Minor CBLAS API changes
Greetings! In putting together a new atlas package, a few errors in
the existing upstream CBLAS reference implementation have been
discovered. These basically are:
1) cblas_?iamax routines should return a value starting from 0, not
from 1 as in the Fortran case. The existing implementation starts
from 1.
2) cblas switches certain error codes in the case of row major
matrices.
3) cblas exports a symbol RowMajorStrg used by the existing reference
testers, but which should not be there.
I have two choices:
1) release blas/atlas packages with a new soname (3). Advantages --
nothing will silently break, coexistence with existing stable
soname (2) packages. Disadvantages -- everything depending on blas
must be recompiled to use the new stuff, atlas is *huge* and it
will be in the archives twice.
2) release the new stuff with soname (2) and a big debconf warning.
Advice/comments appreciated!
Take care,
--
Camm Maguire camm@enhanced.com
==========================================================================
"The earth is but one country, and mankind its citizens." -- Baha'u'llah
Reply to: