by Julien Langou » Sun Oct 02, 2005 5:32 am
Hello,
yes you are correct.
Basically IZAMAX can fail its testing routine depending on the BLAS tested.
We already have spotted this but never take any action on it. The problem is more on the tester (or the definition of IZAMAX) than in the reference BLAS routine itself (or the compiler) .
The problems comes from the testing routines. It asked to IZAMAX which elements is the biggest in term of absolute value [abs(z) = |Re(z)| + |Im(z)|] amongst the two following double-complex-entry vectors
(0.3,0.1) (0.1,0.4) (0.4,0.1) (0.1,0.2) [first failure]
(0.3,0.1) (0.1,0.4) (0.4,0.1) (0.1,0.2) [second failure]
The expected answer by the tester is the second entry in both cases.
The computed answer (on GNU-Fortran-g77-compiler BLAS) is the third one.
So the tester indicates a failure.
This failure can be removed at the tester level by removing the test...
Or at the reference BLAS level by defining the function dcabs1 (basically do in IZAMAX as is done in ICAMAX )
I guess nothing dramatic for user application but we should fix this.
I'll contact the other developers and we take action on this and report.
Julien