fixed writing out GPU options for ELPA

See merge request siesta-project/siesta!286

This removes nprocs writings, and also
correctly reads the configured values
depending on the CMake variables and
the environment variable.

I have removed the 3 print-outs where
only 1 is necessary for the end user.
The env-var is written at the top level
so that can be used to track whether env
vars are used. Otherwise it should be
trivial to check ones fdf whether one
has the GPU option there.

Also, the options are only written
out if ELPA is selected as the kernel.
This should clarify whether the users
are actually using ELPA with GPU support
much simpler.

Signed-off-by: Nick Papior <nickpapior@gmail.com>

Remove mentions of 'off-site' SOC in manual

See merge request siesta-project/siesta!285

Also, add comments on the 'on-site' approximation.

Rolled back NAME_ALTER usage in parseout.

See merge request siesta-project/siesta!282

Tidy up tools (master)

See merge request siesta-project/siesta!279

(cherry picked from commit 376073c9)

moved CH_OVERLAP print-out into debug print

Closes #345

See merge request siesta-project/siesta!276

fixes reading XV files when only UseSaveData is set

Closes #361

See merge request siesta-project/siesta!277

(cherry picked from commit 1bff3fa6)

(cherry picked from commit bacc86a0)

(cherry picked from commit 9fcd2518)

enabled running tests with spaces in build-dir

See merge request siesta-project/siesta!269

Signed-off-by: Nick Papior <nickpapior@gmail.com>

Some restructuring of the tests was required.

The Python interpreter is now found using CMake.
When the comparison should not be made, then the python script will return the error
code that is catched in CMake.

Ran black on the yaml_compare.py script.
It is now simpler to read.

Signed-off-by: Nick Papior <nickpapior@gmail.com>

Update INSTALL.md

See merge request siesta-project/siesta!272

Allow control of the string key for enabling GPU operation in the ELPA interface

Closes #349 and #359

See merge request siesta-project/siesta!266

The ELPA interface uses the idiom

    elpa%set("nvidia-gpu",1)

to enable GPU support.

The appropriate value of the string key changes with the kind of GPU
(e.g. "amd-gpu"), or the ELPA version (e.g. "gpu" in older
versions). There is no simple way to know what the right string is.

We provide several ways to configure and set the string. In increasing
precedence:

- At build time, by setting the CMake variable SIESTA_ELPA_GPU_STRING
  (which might be set upon discovery of the version or architectural
  details of the ELPA library). If this variable is not set, the
  default is "nvidia-gpu".

- Through a fdf variable 'diag-elpa-gpu-string', which defaults to the
  configured value.

- Through an environment variable SIESTA_ELPA_GPU_STRING.

Detect the use of the NVHPC compiler and set some options

See merge request siesta-project/siesta!270

The Nvidia compiler nvfortran (NVHPC) has some issues with the code.

* We detect the use of nvfortran and set a needed Fortran flag. 

* Also, deactivate the use of the legacy MPI interfaces and disable 
  by default the flook library.

Clean up of the README file

See merge request siesta-project/siesta!236

It now contains badges and direct links to relevant information.

fixed version-printout and flags print-out

Closes #353 and #355

See merge request siesta-project/siesta!263

The print-out of version, libraries and flags
has been updated.

The Src/SIESTA_cmake_vgen.sh script has been renamed
to Tools/version_generate.sh

The SIESTA prefix is superfluous, and the placement
is now better.

The build for the various version implementations
are now much simpler, and should not require any
copying of files (which can be error-prone).
Utilites who wants to write out a version and
associated flags/ppflags can now follow
the quite simple recipe in the TS/TBtrans
utility.

It boils down to a couple of things:

1. translating the FLAGS into a string
2. extracting the FPPFLAGS for the executable (not all gathering)
3. translate the FPPFLAGS into a string
4. configure_file the Src/version-info-template.inc
   into the current utility's binary-dir
5. add the binary-dir to the include directories

The above will only use the siesta-used files
for everything. So no code duplication etc.
One could argue that it should be done in a library.
However, the different utilities may use different compile
definitions, hence different capabilities.
For now, I think this is the simplest, yet very effective
way to propagate how its shown to all utilities.

All uncessesary files have been deleted.

The Docs/{release.sh,Makefile} are updated
to respect this change as well.

As for the building, now Siesta prints out the
error messages a bit more stricter, with
some more clear indication of how it interprets
the versions.

This fixes #353, #355.

Signed-off-by: Nick Papior <nickpapior@gmail.com>

Prefix options and some variables with 'SIESTA_'

See merge request siesta-project/siesta!261

Re-enabled 2-stage diagonalization algorithms.

See merge request siesta-project/siesta!257

In an earlier commit HAS_2STAGE	was renamed to LAPACK_HAS_2STAGE
within the CMake machinery, but the new names were not properly used
in the main and Unfolding CMakeLists.txt files.

This fixes the issue to properly handle the symbol and enable the
functionality.

Re-structuring and upgrades to CMake infrastructure (Updated log message)

Closes #318

See merge request siesta-project/siesta!256

(The log message in this empty commit was intended for commit
 f9158b2e, but it was not used due to an error.)

* Bumped minimal version to 3.20 which allows one to DEFER calls
  (post-pone a call until the end of the add_subdirectory step). (3.19
  for DEFER, 3.20 for fixing #318)

* Added:

    SIESTA_TESTS
    SIESTA_SHARED_LIBS
    SIESTA_INSTALL

  mainly for future-proofing the codebase when being inherited in a
  project.

* Added some more print-build-options at the end of the configure
  step.  The shown options are mainly those that affect only
  siesta. I.e. units-convention, suffix names etc.  It gives a better
  overview.
  
* Added the CMAKE_MESSAGE_CONTEXT which can be quite handy.

* Changed lots of explicit function names in debug messages to use
  CMAKE_CURRENT_FUNCTION (does not work in macro's for obvious reasons).

* Enabled the flags:

    SIESTA_LINKER_FLAGS_PRE
    SIESTA_LINKER_FLAGS_POST
    SIESTA_LINKER_FLAGS

  initially for the siesta and libsiesta targets (more will be added
  in a later commit).
  
  The scheme works by creating a wrapper for add_executable|library,
  which defines the target with all its arguments, and then
  immediately calls

    target_link_libraries(<target> ${SIESTA_LINKER_FLAGS_PRE})
   
  It then installs a deferred call when the directory is
  done parsing which calls:

    target_link_libraries(<target> ${SIESTA_LINKER_FLAGS_POST|})

  (We might also consider doing the deferred call only when the
  PROJECT_SOURCE_DIR is done.)

  Basically this wraps all of this:

    add_executable(target ...)
    target_link_libraries(target SIESTA_LINKER_FLAG_PRE)
    < add more link libraries >
    target_link_libraries(target SIESTA_LINKER_FLAG_POST)

  into:

    siesta_add_executable(target ...)
    < add more link libraries >

  It is not completely clear that this is the best way to do things.
  An alternative could be to create other functions which install the
  PRE and POST flags, but that would require two more function calls at
  determined positions.

* Beautified the printbuildinfo scripts.

  When using different modes in the MESSAGE one needs to worry about
  the -- prefix for STATUS and above (i.e. NOTICE won't have these).
  For better formatting now everything is STATUS.  (Once 3.25 is in,
  one can bypass the mode flag and use a custom msg function which
  prints with NOTICE mode but only for DEBUG.)  Also reduced the
  spacing a bit, and just generally made it look prettier.

* The test-suite now used input flags with new names:

   SIESTA_TESTS_MPI_MIN|MAX_NUMPROCS (follows MPI naming convention)
   SIESTA_TESTS_MPI_NUMPROCS (actual used number of MPI ranks, defaults to 4)
   SIESTA_TESTS_VERIFY
   
* A function siesta_test_get_mpi has been added, but it is currently not
  used, pending some feedback.

* The test functions have been shortened to a great extent.

* The `IS_SIMPLE` option has been removed since it got superseeded by
  the `LABELS` (see internal #344).

* Changed the test names to be a bit shorter, and also changed the
  amount of printed information. Much of it can be regarded as debug
  information.

* The SUFFIX handling in Siesta has now a reduced output scope, with
  reduced verbosity.

Re-structuring and upgrades to cmake-infrastructure

Closes #318, #342, and #344

See merge request siesta-project/siesta!247

* Bumped minimal version to 3.20 which allows one to DEFER
  qcalls (post-pone a call until the end of the add_subdirectory
  step). (3.19 for DEFER, 3.20 for fixing #318)

* Added:

SIESTA_TESTS
SIESTA_SHARED_LIBS
SIESTA_INSTALL
The above is mainly for future-proofing the codebase against
being inherited in a project.

After advice from Christian Le, (LecrisUT).
One should generally define the exposed executables through
the top-level CMakeLists.txt. I tried moving them up, but
then the source compilations happen at the directory of definition.
So I decided not to do it.
Added some more print-build-options at the end of the configure step.
The shown options are mainly those that
affect only siesta. I.e. units-convention, suffix names etc.
It gives a better overview.
Added the CMAKE_MESSAGE_CONTEXT which can be quite handy.
Changed lots of explicit function names in debug messages to use
CMAKE_CURRENT_FUNCTION (does not work in macro's for obvious
reasons).
Enabled these flags:

SIESTA_LINKER_FLAGS_PRE
SIESTA_LINKER_FLAGS_POST
SIESTA_LINKER_FLAGS

these will be added to the siesta and libsiesta targets.
(more will be added in a later commit).
These flags are the reason we have to bump the cmake dependency
version.
The reason is that we now require a new way of adding libraries.
(I am up for suggestions to streamline this).
The way this works is by creating a wrapper for add_executable|library.
This wrapper then defines the executable with all its arguments,
and then immediately installs a
target_link_libraries( ${SIESTA_LINKER_FLAGS_PRE})
It then installs a deferred call when the directory is
done parsing which calls:
target_link_libraries( ${SIESTA_LINKER_FLAGS_POST|})
We might consider doing the deferred call only when the PROJECT_SOURCE_DIR
is done (can easily be done).
Basically this wraps all of this:
add_executable(target ...)
target_link_libraries(target SIESTA_LINKER_FLAG_PRE)
< add more link libraries >
target_link_libraries(target SIESTA_LINKER_FLAG_POST)
into:
siesta_add_executable(target ...)
< add more link libraries >
Whether or not this is the best way to do things, I don't know.
We could also create other functions which installs the PRE and POST
flags, but that would require 2 more function calls at determined
positions.
Beautified the printbuildinfo scripts.
when using different modes in the MESSAGE one needs to
worry about the -- prefix for STATUS and above (i.e. NOTICE
won't have these).
For better formatting now everything is STATUS.
Once 3.25 is in, one can bypass the mode flag and use
a custom msg function which prints with NOTICE mode but only
for DEBUG.
Reduced the spacing a bit, and just generally made it
look prettier.
The test-suite got changed input flags:

SIESTA_TESTS_MPI_MIN|MAX_NUMPROCS (follows MPI naming convention)
SIESTA_TESTS_MPI_NUMPROCS (actual used number of ranks, defaults to 4)
SIESTA_TESTS_VERIFY
A function
siesta_test_get_mpi has been added. It is currently not used,
as I would like some feedback on it.
The test functions were shortened to a great extend due
to equivalent branches, the IS_SIMPLE option is now also removed
since it got superseeded by the LABELS (see internal #344).

Changed the test names to be a bit shorter, and also changed
the amount of printed information. Much of it I would regard
as debug information.
Should we simply use prefix SIESTA_TEST_?
The SUFFIX handling in Siesta got a reduced output scope.
It generally shouldn't require a big investigation, and hence
it has reduced its verbosity.
Signed-off-by: Nick Papior <nickpapior@gmail.com>

Details
Please detail what this MR introduces


 Describe in details what this MR does? (link to the issue by writing # here)

 Does this MR change the behaviour of any fdf-flags? If so, ensure this is documented in the manual

 If new fdf-flags are added, please add thorough descriptions to the manual

 add a summary sentence in the ReleaseNotes.md file

 if this breaks compatibility, add mentions to the Docs/compatibility.tex

 backport to rel-5

Also needs to fix the last details of the SiestaSubroutine utils (which is currently commented out).

Why is it needed


Authors
Closes #318, #342, #344 (and possibly others)

Fix supercell-related bug in Mulliken overlaps

See merge request siesta-project/siesta!254

When an auxiliary supercell was used, the 'column' in the DM data
structure was not correctly mapped to the unit cell.

This fix corrects both the 'orbital-based' and the 'atom-based'
overlaps that are computed when write-mulliken-pop is 2 or 3.
Diagonal overlap values are now consistent with the corresponding
populations.

Assorted fixes in utilities

Closes #346

See merge request siesta-project/siesta!252