Currently failing tests (Unit Test #469)


Added by Christopher Pinke over 6 years ago. Updated about 5 years ago.


Status:In Progress Start date:28 Apr 2013
Priority:Normal Due date:
Assignee:Alessandro Sciarra % Done:

0%

Category:-
Target version:-

Description

Not all implemented tests pass.
For example, the REC12 tests fail on GPUs.

This issue may be used to collect these tests.


History

Updated by Christopher Pinke over 5 years ago

Today I ran the tests on loewe (login-node) and some failed:
(these include only CPU tests, all GPU tests failed because there is no GPU on the login nodes)

pbp staggered (volume one)
sources_stagg (gaussian)
all RHMC

profiling data test
Inverter 31, 33, 35, 37
Heatbath 5
HMC 2
geometry

  • Status changed from New to In Progress

Updated by Christopher Pinke over 5 years ago

Alessandro, can you please check on the staggered tests above?
I checked that they are properly executed already, it seems that the results are just too far off, perhaps related to rnd numbers.

  • Assignee changed from Christopher Pinke to Alessandro Sciarra

Updated by Alessandro Sciarra about 5 years ago

The tests on the staggered chiral condensate on LOEWE (login node) are now all passing (a66198d6bbfc5378dae9126dce945d2067f4b4b4). For some reason, a test whose result should be 0 gives something.e-17. This is ok, but BOOST_CHECK_CLOSE failed because the reference value was zero. Now BOOST_CHECK_SMALL(..., 1.e-15) is used.

Updated by Alessandro Sciarra about 5 years ago

The tests on the staggered sources on LOEWE (login node) are now all passing (04e57f858d44a2bb198dc630d84c9d5db8cda567). I put a too small tolerance in the gaussian test, whose result cannot be predicted analitically. Now the squarenorm is compared with that obtained from the field drawn on gpu-dev03 (used as reference number) up to 1% of tolerance.

Updated by Alessandro Sciarra about 5 years ago

The tests on the RHMC staggered algorithm on LOEWE (login node) are now all passing (6aa8161729b972bfa061480bbdcdc8b7ff7c3d33). Again the tolerance was too small and it was increased to 0.5%. This is unavoidable, since changing the machine, the drawn random number are not guaranteed to be the same.

Also available in: Atom PDF