Mercurial > hg > octave-max
changeset 11596:a9cf422ed849
avoid apparent bug in ATLAS versions of CHERK/ZHERK
| author | John W. Eaton <jwe@octave.org> |
|---|---|
| date | Fri, 21 Jan 2011 11:05:21 -0500 |
| parents | 5ec6aa05638d |
| children | a066673566da |
| files | liboctave/CMatrix.cc liboctave/ChangeLog liboctave/fCMatrix.cc |
| diffstat | 3 files changed, 17 insertions(+), 2 deletions(-) [+] |
line wrap: on
line diff
--- a/liboctave/CMatrix.cc +++ b/liboctave/CMatrix.cc @@ -3754,7 +3754,12 @@ { octave_idx_type lda = a.rows (); - retval = ComplexMatrix (a_nr, b_nc); + // FIXME -- looking at the reference BLAS, it appears that it + // should not be necessary to initialize the output matrix if + // BETA is 0 in the call to ZHERK, but ATLAS appears to + // use the result matrix before zeroing the elements. + + retval = ComplexMatrix (a_nr, b_nc, 0.0); Complex *c = retval.fortran_vec (); const char ctra = get_blas_trans_arg (tra, cja);
--- a/liboctave/ChangeLog +++ b/liboctave/ChangeLog @@ -1,3 +1,8 @@ +2011-01-21 John W. Eaton <jwe@octave.org> + + * CMatrix.cc (xgemm): Initialize output matrix for call to ZHERK. + * fCMatrix.cc (xgemm): Initialize output matrix for call to CHERK. + 2011-01-20 John W. Eaton <jwe@octave.org> * Array-b.cc, Array-d.cc, Array-f.cc, Array-util.cc,
--- a/liboctave/fCMatrix.cc +++ b/liboctave/fCMatrix.cc @@ -3750,7 +3750,12 @@ { octave_idx_type lda = a.rows (); - retval = FloatComplexMatrix (a_nr, b_nc); + // FIXME -- looking at the reference BLAS, it appears that it + // should not be necessary to initialize the output matrix if + // BETA is 0 in the call to CHERK, but ATLAS appears to + // use the result matrix before zeroing the elements. + + retval = FloatComplexMatrix (a_nr, b_nc, 0.0); FloatComplex *c = retval.fortran_vec (); const char ctra = get_blas_trans_arg (tra, cja);