Hi,
I use the routine
PDL::LinearAlgebra::Complex::cgtsv
to solve systems of linear equations described by a tridiagonal
matrix. The arguments are the sub-diagonal, the diagonal, the
supradiagonal and the right hand side of the equations. I found one of
my programs has a misterious error, as it uses cgtsv to solve a
symmetric tridiagonal system,
my $sub_diagonal=my $supra_diagonal;
cgstv($sub_diagonal, $diagonal, $supra_diagonal, $right_hand_side);
I didn't realize until today that cgstv modified the contents of the
$sub_diagonal array, and thus, it silently and erroneously modified the
$supra_diagonal, which in my case was an ndarray that shared the
numerical data.
The signature of the function is:
cgtsv
Signature: (complex [phys]DL(n);complex [phys]D(n);complex
[phys]DU(n);complex [io,phys]B(n,nrhs); int [o,phys]info())
As the parameter 'DL' is not marked 'io' I erroneously assumed it wouldn't be
modified, although the documentations states further down that it
does.
My questions are: Is the signature above correct? Should parameters
that are modified be marked 'io' or something else? Or would it be better
for routines such as cgtsv to make a copy of their inputs to avoid
modifying the original arguments? Would it be considered too costly in
general?
Best regards,
Luis
--
o
W. Luis Mochán, | tel:(52)(777)329-1734 /<(*)
Instituto de Ciencias Físicas, UNAM | fax:(52)(777)317-5388 `>/ /\
Av. Universidad s/n CP 62210 | (*)/\/ \
Cuernavaca, Morelos, México | [email protected] /\_/\__/
GPG: 791EB9EB, C949 3F81 6D9B 1191 9A16 C2DF 5F0A C52B 791E B9EB
_______________________________________________
pdl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pdl-devel
