[SciPy-user] import failure using g77

Jeffrey B. Layton laytonjb at bellsouth.net
Sun Oct 20 11:22:13 EDT 2002


Pearu Peterson wrote:

>>It appears to be there (I think).
>>    
>>
>
>No, it is not there (U means undefined, see 'man nm'). I have
>
>ATLAS-3.4.1/lib/Linux_PII$ nm liblapack.a | grep atl_f77wrap_sgetrf__
>         U atl_f77wrap_sgetrf__
>00000000 T atl_f77wrap_sgetrf__
>
>
>atl_f77wrap_sgetrf__ appears in ATL_f77wrap_sgetrf.o
>

OK. Here's the problem. If I look at the symbols in the original
liblapack.a that ATLAS creates for sgetrd, here's what I find:

ATL_sgetrf.o:
00000000 T ATL_sgetrf
         U ATL_sgetrfC
         U ATL_sgetrfR
ATL_sgetrfC.o:
00000000 T ATL_sgetrfC
ATL_sgetrfR.o:
00000000 T ATL_sgetrfR
         U ATL_sgetrf
clapack_sgetrf.o:
         U ATL_sgetrf
00000000 T clapack_sgetrf
         U ATL_sgetrf
ATL_f77wrap_sgetrf.o:
         U ATL_sgetrf
sgetrf.o:
         U atl_f77wrap_sgetrf__
00000000 T sgetrf_

So that means that atl_f77wrap_sgetrf is there, correct?
   Now, if I lok at liblapack.a after I combine LAPACK
and ATLAS for sgetrf, here's what I find:

         U sgetrf_
sgetrf.o:
         U atl_f77wrap_sgetrf__
00000000 T sgetrf_
         U ATL_sgetrf
ATL_f77wrap_sgetrf.o:
         U ATL_sgetrf
ATL_sgetrfC.o:
00000000 T ATL_sgetrfC
ATL_sgetrf.o:
00000000 T ATL_sgetrf
         U ATL_sgetrfC
         U ATL_sgetrfR
ATL_sgetrfR.o:
00000000 T ATL_sgetrfR
         U ATL_sgetrf
clapack_sgetrf.o:
         U ATL_sgetrf
00000000 T clapack_sgetrf

So, I'm assuming that in recombining the libs, atl_f77wrap_sgetrf__
did not make it. Is that correct? How could this happen? Could it
be from using different compile flags (I used some other flags in
compiling LAPACK, particularly "-march=athlon")? I'll try compiling
LAPACK using -really- generic flags and see what happens.

Thanks for any help on this!

Jeff "Closer to giving up" Layton

>
>Pearu
>
>_______________________________________________
>SciPy-user mailing list
>SciPy-user at scipy.net
>http://www.scipy.net/mailman/listinfo/scipy-user
>
>  
>






More information about the SciPy-User mailing list