distributors.py 7.29 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
#
# Copyright(C) 2013-2018 Max-Planck-Society
#
# NIFTy is being developed at the Max-Planck-Institut fuer Astrophysik
# and financially supported by the Studienstiftung des deutschen Volkes.

Martin Reinecke's avatar
Martin Reinecke committed
19
from __future__ import absolute_import, division, print_function
20

Martin Reinecke's avatar
Martin Reinecke committed
21
import numpy as np
22

Martin Reinecke's avatar
Martin Reinecke committed
23
from .. import dobj
24
25
from ..compat import *
from ..domain_tuple import DomainTuple
Martin Reinecke's avatar
Martin Reinecke committed
26
from ..domains.dof_space import DOFSpace
Martin Reinecke's avatar
cleanup  
Martin Reinecke committed
27
from ..domains.power_space import PowerSpace
28
from ..field import Field
Martin Reinecke's avatar
Martin Reinecke committed
29
from ..utilities import infer_space, special_add_at
30
from .linear_operator import LinearOperator
Martin Reinecke's avatar
Martin Reinecke committed
31
32


Martin Reinecke's avatar
Martin Reinecke committed
33
class DOFDistributor(LinearOperator):
Martin Reinecke's avatar
PEP8  
Martin Reinecke committed
34
35
36
37
38
39
    """Operator which distributes actual degrees of freedom (dof) according to
    some distribution scheme into a higher dimensional space. This distribution
    scheme is defined by the dofdex, a degree of freedom index, which
    associates the entries within the operators domain to locations in its
    target. This operator's domain is a DOFSpace, which is defined according to
    the distribution scheme.
40
41
42
43

    Parameters
    ----------
    dofdex: Field of integers
Martin Reinecke's avatar
PEP8  
Martin Reinecke committed
44
        An integer Field on exactly one Space establishing the association
Martin Reinecke's avatar
fixes  
Martin Reinecke committed
45
        between the locations in the operator's target and the underlying
Martin Reinecke's avatar
PEP8  
Martin Reinecke committed
46
        degrees of freedom in its domain.
Martin Reinecke's avatar
fixes  
Martin Reinecke committed
47
        It has to start at 0 and it increases monotonically, no empty bins are
Martin Reinecke's avatar
PEP8  
Martin Reinecke committed
48
        allowed.
49
    target: Domain, tuple of Domain, or DomainTuple, optional
Martin Reinecke's avatar
PEP8  
Martin Reinecke committed
50
51
        The target of the operator. If not specified, the domain of the dofdex
        is used.
52
53
54
55
    space: int, optional:
       The index of the sub-domain on which the operator acts.
       Can be omitted if `target` only has one sub-domain.
    """
Martin Reinecke's avatar
Martin Reinecke committed
56

Martin Reinecke's avatar
Martin Reinecke committed
57
58
59
60
61
62
    def __init__(self, dofdex, target=None, space=None):
        if target is None:
            target = dofdex.domain
        self._target = DomainTuple.make(target)
        space = infer_space(self._target, space)
        partner = self._target[space]
Martin Reinecke's avatar
Martin Reinecke committed
63
64
        if not isinstance(dofdex, Field):
            raise TypeError("dofdex must be a Field")
Martin Reinecke's avatar
Martin Reinecke committed
65
66
67
        if not len(dofdex.domain) == 1:
            raise ValueError("dofdex must live on exactly one Space")
        if not np.issubdtype(dofdex.dtype, np.integer):
Martin Reinecke's avatar
Martin Reinecke committed
68
            raise TypeError("dofdex must contain integer numbers")
Martin Reinecke's avatar
Martin Reinecke committed
69
        if partner != dofdex.domain[0]:
Martin Reinecke's avatar
Martin Reinecke committed
70
71
            raise ValueError("incorrect dofdex domain")

Martin Reinecke's avatar
fixes  
Martin Reinecke committed
72
        ldat = dofdex.local_data
Martin Reinecke's avatar
Martin Reinecke committed
73
        if ldat.size == 0:  # can happen for weird configurations
Martin Reinecke's avatar
fixes  
Martin Reinecke committed
74
75
76
77
            nbin = 0
        else:
            nbin = ldat.max()
        nbin = dobj.np_allreduce_max(np.array(nbin))[()] + 1
Martin Reinecke's avatar
Martin Reinecke committed
78
        if partner.scalar_dvol is not None:
Martin Reinecke's avatar
Martin Reinecke committed
79
            wgt = np.bincount(dofdex.local_data.ravel(), minlength=nbin)
Martin Reinecke's avatar
Martin Reinecke committed
80
            wgt = wgt*partner.scalar_dvol
Martin Reinecke's avatar
Martin Reinecke committed
81
        else:
Martin Reinecke's avatar
fixes  
Martin Reinecke committed
82
            dvol = Field.from_global_data(partner, partner.dvol).local_data
Martin Reinecke's avatar
Martin Reinecke committed
83
            wgt = np.bincount(dofdex.local_data.ravel(),
Martin Reinecke's avatar
Martin Reinecke committed
84
85
86
87
88
89
90
91
92
                              minlength=nbin, weights=dvol)
        # The explicit conversion to float64 is necessary because bincount
        # sometimes returns its result as an integer array, even when
        # floating-point weights are present ...
        wgt = wgt.astype(np.float64, copy=False)
        wgt = dobj.np_allreduce_sum(wgt)
        if (wgt == 0).any():
            raise ValueError("empty bins detected")

Martin Reinecke's avatar
Martin Reinecke committed
93
94
95
        self._init2(dofdex.val, space, DOFSpace(wgt))

    def _init2(self, dofdex, space, other_space):
Martin Reinecke's avatar
Martin Reinecke committed
96
        self._space = space
Martin Reinecke's avatar
Martin Reinecke committed
97
98
99
        dom = list(self._target)
        dom[self._space] = other_space
        self._domain = DomainTuple.make(dom)
Martin Reinecke's avatar
Martin Reinecke committed
100
        self._capability = self.TIMES | self.ADJOINT_TIMES
Martin Reinecke's avatar
Martin Reinecke committed
101

Martin Reinecke's avatar
Martin Reinecke committed
102
103
        if dobj.default_distaxis() in self._domain.axes[self._space]:
            dofdex = dobj.local_data(dofdex)
Martin Reinecke's avatar
Martin Reinecke committed
104
        else:  # dofdex must be available fully on every task
Martin Reinecke's avatar
Martin Reinecke committed
105
            dofdex = dobj.to_global_data(dofdex)
Martin Reinecke's avatar
Martin Reinecke committed
106
        self._dofdex = dofdex.ravel()
Martin Reinecke's avatar
Martin Reinecke committed
107
108
109
        firstaxis = self._target.axes[self._space][0]
        lastaxis = self._target.axes[self._space][-1]
        arrshape = dobj.local_shape(self._target.shape, 0)
Martin Reinecke's avatar
Martin Reinecke committed
110
111
        presize = np.prod(arrshape[0:firstaxis], dtype=np.int)
        postsize = np.prod(arrshape[lastaxis+1:], dtype=np.int)
Martin Reinecke's avatar
Martin Reinecke committed
112
        self._hshape = (presize, self._domain[self._space].shape[0], postsize)
Martin Reinecke's avatar
Martin Reinecke committed
113
114
        self._pshape = (presize, self._dofdex.size, postsize)

Martin Reinecke's avatar
Martin Reinecke committed
115
    def _adjoint_times(self, x):
Martin Reinecke's avatar
Martin Reinecke committed
116
        arr = x.local_data
Martin Reinecke's avatar
Martin Reinecke committed
117
118
        arr = arr.reshape(self._pshape)
        oarr = np.zeros(self._hshape, dtype=x.dtype)
Martin Reinecke's avatar
tweaks  
Martin Reinecke committed
119
        oarr = special_add_at(oarr, 1, self._dofdex, arr)
Martin Reinecke's avatar
Martin Reinecke committed
120
        if dobj.distaxis(x.val) in x.domain.axes[self._space]:
Martin Reinecke's avatar
cleanup  
Martin Reinecke committed
121
122
            oarr = oarr.reshape(self._domain.shape)
            res = Field.from_global_data(self._domain, oarr, sum_up=True)
Martin Reinecke's avatar
Martin Reinecke committed
123
        else:
Martin Reinecke's avatar
cleanup  
Martin Reinecke committed
124
125
            oarr = oarr.reshape(self._domain.local_shape)
            res = Field.from_local_data(self._domain, oarr)
126
        return res
Martin Reinecke's avatar
Martin Reinecke committed
127

Martin Reinecke's avatar
Martin Reinecke committed
128
    def _times(self, x):
Martin Reinecke's avatar
Martin Reinecke committed
129
        if dobj.distaxis(x.val) in x.domain.axes[self._space]:
130
            arr = x.to_global_data()
Martin Reinecke's avatar
Martin Reinecke committed
131
        else:
Martin Reinecke's avatar
Martin Reinecke committed
132
            arr = x.local_data
Martin Reinecke's avatar
Martin Reinecke committed
133
        arr = arr.reshape(self._hshape)
Martin Reinecke's avatar
Martin Reinecke committed
134
135
136
137
        oarr = np.empty(self._pshape, dtype=x.dtype)
        oarr[()] = arr[(slice(None), self._dofdex, slice(None))]
        return Field.from_local_data(
            self._target, oarr.reshape(self._target.local_shape))
Martin Reinecke's avatar
Martin Reinecke committed
138

Martin Reinecke's avatar
Martin Reinecke committed
139
140
    def apply(self, x, mode):
        self._check_input(x, mode)
Martin Reinecke's avatar
Martin Reinecke committed
141
        return self._times(x) if mode == self.TIMES else self._adjoint_times(x)
Martin Reinecke's avatar
cleanup  
Martin Reinecke committed
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175


class PowerDistributor(DOFDistributor):
    """Operator which transforms between a PowerSpace and a harmonic domain.

    Parameters
    ----------
    target: Domain, tuple of Domain, or DomainTuple
        the total *target* domain of the operator.
    power_space: PowerSpace, optional
        the input sub-domain on which the operator acts.
        If not supplied, a matching PowerSpace with natural binbounds will be
        used.
    space: int, optional:
       The index of the sub-domain on which the operator acts.
       Can be omitted if `target` only has one sub-domain.
    """

    def __init__(self, target, power_space=None, space=None):
        # Initialize domain and target
        self._target = DomainTuple.make(target)
        self._space = infer_space(self._target, space)
        hspace = self._target[self._space]
        if not hspace.harmonic:
            raise ValueError("Operator requires harmonic target space")
        if power_space is None:
            power_space = PowerSpace(hspace)
        else:
            if not isinstance(power_space, PowerSpace):
                raise TypeError("power_space argument must be a PowerSpace")
            if power_space.harmonic_partner != hspace:
                raise ValueError("power_space does not match its partner")

        self._init2(power_space.pindex, self._space, power_space)