test_omega95_bispinors fails with nagfor

[ohl]

_build_nag/tests/test_omega95_bispinors.log:

...
 veps(-2).ueps(-2)=   0: passed at  100%
 Runtime Error: *** Arithmetic exception: Floating invalid operation - aborting
omega_testtools.f90, line 253: Error occurred in OMEGA_TESTTOOLS:EXPECT_REAL
omega_testtools.f90, line 326: Called by OMEGA_TESTTOOLS:EXPECT_REAL_INTEGER
test_omega95_bispinors.f90, line 243: Called by TEST_OMEGA95_BISPINORS
*** Majorana properties of gravitino vertices: ***
...

This can also be an error with the testing routine EXPECT_REAL

[ohl]

Fixed the EXPECT* problem with NaNs? in r2258.

As of build 686 nagfor produce several NaNs? in test_omega95_bispinors. I'm not sure if this is an compiler issue.

[ohl]

Illustration (depends on the random number seed, but occurs fairly often):

 f_potgr   - gr_potf   = 0: failed at    0% [expected  0.000E+00, got        NaN]
 pot_fgr   - pot_grf   = 0: failed at    0% [expected  0.000E+00, got        NaN]
 f_s2gr    - gr_s2f    = 0: failed at    0% [expected  0.000E+00, got        NaN]
 s2_fgr    - s2_grf    = 0: failed at    0% [expected  0.000E+00, got        NaN]
...
 *** Testing the gravitino propagator: ***
 Transversality:
Runtime Error: *** Arithmetic exception: Floating overflow - aborting
omega_bispinor_couplings.f90, line 414: Error occurred in OMEGA_BISPINOR_COUPLINGS:F_VF
omega_bispinor_couplings.f90, line 2183: Called by OMEGA_BISPINOR_COUPLINGS:PR_GRAV
test_omega95_bispinors.f90, line 287: Called by TEST_OMEGA95_BISPINORS
Aborted

[Juergen Reuter]

For JRR the test ran through two times now without any problems (using NAG 5.2 (724) on 32 bit). Could this be a bug that has been erased meanwhile from the NAG compiler? Or it is 64bit specific.

[Juergen Reuter]

Passed all times now, WK reports passing for him on 64bit as well. Please recheck TO!

[Juergen Reuter]

Wow, with -O3 and NAG I get now also a runtime error:

=======================================
   omega 2.0.0: tests/test-suite.log   
=======================================

1 of 5 tests failed.  

.. contents:: :depth: 2


FAIL: test_omega95 (exit: 134)
==============================

Runtime Error: *** Arithmetic exception: Floating divide by zero - aborting
 *** Checking the equations of motion ***:
 |[p-m]u(+)|=0: passed at   86%
 |[p-m]u(-)

Needs really further investigation

[ohl]

nagfor build 724 with -O2 on 64bit works for me reliably. We should keep an eye on this, but I'm closing the ticket for now.

[ohl]

Just a quick followup: something's rotten with RANDOM_NUMBER in the current nagfor. It may happen that identical four vectors are generated and the tests implicitely assume that they are different.

[Juergen Reuter]

Milestone v2.0.0final deleted