Re: [eigen] Altivec port status

[ Thread Index | Date Index | More Archives ]

On Tue, 15 Jul 2014 10:03:12 +0200
Gael Guennebaud <gael.guennebaud@xxxxxxxxx> wrote:

> Hi Konstantinos,

Hi Gael,

> Thank you for looking at it.
> I've tested the Altivec support a few months ago using the GCC's
> compile farm after we introduced several new vectorization functions.
> So the situation should not be too bad.

No, it's actually much better than I believed after reading some mails
on this list, thanks for keeping the port up to speed! :)

> That's probably a shortcoming in vectorization_logic itself. Perhaps
> that one is already fixed by the very recent Christoph's change?

Yes, as Christoph just said, this was an easy one to fix, just
HasDiv=1 and it worked. the second one is a bit trickier.
> What's the second one?

$ ./product_trsolve_8 
Initializing random number generator with seed 1405413527
Repeating each test 10 times
Test (trsolve<std::complex<float>,4,1>()) failed
in /home/markos/eigen/test/product_trsolve.cpp (46) test_isApprox
((cmLhs.conjugate().template triangularView<Lower>()).toDenseMatrix() *
(cmRhs), ref) Stack:
  - (trsolve<std::complex<float>,4,1>())
  - product_trsolve
  - Seed: 1405413527


I'm trying to follow the code paths that lead to this failure, but I
have to say it's not that easy :)



Attachment: pgpCZRLO1fQXN.pgp
Description: PGP signature

Mail converted by MHonArc 2.6.19+