Compatibility with PGI Fortran

[eos_pengwern]

In the article describing use of the data fitting component, it mentions that MKL 10.3 update 8 does not support the PGI Fortran compiler. However, I can't find this mentioned anywhere in the release notes, and normally the stand-alone MKL does support the PGI compiler.

Does the restriction only apply to the data fitting component? Or is it the intention to drop support for the PGI (and/or other third-party Fortran compilers)?

I normally use both the Intel and PGI compilers for different parts of my project, and it will be very inconvenient to have to use two different versions of the MKL.

Thanks,
Stephen.

[TimP]

In general, ifort and pgf90 won't be compatible, because of conflicts in run-time libraries. It's possible, if you avoid using the threaded run-time in both together, and also restrict the use of the compiler's other run-time functions to one or the other, it may work. Many of the MKL functions, with the sequential library, may avoid using any ifort run-time functions which would conflict with PGI run-time. When you say "the" PGI Fortran, I suppose you mean one particular version, among Windows, linux, 32-bit, 64-bit, as well as among the various release levels.
You will note that although MKL for linux has specific support for gnu compilers, this doesn't support objects from ifort and gfortran linking together, nor the simultaneous use of libgomp and libiomp5.

[Gennady_F_Intel · Accepted solution]

Stephen,

Thanks for pointing this inconsistency into Release Notes.

Yes, this restriction applies only data fitting component and this is not intention to drop supporting PGI Compiler.

--Gennady

[Andrey_N_Intel]

Hi Stephen,
As Gennady mentioned, this limitation is applied to Data FittingComponent of Intel Math Kernel Library 10.3.8 only. We work toeliminate the limitation in future versions of the library.
Thanks,
Andrey

[eos_pengwern]

Thank you Tim; I understand these restrictions, and don't expect code compiled on the Intel and PGI compilers to play nicely together within a single application. My project consists of a suite of applications working together in concert, and I find that different compilers provide optimal performance in different parts of the suite. It's handy to be able to link both to the same MKL installation, though.

[eos_pengwern]

Thank you, Gennady and Andrey. As it happens I don't need to use the Data Fitting code, but I was reading about it out of curiosity when I noticed the comment.

I'll go ahead and install v10.3.8, which looks like it will be fine for what I'm doing.

Warm regards,
Stephen.