Commit ff31135d authored by Ask Hjorth Larsen's avatar Ask Hjorth Larsen

update documentation/docstrings that mention scalapack variables. (Several...

update documentation/docstrings that mention scalapack variables.  (Several are probably way outdated anyway)
parent b63bc58c
......@@ -550,15 +550,13 @@ The number of bands (1056) and cores are chosen to make comparisons
of different band parallelizations (:ref:`band_parallelization`) possible.
**Note**: to achieve optimal performance diagonalization steps are performed
on `4x4` blacs grid with block size of `64` specified by adding ``--gpaw=blacs=1 --sl_diagonalize=4,4,64`` options.
on `4x4` blacs grid with block size of `64`. This can be specified
using the keyword `parallel={'sl_diagonalize': (4, 4, 64)}`.
**Note** also that a default domain decomposition is appplied, and different
results can be obtained by tuning ``--domain-decomposition`` argument
to your platform.
**Note**: the ``--gpaw=usenewlfc=1`` option is required to skip the calculation of forces
and decrease **memory** usage.
The results of the benchmark is scaling of execution time of different stages
of GPAW run with the number of processes (CPU cores).
......@@ -707,15 +705,13 @@ The number of bands (1728) and cores are chosen to make comparisons
of different band parallelizations (:ref:`band_parallelization`) possible.
**Note**: to achieve optimal performance diagonalization steps are performed
on `5x5` blacs grid with block size of `64` specified by adding ``--gpaw=blacs=1 --sl_diagonalize=5,5,64`` options.
on `5x5` blacs grid with block size of `64`. This can be specified
using the keyword `parallel={'sl_diagonalize': (4, 4, 64)}`.
**Note** also that a default domain decomposition is appplied, and different
results can be obtained by tuning ``--domain-decomposition`` argument
to your platform.
**Note**: the ``--gpaw=usenewlfc=1`` option is required to skip the calculation of forces
and decrease **memory** usage.
The results of the benchmark is scaling of execution time of different stages
of GPAW run with the number of processes (CPU cores).
......
......@@ -339,10 +339,8 @@ benefit from ScaLAPACK; otherwise, the default serial LAPACK might as
well be used.
The ScaLAPACK parameters
are defined either using the aforementioned ``'sl_...'`` entry in the parallel
keyword dictionary (recommended) such as ``sl_default=(m, n, block)``,
or alternatively using a command line argument such as
``--sl_default=m,n,block``.
are defined using the parallel
keyword dictionary, e.g., ``sl_default=(m, n, block)``.
A block size of 64 has been found to be a universally good choice both
in all modes.
......@@ -372,13 +370,9 @@ possible to use different ScaLAPACK parameters in the LCAO
initialization and the FD calculation by using two of the ScaLAPACK
keywords in tandem, e.g::
--sl_lcao=p,q,pb --sl_default=m,n,mb
GPAW(..., parallel={'sl_lcao': (p, q, p), 'sl_default': (m, n, mb)})
where ``p``, ``q``, ``pb``, ``m``, ``n``, and ``mb`` all
have different values. The most general case is the combination
of three ScaLAPACK keywords, e.g::
--sl_lcao=p,q,pb --sl_diagonalize=m,n,mb --sl_inverse_cholesky=r,s,rb
however, we do not presently support ``m != r``, ``n != s``, and
``mb != rb``. We may implement this in the future.
of three ScaLAPACK keywords.
Note that some combinations of keywords may not be supported.
......@@ -139,7 +139,7 @@ The p-valence basis sets can be easily obtained for the appropriate elements
with the :command:`gpaw install-data` tool using the following options::
$ gpaw install-data {<directory>} --basis --version=pvalence
See :ref:`installation of paw datasets` for more information on basis set
installation. It is again reminded that these basis sets are not thoroughly
tested and **it is essential to benchmark the performance of the basis sets
......@@ -159,9 +159,9 @@ silver, and gold clusters.
For further details of the basis sets, as well as their construction and
performance, see [#Rossi2015]_. For convenience, these basis sets can be easily
obtained with::
$ gpaw install-data {<directory>} --basis --version=coopt
See :ref:`installation of paw datasets` for basis set installation. Finally,
it is again emphasized that when using the basis sets, **it is essential to
benchmark their suitability for your application**.
......@@ -172,8 +172,7 @@ Parallelization
LCAO-TDDFT is parallelized using ScaLAPACK. It runs without ScaLAPACK, but in this case only a single core is used for linear alrebra.
* Use ``parallel={'sl_default':(N,M,64)}``; See :ref:`manual_parallel`.
* ScaLAPACK can be also enabled by specifying --sl_default=N,M,64 in command line.
* Use ``parallel={'sl_default':(N, M, 64)}``; See :ref:`manual_parallel`.
* It is necessary that N*M equals the total number of cores used by the calculator, and ``max(N,M)*64 < nbands``. The block size 64 can be changed to, e.g., 16 if necessary.
* Apart from parallelization of linear algrebra, normal domain and band parallelizations can be used. As in ground-state LCAO calculations, use band parallelization to reduce memory consumption.
......
......@@ -87,7 +87,7 @@ Example batch job script for GPAW (512 cores, 30 minutes)::
#!/bin/bash -x
#MSUB -l nodes=64:ppn=8
#MSUB -l walltime=0:30:00
cd $PBS_O_WORKDIR
export PYTHONPATH="MY_INSTALLATION_DIR/ase/lib64/python"
export PYTHONPATH="$PYTHONPATH":"MY_INSTALLATION_DIR/gpaw/svn/lib64/python"
......@@ -96,7 +96,7 @@ Example batch job script for GPAW (512 cores, 30 minutes)::
export PSP_ONDEMAND=1
mpiexec -np 512 -x $GPAW_PYTHON my_input.py --sl_default=4,4,64
mpiexec -np 512 -x $GPAW_PYTHON my_input.py
Note that **-x** flag for *mpiexec* is needed for exporting the environment
variables to MPI tasks. The environment variable ``PSP_ONDEMAND`` can decrease
......
......@@ -160,5 +160,5 @@ The installation of user's packages on neolith described below uses
module load campos-gpaw-setups
module load campos-gpaw
mpprun --force-mpi="openmpi/1.2.7-i101017" `which gpaw-python` ./CH4.py --sl_diagonalize=2,1,2
mpprun --force-mpi="openmpi/1.2.7-i101017" `which gpaw-python` ./CH4.py
......@@ -27,10 +27,3 @@ Try also::
This will perform a calculation for a single spin-polarized hydrogen atom
parallelized with spin up on one processor and spin down on the other.
If you enabled ScaLAPACK, do::
$ mpirun -np 2 gpaw-python ~/gpaw/test/CH4.py --sl_default=1,2,2
This will enable ScaLAPACK's diagonalization on a 1x2 BLACS grid
with the block size of 2.
......@@ -17,17 +17,17 @@ class AGTSJob:
Example:
>>> job = AGTSJob('doc/devel/256H2O', 'b256H2O.py --sl_default=4,4,16')
>>> job = AGTSJob('doc/devel/256H2O', 'b256H2O.py --gpaw=debug=True')
>>> job.dir
'doc/devel/256H2O'
>>> job.script
'b256H2O.py'
>>> job.args
'--sl_default=4,4,16'
'--gpaw=debug=True'
>>> job.name
'b256H2O.py_--sl_default=4,4,16'
'b256H2O.py_--gpaw=debug=True'
>>> job.absname
'doc/devel/256H2O/b256H2O.py_--sl_default=4,4,16'
'doc/devel/256H2O/b256H2O.py_--gpaw=debug=True'
"""
if ' ' in script:
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment