Skew

See merge request !26

Pkus/devel

See merge request !25

Gitlab CI: Test for 64bit blas and 32bit MPI

See merge request !24

This commit addresses several issues. It essentially forbids the use of
the GPU kernel, which become obsolete and caused problems. But it
does not complete remove the related code, nor does it forbid from
explicitly selecting the GPU kernel. However, if the user does select
it, the warning will be issued and the GENERIC kernel would be used
instead. In the more details:
* Commentin out operations in the GPU kernel, which do not compile with
  CUDA 10.1. This makes the kernel deffinitely not ussable (but it was
  true even before)
* removing the gpu_tridiag_band option, sincie the tridiag->banded routine
  is actually not ported to GPU at all. This step will thus always be
  run on the CPU
* removing the gpu_trans_ev_tridi_to_band option, since the GPU version
  of this step cannot run without the GPU kernel and it is not usable.
  This step will thus also be performed on the CPU
* modifying REAL_GPU_KERNEL_ONLY_WHEN_GPU_IS_ACTIVE and
  COMPLEX_GPU_KERNEL_ONLY_WHEN_GPU_IS_ACTIVE such that the GPU kernel is
  not considered during the autotuning

* TODO however, the GPU kernel can still be enforced by the user. In
  this case, during the calculation, a warning is issued and the kernel
  is switched to the GENERIC one. This should be improved and there
  should not even be the possibility to choose the GPU kernel at the
  begining.

Long int scalapack

See merge request !23

Add support for NVTX profiling

See merge request !22

with cudaDeviceSynchronize

a_dev was never freed on the GPU
However, this might not be enough. What if bandred runs on GPU and
band_to_tridi on CPU? a_dev is then not allocated. Has to be re-thinked
in general

When profiling the GPU version, NVTX can be used to highlight the
corresponding regions of the code in the timeline of the profiling tool
(nvvp or nsight systems). This is very useful to correlate what happens
on the GPU with what part of the code we are in.

ELPA can now be linked against a 64bit integer version of MPI and
ScalaPack. This is an experimental feature

The following points are still to be done
- does not work with real QR-decomposition
- generalized routines return wrong results
- the C tests and the C Cannon algorithm implementation do not
  work (no 64bit header files for MPI *at least* with Intel MPI)

ELPA can now link agains a 64bit integer verion of BLAS/LAPACK.
Currently this only works if ELPA is compiled with MPI=OFF!

The 64bit support is not available in the legacy interface

Seperate the variable type definition of the library and the test programs

See merge request !21

Auto detect

See merge request !20

On heterogenous cluster, of nodes with different CPUs the _experimental_
feature (--enable-heterogenous-cluster-support) can be used:

It compares the (Intel) cpuid set of all CPUs which are used by ELPA MPI
processes and finds the SIMD instruction set, which is supported by all
used CPUs. The ELPA 2stage back-transformation kernel (a.k.a "kernel")
will be set accordingly on all MPI processes.

This feature, can override the setting of the kernel done previously by
the user!

At the moment it will only work for Intel CPUs, i.e. clusters consisting
of nodes with Intel CPUs and e.g. AMD CPUs are at the moment _NOT_
supported.

Since this is an experimental feature, it might be dropped again in the
future, if it turns out not to be useful for the users