Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Abo__Gavin
Beginner
69 Views

mkl ILAENV question

I'm using Update 11 of composerxe 2011, but the following may occur in earlier and later verisons as well.  When I run the attached test case (tilaenv.f) for intel64, ILAENV returns 128.  However, when I run the following case for ia32, ILAENV returns 640.  Is 640 the correct return value?  It just seems large compared to the return value for intel64, so I'm wondering if iILAENV should be returning 64.  If so, how can I fix it?  I would appreciate any feedback.

0 Kudos
6 Replies
mecej4
Black Belt
69 Views

I obtain 64 as the return value with MKL 10.3.11 Product Build 20120606 (IA32) and compiler 12.1.5.344 Build 2012-0612 .
TimP
Black Belt
69 Views

With the Intel64 13.0 update 1 or 12.1 update 13 packages for Windows, I get 640 from MKL ilaenv. With netlib ilaenv, I get 64, so it seems MKL doesn't work the same as lapack 3.2.1 reference source. The source code appears to date from over 30 years ago.
Gennady_F_Intel
Moderator
69 Views

the same with me - ia32 version returned - 64, but intel64 - 640. windows, MKL v.11.0 we will check what' going with this code.
Abo__Gavin
Beginner
69 Views

Ok and thanks. By the way, I'm using the linux package.
mecej4
Black Belt
69 Views

After thinking a bit more about this issue, I begin to ask why the returned value, being a tuning parameter (= optimum block size), should be the same between the MKL and ref. Lapack implementations. If the MKL optimum value is related to optimized assembler code, there is nothing wrong with such a discrepancy. Furthermore, there is nothing wrong with the value being different for different machine architectures (IA32 and X64).
Gennady_F_Intel
Moderator
69 Views

Hello,

This issue has been fixed in MKL v.11.0 update 2 released yesterday.

You can download this update from intel registration center and check the problem on your side.

--Gennady

Reply