GROMACS version: 2021.4
GROMACS modification: No
Dear all:
In my hands, regressiontests always fails for GROMACS 2021.4 (when compiled in double precision) on all Linux systems tested (Debian 10, Debian 11, CentOS 8 using various gcc versions).
Specifically, GammaDistributionTest (RandomUnitTests) fails (see below).
On all systems, regressionstests are successfully passed for GROMACS 2021.4 (compiled in single/mixed precision) and for all other GROMACS versions tested (for example, 2021.3 or 2020.6) in both mixed and double precision.
Does anyone know the implications of this failure and/or knows what can be done to make a double-precision compilation of 2021.4 pass the tests?
Best wishes,
Jan-Philipp
PS: Here the relevant terminal output:
32/71 Test #32: RandomUnitTests …***Failed 0.01 sec
[==========] Running 44 tests from 10 test cases.
[----------] Global test environment set-up.
[----------] 4 tests from ExponentialDistributionTest
[ RUN ] ExponentialDistributionTest.Output
[ OK ] ExponentialDistributionTest.Output (0 ms)
[ RUN ] ExponentialDistributionTest.Logical
[ OK ] ExponentialDistributionTest.Logical (0 ms)
[ RUN ] ExponentialDistributionTest.Reset
[ OK ] ExponentialDistributionTest.Reset (0 ms)
[ RUN ] ExponentialDistributionTest.AltParam
[ OK ] ExponentialDistributionTest.AltParam (0 ms)
[----------] 4 tests from ExponentialDistributionTest (0 ms total)
[----------] 4 tests from GammaDistributionTest
[ RUN ] GammaDistributionTest.Output
/tmp/inst/gromacs-2021.4/src/testutils/refdata.cpp:927: Failure
In item: /GammaDistribution/[28]
Actual: 24.49245196388771
Reference: 24.492451963887692
Difference: 1.77636e-14 (5 double-prec. ULPs, rel. 7.25e-16)
Tolerance: abs. 8.88178e-16, 4 ULPs
[ FAILED ] GammaDistributionTest.Output (0 ms)