Re: [ublas] [BLAS bindings] help appreciated on using trmv and gbmv

Subject: Re: [ublas] [BLAS bindings] help appreciated on using trmv and gbmv
From: Florent Teichteil (florent.teichteil_at_[hidden])
Date: 2012-10-07 16:45:24

> Hello Florent,
>
> I'm not able to reproduce your column_major / row_major differences.
>
> Please find attached your first test case. I get identical results for both
> row_major and column_major. And, indeed, those results are also identical to
> a ublas::prod(). g++ 4.7.2, ATLAS 3.8.4.
>
> Am I missing something?
>
> Regards,
>
> Rutger
>

That's very strange.
Indeed, I get identical results with your tests.
But I don't with mines (attached to previous e-mail), except if I
choose dim=7 and if divide random generated numbers by 1e9, exactly
like in your tests.
So it seems that the observed results depend on the dimension of the
matrix and on the magnitude of numbers stored in the matrix? That's
strange, isn't it?

(anyway, other tests, for instance with transposed triangular matrices
or banded matrices, are still wrong)

Florent