And vzecca@Domain.Removed writes:
At zcgesv line 199 the routine zlange is called with sixth actual
argument (WORK) complex*16. However in zlange the corrssponding dummy
argument is double precision. This is against the Fortran 95
Actually, I think it is standard. I don't have a text handy, but I
believe sequence association combined with the number of numerical
units in COMPLEX does guarantee that the code is standard-correct
as it stands.
slaneg.f line 108 because DPLUS may be zero. This is producing
ssep.out, csep.out, and zsep.out.
dlaneg.f line 108 ditto. Producing dsep.out.
This is intentional and is handled correctly. See LAWN 172 for
too many details:
dlar1v.f line 199 ditto. Producing dsep.out.
Similar. This eigenvalue/vector algorithm is designed to use
IEEE-754 default handling of many exceptions. Updating xLAR1V and
crew to use the same code as xLANEG is on a too long to-do list
along with optimizing them for appropriate test block sizes, etc.
Contributions gleefully accepted!