Working on documentation

Author: mikaem

docs/source/global.rst

@@ -252,8 +252,10 @@ Multidimensional distributed arrays
 
 The procedure discussed above remains the same for any type of array, of any
 dimensionality. With mpi4py-fft we can distribute any array of arbitrary dimensionality
-using an arbitrary number of processor groups. How to distribute is completely
-configurable through the classes in the :mod:`.pencil` module.
+using any number of processor groups. We only require that the number of processor
+groups is at least one less than the number of dimensions, since one axis must
+remain aligned. Apart from this the distribution is completely configurable through
+the classes in the :mod:`.pencil` module.
 
 We denote a global :math:`d`-dimensional array as :math:`u_{j_0, j_1, \ldots, j_{d-1}}`,
 where :math:`j_m\in\textbf{j}_m` for :math:`m=[0, 1, \ldots, d-1]`.
@@ -263,7 +265,7 @@ than one processor group, the groups are indexed, like :math:`P_0, P_1` etc.
 
 Lets illustrate using a 4-dimensional array with 3 processor groups. Let the
 array be aligned only in axis 3 first (:math:`u_{j_0/P_0, j_1/P_1, j_2/P_2, j_3}`),
-and then redistributed for alignment along axes 2, 1 and finally 0. Mathematically,
+and then redistribute for alignment along axes 2, 1 and finally 0. Mathematically,
 we will now be executing the three following global redistributions:
 
 .. math::
@@ -273,6 +275,13 @@ we will now be executing the three following global redistributions:
     u_{j_0/P_0, j_1, j_2/P_1, j_3/P_2} \xleftarrow[P_1]{2 \rightarrow 1}  u_{j_0/P_0, j_1/P_1, j_2, j_3/P_2} \\
     u_{j_0, j_1/P_0, j_2/P_1, j_3/P_2} \xleftarrow[P_0]{1 \rightarrow 0}  u_{j_0/P_0, j_1, j_2/P_1, j_3/P_2}
 
+Note that in the first step it is only processor group :math:`P_2` that is
+active in the redistribution, and the output (left hand side) is now aligned
+in axis 2. This can be seen since there is no processor group there to
+share the :math:`j_2` index.
+In the second step processor group :math:`P_1` is the active one, and
+in the final step :math:`P_0`.
+
 Now, it is not necessary to use three processor groups just because we have a
 four-dimensional array. We could just as well have been using 2 or 1. The advantage
 of using more groups is that you can then use more processors in total. Assuming

docs/source/howtocite.rst

@@ -7,13 +7,13 @@ Please cite mpi4py-fft using
 
     @article{jpdc_fft,
         author = {{Dalcin, Lisandro and Mortensen, Mikael and Keyes, David E}},
-        year = 2019,
+        year = {{2019}},
         title = {{Fast parallel multidimensional FFT using advanced MPI}},
         journal = {{Journal of Parallel and Distributed Computing}},
-        volume = in press
+        volume = {{in press}}
     }
     @electronic{mpi4py-fft,
-	    author = {{Lisandro Dalcin and Mikael Mortensen}},
-	    title = {{mpi4py-fft}},
-	    url = {https://bitbucket.org/mpi4py/mpi4py-fft}
+        author = {{Lisandro Dalcin and Mikael Mortensen}},
+        title = {{mpi4py-fft}},
+        url = {{https://bitbucket.org/mpi4py/mpi4py-fft}}
     }

docs/source/installation.rst

@@ -37,7 +37,7 @@ is not installed. This can be achieved with, e.g.,
 
 ::
 
-    conda create --name mpi4py-fft -c conda-forge mpi4py-fft mpich nomkl
+    conda create --name mpi4py-fft -c conda-forge mpi4py-fft mpich nomkl h5py=*=mpi*
 
 Note that the nomkl package makes sure that numpy is installed without
 mkl, whereas mpich here chooses this backend over openmpi.
@@ -50,11 +50,13 @@ any version of mpi4py-fft hosted on `pypi`_ using `pip`_
 
     pip install mpi4py-fft
 
-whereas the following will install the latest version from github
+whereas either one of the following will install the latest version
+from github
 
 ::
 
     pip install git+https://bitbucket.org/mpi4py/mpi4py-fft@master
+    pip install https://bitbucket.org/mpi4py/mpi4py-fft/get/master.zip
 
 You can also build mpi4py-fft yourselves from the top directory,
 after cloning or forking

mpi4py_fft/distributedarray.py

@@ -1,4 +1,5 @@
 import os
+from numbers import Number
 import numpy as np
 from mpi4py import MPI
 from .pencil import Pencil, Subcomm
@@ -85,7 +86,7 @@ def __new__(cls, global_shape, subcomm=None, val=None, dtype=np.float,
         if rank > 0:
             subshape = global_shape[:rank] + subshape
         obj = np.ndarray.__new__(cls, subshape, dtype=dtype, buffer=buffer)
-        if buffer is None and isinstance(val, int):
+        if buffer is None and isinstance(val, Number):
             obj.fill(val)
         obj._p0 = p0
         obj._rank = rank