Redundant copies of scalars in OpenACC parallel directive

[fspiga]

https://github.com/QEF/qe_test_openacc/blob/phonon_gpu_develop/PHonon/PH/dvqpsi_us.f90#L141

In both OpenACC parallel and kernel directives it is not needed to do copyin/copyout/create (any data clause actually) for scalars. Scalars are passed to the GPu kernel generated by the compiler as parameters of the kernel function call. Forcing them with copyin/copyout results is extra unnecessary data movement.

This:

copyin(fact, u1, u2, u3, gu0)

can be purged.

Similar here

qe_test_openacc/PHonon/PH/dvqpsi_us.f90

Line 183 in a37c0ec

!$acc parallel loop private(itmp,gtau,gu) present(eigts1, eigts2, eigts3, g, mill,drhoc) copyin(u1,u2,u3,drc,fact,mu,gu0)