[Q-e-developers] Compiling QE-GPU without openmp

Filippo Spiga filippo.spiga at quantum-espresso.org
Thu Jul 14 18:41:32 CEST 2016


Hello Pietro,

I will look into this hopefully tomorrow. Do you know which version of Intel compiler is used in the CRAY environment and do you know which additional flags (if any) are added by ftn?

Cheers

On 14 Jul 2016, at 17:13, Pietro Bonfa' <pietro.bonfa at fis.unipr.it> wrote:
> Dear All,
> 
> I hope this is the right place to report compilation errors for QE-GPU.
> If not, please tell me where should I post this (maybe an issue on github?).
> 
> I'm testing QE-GPU on the Piz Daint cluster of CSCS.
> 
> The compilation fails when I use the --disable-openmp option (see here:
> https://gist.github.com/bonfus/05d8a574c6a36fea7a02595fc0e9fd58#file-compile-sh-L47).
> 
> I get the following error message:
> 
> ---------------------------------------------
> 
> !$omp parallel default(shared), private(na,qgm_na,is,dtmp,ig,mytid,ntids)
>                                                             ^
> ftn-113 crayftn: ERROR NEWQ_COMPUTE_GPU, File = newq_compute_gpu.F90,
> Line = 133, Column = 62
>  IMPLICIT NONE is specified in the local scope, therefore an explicit
> type must be specified for data object "MYTID".
>                                                                   ^
> ftn-113 crayftn: ERROR NEWQ_COMPUTE_GPU, File = newq_compute_gpu.F90,
> Line = 133, Column = 68
>  IMPLICIT NONE is specified in the local scope, therefore an explicit
> type must be specified for data object "NTIDS".
> 
> 
> (Long report here http://pastie.org/10907277 )
> 
> ---------------------------------------------
> 
> I solved the problem by adding a couple of ifdef(__OPENMP) directives in
> newq_compute_gpu.f90 (line 133 and 185) but I don't know if this is
> actually a compiler problem.
> 
> 
> Thanks,
>    Pietro





More information about the developers mailing list