[Nek5000-users] libxsmm within Nek5000

[nek5000-users at lists.mcs.anl.gov]

At the moment there are no tuned "high-level "operators like axhelm(). Also my experience is that libxsmm is faster for lx1>14 which is rarely used for production runs.

Cheers,
Stefan

-----Original message-----
> From:nek5000-users at lists.mcs.anl.gov <nek5000-users at lists.mcs.anl.gov>
> Sent: Monday 20th November 2017 13:49
> To: nek5000-users at lists.mcs.anl.gov
> Subject: [Nek5000-users] libxsmm within Nek5000
> 
> Hi,
> 
> Do you have any performance results of the libxsmm comparing with stand-alone "mxm" ?
> 
> In the master branch of Nek5000, the libxsmm function "libxsmm_dgemm" is only called in file "
> mxm_wrapper.f"
> ...
> #ifdef XSMM
>          call libxsmm_dgemm('N','N',n1,n3,n2,1.0,a,n1,b,n2,0.0,c,n1)
> #endif
> ...
> 
> I could not find any optimization works such as dispatch and streaming updates in the kernels that were used in e.g. NekBox.
> 
> What is the status about the implementation of libxsmm or other SIMD techniques within Nek5000 right now?
> 
> Thanks. /Jing 
> _______________________________________________
> Nek5000-users mailing list
> Nek5000-users at lists.mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users