gaussianNormalPair does not work with _use_same_rnd_numbers (Defect #472)


Added by Christopher Pinke almost 6 years ago. Updated almost 6 years ago.


Status:Feedback Start date:30 Apr 2013
Priority:Normal Due date:
Assignee:Matthias Bach % Done:

0%

Category:-
Target version:-

Description

In e7b7aafe I added test for volume sources with use_same_rnd_numbers.

However, I observed nans on CPU and GPU with this setting.

Different to #470, this was not caused by the looping over the sites, all sites are reached. It has to do with the gaussian normal pair.

In fact, the kernel works with a z4 sourcecontent.
Also if one excludes the for-loop by inserting

        hmc_complex tmp2 = gaussianNormalPair(&rnd);
        printf("%f %f\n", tmp2.re, tmp2.im);
        return;

before the loop one can see that the output of gaussianNormalPair is undefined.

However, the same test passes for the spinorfield, with and without eo (see cdd82e57 and 9abb6a0c).
The test for the volume source is a modified version of the spinorfield_gaussian test.


History

Updated by Christopher Pinke almost 6 years ago

  • Description set to In commit:e7b7aafe I added test for volume sources with _use_same_rnd_numbers... More

Updated by Christopher Pinke almost 6 years ago

  • Description changed from In commit:e7b7aafe I added test for volume sources with _use_same_rnd_numbers... to In commit:e7b7aafe I added test for volume sources with _use_same_rnd_numbers... More

Updated by Christopher Pinke almost 6 years ago

  • Description changed from In commit:e7b7aafe I added test for volume sources with _use_same_rnd_numbers... to In commit:e7b7aafe I added test for volume sources with _use_same_rnd_numbers... More

Updated by Christopher Pinke almost 6 years ago

I checked the compile parameters of the two kernels "spinorfield_volume_source" and "spinorfield_gaussian", they differ by "KAPPA=0.125 -D MKAPPA=-0.125 -D NUM_SOURCES=12 -D SOURCE_CONTENT=2". The first 3 are not needed in the kernel, but commenting them out does not change the result.

Updated by Christopher Pinke almost 6 years ago

Matthias, do you have any idea what might be the reason? I really don`t see much difference, neither in the kernels nor in the corresponding tests.

  • Status changed from New to Feedback
  • Assignee set to Matthias Bach

Also available in: Atom PDF