Still missing is the compatibility layer, currently it only compiles
when configure is called with

  --disable-legacy

Also, a more general solution to parameter passing via environment
variables would be nice.

This reverts commit 64ec6ec2.

PGI's compiler wants

  -module

as the flag to specify the module output directory. Sadly, this is also
a libtool argument, resulting in much sorrow.

Thus, we have to escape -module by prefixing it with -Xcompiler. Sadly
libtool is not used everywhere by automake, thus we have to remove it
again in those cases were it would be passed to the compiler and cause
sorrow there.

Sad story. But with a happy ending?

Sometimes the compiler is used with the manual_cpp wrapper, do not
invoke this for linking

This is used to check if we compile Fortran with PGI or Intel and to
then use the CPP wrapper script.

It would be nice to change this check to test for actual misbehaviour of
the preprocessor. TBD

Enough is enough, C99 is not too much to ask in 2017

For some reason, and only in certain cases, the built-in
preprocessor of ifort cannot handle constructs such as

  #define FOO bar

  print *, "foo&
  &FOO&
  &baz"

or more specifically

  #define DATATYPE double

  print *, some_&
  &DATATYPE&
  &_routine

which, however, are quite handy. GCC seems to handle those as
expected.

This gives a speed-up of up to a factor 2 in the complex version

Now the CUDA compute capability is not hard-coded anymore.
Setting

--with-gpu-compute-capability=value

set the nvcc compiler option "-arch value".

A simple sanity checking is done to ensure that "value" is not
completely unreasonable

- the version number has been updated
- a warning that this is a pre-release has been added to the configure
  script
- the changelog has been updated
- the test_project versioning is consistent with the ELPA versioning

Conflicts:
	src/elpa1_solve_tridi_real_template.X90

configure can now distinguish the options --enable-assumed-size and
--disable-assumed-size

This can be switched off via --disable-assumed-size

configure can now distinguish the options --enable-assumed-size and
--disable-assumed-size

This can be switched off via --disable-assumed-size

Sometimes it might be desirable that only one of the specifc AVX
kernels (AVX, AVX2, and in the future AVX512) are build, even though the processor
supports all (or several) of these instructions.

The flag --disable-[avx | avx2 | avx512] allows to define which kernels
should NOT be build. The default is: what can be build will be build

Sometimes it might be desirable that only one of the specifc AVX
kernels (AVX, AVX2, AVX512) are build, even though the processor
supports all (or several) of these instructions.

The flag --disable-[avx | avx2 | avx512] allows to define which kernels
should NOT be build. The default is: what can be build will be build

The actual reason for the linker problems was that the Fortran libraries
where listed before the object files, by modifying the _LINK automake
variables. The proper way to do is of course to add the necessary
libraries after the object files by appending them to the _LDADD
variables.

As the MPI module was not responsible for the linker problems it is now
used by default, unless explicitly switched off with

  --disable-mpi-module

Also, the C test programs that had these linker errors where previously
not compiled in the OpenMP case, for no obvious reason. Now they are
also included there.