Re: intel mkl

[javaloye]

Dear Jaroslav,

thanks for the comments. I followed your advice and compiled 3.3.52.
Using icc + icpc + ifort + mkl is still not working for me.

However I finally got a functional system by using gcc + g++ + ifort + mkl.
I had to compile qrupdate with ifort otherwise test qr was failing.
I also had to remove support for arpack, that is still giving me some
SIGSEGV during make check.

Other than that, all test were successful (excepted the few that are known
to fail).
Everything is functional and I can even zoom with the fltk terminal, sweet !
The linear algebra part is working at the speed of mkl, that is to say,
fast. 

For the sake of curiosity, you mentioned that you compiled octave with intel
compiler and mkl without difficulty, would you be so kind as to post your
configure line ?

I am insisting on using icc (with flags as -xT -O2 or -fast) since I am
using a home made oct file that need top speed. 
1) It is already quite fast using a mkoctfile linking to g++.
2)It is already using nicely openmp.
However, I was curious to know it I can still squeeze out some juice.
( Notice that I CAN use icc to compile and then link with mkoctfile, but I
cannot use the flag -xT.)
My hope was that if I can successfully compile octave with icc -xT,
mkoctfile would automatically use it.

regards.

Julien.
-- 
View this message in context: 
http://octave.1599824.n4.nabble.com/intel-mkl-tp2246821p2314623.html
Sent from the Octave - General mailing list archive at Nabble.com.