Hi, please build a 64-bit flavor of lapack as well, say lapack64. I'll need the _pic static library to build a 64-bit version of OpenBLAS when the review is approved.
.. and what this mean is that you should add -fdefault-integer-8 to the fortran flags.
It seems like all this needs is for -fdefault-integer-8 to be added when compiling for x86_64. Is there any benefit to the FORTRAN INTEGER being 32bit on x86_64?
Arrays can be bigger since there's no 2^32 ~ 4.29 billion element size limit. See comment #21 in bug #739398.
No, I know the PROs of having 64bit FORTRAN INTEGER. What I'm asking is, on x86_64, are there any CONs to having 64bit FORTRAN INTEGER?
Oh, whoops! :D Well, legacy code might break. There also might be some speed and memory issues. I'd be quite wary of changing the way things work by default. Fortran integers are 4 byte by default, so ...
For example, if C code calls a routine with integer arguments, things will go haywire straight away.
Okay, good enough for me. I'll make a lapack64 variant in rawhide.
As of 3.4.1-3.fc19, there are the following new subpackages (on x86_64 only): lapack64 lapack64-devel lapack64-static blas64 blas64-devel blas64-static Lemme know if these do not work for you.
Provides: blas64 = 3.4.1-3.fc19 blas64(x86-64) = 3.4.1-3.fc19 libblas.so.3()(64bit) This doesn't look right. You should change the sonames as well. Otherwise looks good.
Hmm, okay. Hacking the sonames will be trickier, but I'm on it.
3.4.1-4.fc19 will have the fixed sonames. http://koji.fedoraproject.org/koji/taskinfo?taskID=4847088