topic Quad Precision with MKL? in Intel® oneAPI Math Kernel Library

Quad Precision with MKL?

syeganeh — Thu, 16 Jul 2009 00:29:47 GMT

Hi,

For starters, I'm running ifort version 11.0 with MKL 10.1.1.019 (on Linux).
I'm currently using the MKL subroutines gesvd and gesv (fortran 95 wrappers), and everything is working fine with double precision inputs. I was wondering if there was anyway to do quad precision with these subroutines? I tried just passing a quad variable (as defined with selected_real_kind) and it didn't work:
"There is no matching specific subroutine for this generic subroutine call."

Thanks in advance for your help.

Re: Quad Precision with MKL?

TimP — Thu, 16 Jul 2009 00:57:14 GMT

You would have to compile the public source code (netlib.org) to make the quad precision versions of the functions you want. You would be giving up on the "performance" aspect of MKL, as there is no support for vectorization in quad precision.

Re: Quad Precision with MKL?

syeganeh — Thu, 16 Jul 2009 13:53:01 GMT

Quoting - tim18

Thanks for the reply. Do you you know if there's already a generic interface that provides quad precision or would I have to edit the code for the function I want and change the type of the variables?

Re: Quad Precision with MKL?

TimP — Thu, 16 Jul 2009 14:32:48 GMT

Quoting - syeganeh

Thanks for the reply. Do you you know if there's already a generic interface that provides quad precision or would I have to edit the code for the function I want and change the type of the variables?

I suppose you would have to add your quad precision to the generic interface and recompile it. Probably, as you say, this would require explicit types with appropriate KIND rather than use of -r16.