Hi!

* Sébastien Fabbro [2007-08-18 20:42]:
>> [ stuff about blas-atlas not linking ]
> should be fixed now.

Yes. Now it emerges fine...

>> [ stuff about mkl not extracting ]
> should be fixed too.

Same here. Emerges fine as well.

BTW, I have two more issues related to MKL and ACML. The ACML package
from official portage tree installs it's header (acml.h) via a symlink
into /usr/include. I remember that Donnie did this due to my report
related to the IT++ library I maintain (sci-libs/itpp). This would be
possible for acml.h (only one header file), but not for MKL headers
(there are many of them).

The problem is that now the IT++ ebuild can not automatically configure
the use of FFT native interfaces from MKL and ACML, when they are used
as BLAS and LAPACK external libs. It should work as follows: If one have
ACML or MKL installed, there is no need to install FFTW additionally. To
configure these specific FFT interfaces, the existence of <mkl_dfti.h>
and <acml.h> headers is verified, but of course it does not work,
because these headers are installed in vendor specific directories.

I discussed this issue with Markus some time ago and I proposed to
append unconditionally "-I/path/to/acml.h -I/path/to/mkl_dfti.h" to
CPPFLAGS and pass it to the configure command. But this solution is a
bit "dirty". Now I wonder how to make this work properly in the IT++
ebuilds. Maybe you have some ideas for this?

>> PS. I also noticed that blas-goto does not work properly on my core2duo
>> laptop (a bunch of BLAS-related tests in the IT++ library fail, whereas
>> ACML, MKL, ATLAS and NetLib's BLAS works fine). But I guess this is an
>> upstream issue...
> 
> blas-goto builds and tests fine on all my boxes. Send me an off-list
> email or file a bug to see what we can do.

OK. Will do it.

BR,
/Adam

-- 
.:.  Adam Piatyszek - "ediap"       .:.  JID: ediap(at)jabber.org .:.
.:.  ediap(at)users.sourceforge.net .:.  PGP key ID: 0x1F115CCB   .:.

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to