than the one used by buildbot for the nightly build. I will check but I think the problem came from using the Intel 13.0 compilers for the reference seismograms.

This brings an important issue. We should compare results from codes compiled with different compilers.

komatits commented 9 years ago

Hi David,

Thanks! That is weird though, because testing the correlation coefficient for being greater than 0.99 should be insensitive to very small noise / roundoff differences and thus to the compiler used (i.e., "visually" the seismograms are unchanged, and that is what we want to check).

Thus the test is either too sensitive (what threshold value do you use? 0.99? just to make sure you do not use 0.9999999 or so), or there is another problem elsewhere.

If you send me your test script I can have a look on Friday.

Thanks, Dimitri.

On 05/20/2015 05:05 PM, David Luet wrote:

It was a problem on my side. The reference seismograms were generated with a code compiled with a different compiler than the one used by buildbot for the nightly build. I will check but I think the problem came from using the Intel 13.0 compilers for the reference seismograms.

This brings an important issue. We should compare results from codes compiled with different compilers.

— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem2d/issues/353#issuecomment-103920171.

Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr