|
|
Ublas : |
Subject: Re: [ublas] LAPACK/BLAS bindings
From: Peter Man (plwm2_at_[hidden])
Date: 2010-06-07 04:43:37
Hi,
Thanks for the quick reply. So I managed to get the ublas_getri.cc
working, with the following information:
1) I'm using cygwin g++ version 4.3.4
2) I built atlas myself - all the library files appear to be correctly
generated.
3) I'm using cygwin version 1.7.x
4) Compilation commands used:
" g++ -Wall -O2 --pedantic -o test ublas_getri.cpp -I
/cygdrive/c/Program\ Files/boost/boost_1_42/ -I
/cygdrive/c/cygwin/usr/local/atlas/include/ -L
/cygdrive/c/cygwin/usr/local/atlas/lib/ -llapack -lf77blas -lcblas
-latlas -lgfortran -lm
If I then define BOOST_NUMERIC_BINDINGS_BLAS_CBLAS, still no problem,
and all is well.
However, if I then define BOOST_NUMERIC_BINDINGS_LAPACK_CLAPACK, then I
reproduce my runtime error.
I'm a bit of a moron with this sort of stuff, so I apologise if i'm
doing something really stupid.
And another problem...if I revert the file back to its original state
(without defining the two macros above), and I remove the
include statements for any lapack routines, but in their place i include
<boost/numeric/bindings/lapack.hpp>,
I get a massive number of errors, the first few lines being:
/cygdrive/c/Program
Files/boost/boost_1_42/boost/numeric/bindings/lapack/computational/ggqrf.hpp:169:32:
warning: '$' in identifie
r or number
In file included from /cygdrive/c/Program
Files/boost/boost_1_42/boost/numeric/bindings/lapack/computational.hpp:22,
from /cygdrive/c/Program
Files/boost/boost_1_42/boost/numeric/bindings/lapack.hpp:12,
from ublas_potri.cpp:24:
/cygdrive/c/Program
Files/boost/boost_1_42/boost/numeric/bindings/lapack/computational/ggqrf.hpp:234:
error: ‘$TYPES’ has not been d
eclared
/cygdrive/c/Program
Files/boost/boost_1_42/boost/numeric/bindings/lapack/computational/ggqrf.hpp:235:
error: ‘$ARGUMENTS’ was not de
clared in this scope
I would be extremely grateful for a swift response!!
Peter
On 06/06/2010 23:13, Thomas Klimpel wrote:
> Peter Man wrote:
>
>> However, what about my second question, about ublas_getri.cc producing
>> runtime errors?
>>
> The regression test for ublas_getri.cc works for me on cygwin. I tried guessing your configuration, but had no luck at reproducing the error. What version of ATLAS are you using? Did you build it yourself, or do you use the cygwin package? Are you using cygwin 1.5 or 1.7? Which gcc version are you using? What are your compile and link flags? Your runtime error is actually an assertion indicating that the workspace request on getri has returned a smaller workspace size than the minimal expected workspace size.
>
> Regards,
> Thomas
> _______________________________________________
> ublas mailing list
> ublas_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/ublas
> Sent to:plwm2_at_[hidden]
>