I'd like to suggest placing a README in the directory where xblat2s & company
are located explaining what the output is supposed to be. It would also be
nice if there were some mention in the top level directory of the divide by
zero issue (e.g. slatm7.f line 188). I'm assuming that LAPACK purports to
handle divide by zero internally. While there are sound reasons for doing so,
I really like to trap it at the FPU level.
At the least, please include a note in the Makefile indicating that xblat2s &
company are testing the argument checking in the BLAS routines.
I just spent an entire day sorting out what was going on :-( In the end, I had
to read xblat2s.f and the BLAS documents to understand if I actually had a
problem. Neither LAWN41 nor google were any help.
FWIW I'm running Solaris 10 U8 x86 w/ the Studio 12.2 compiler suite. I'm
trying to build a recent version of Octave. The dependency chain is large
enough that it will probably have consumed 4 days by the time I'm done. Just
building gcc took a full day because of minor bugs (e.g. libfoo.so.1 vs
libfoo.so.1.0.0) and an incompletely documented dependency chain. I very much
prefer to build important codes from source because it gives me control of
optimization and allows me to build debug versions if I need them.