Re: [eigen] problem with MKL bindings for dense Eigen

[ Thread Index | Date Index | More Archives ]

It's not clear to me what you call "mess". Do you get completely different eigenvalues for the same input matrix? Or do your iterations slightly diverge before exploding?


On Fri, Apr 6, 2018 at 11:15 PM, Peter <list@xxxxxxxxxxxxxxxxx> wrote:
Dear Gael,

Am 06.04.2018 um 22:49 schrieb Gael Guennebaud:
> Just a guess, but maybe it's just that MKL gives you different values for the lowest eigenvalues (e.g., 1e-16 vs 1e-30 vs exact 0 while the largest is 1) but your code is lacking some thresholds to properly deal with them (e.g.,
> to ignore them). So tracking the largest and the smallest eigenvalues might a be a good start.

It's all about tracking the lowest eigen values and that's done properly, at least
I intended to do it properly.
That the MKL BLAS version is giving slightly different results is fine.
E.g.  if FMA are handled differently it will already influence the result.

My concern is about the mess I get from the MKL diagonalization.
I use MKL a lot in my main code (that one is still without Eigen) and there
everything is fine.

Best regards,

Mail converted by MHonArc 2.6.19+