distributed_do.py 15.7 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 print_function
20
21
22
23
import numpy as np
from .random import Random
from mpi4py import MPI

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
24
25
26
_comm = MPI.COMM_WORLD
ntask = _comm.Get_size()
rank = _comm.Get_rank()
Martin Reinecke's avatar
Martin Reinecke committed
27
master = (rank == 0)
28
29


Martin Reinecke's avatar
Martin Reinecke committed
30
31
32
33
34
def mprint(*args):
    if master:
        print(*args)


Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
35
def _shareSize(nwork, nshares, myshare):
Martin Reinecke's avatar
Martin Reinecke committed
36
    return (nwork//nshares) + int(myshare < nwork % nshares)
Martin Reinecke's avatar
Martin Reinecke committed
37

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
38
39

def _shareRange(nwork, nshares, myshare):
Martin Reinecke's avatar
Martin Reinecke committed
40
41
    nbase = nwork//nshares
    additional = nwork % nshares
Martin Reinecke's avatar
Martin Reinecke committed
42
    lo = myshare*nbase + min(myshare, additional)
Martin Reinecke's avatar
Martin Reinecke committed
43
    hi = lo + nbase + int(myshare < additional)
Martin Reinecke's avatar
Martin Reinecke committed
44
45
    return lo, hi

46

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
47
def local_shape(shape, distaxis):
Martin Reinecke's avatar
Martin Reinecke committed
48
    if len(shape) == 0 or distaxis == -1:
49
        return shape
Martin Reinecke's avatar
Martin Reinecke committed
50
51
    shape2 = list(shape)
    shape2[distaxis] = _shareSize(shape[distaxis], ntask, rank)
52
53
    return tuple(shape2)

Martin Reinecke's avatar
Martin Reinecke committed
54

55
56
class data_object(object):
    def __init__(self, shape, data, distaxis):
Martin Reinecke's avatar
Martin Reinecke committed
57
        self._shape = tuple(shape)
Martin Reinecke's avatar
Martin Reinecke committed
58
        if len(self._shape) == 0:
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
59
            distaxis = -1
60
61
62
        self._distaxis = distaxis
        self._data = data

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
63
    def _sanity_checks(self):
64
        # check whether the distaxis is consistent
Martin Reinecke's avatar
Martin Reinecke committed
65
        if self._distaxis < -1 or self._distaxis >= len(self._shape):
66
            raise ValueError
Martin Reinecke's avatar
Martin Reinecke committed
67
68
69
70
        itmp = np.array(self._distaxis)
        otmp = np.empty(ntask, dtype=np.int)
        _comm.Allgather(itmp, otmp)
        if np.any(otmp != self._distaxis):
71
72
            raise ValueError
        # check whether the global shape is consistent
Martin Reinecke's avatar
Martin Reinecke committed
73
74
75
        itmp = np.array(self._shape)
        otmp = np.empty((ntask, len(self._shape)), dtype=np.int)
        _comm.Allgather(itmp, otmp)
76
        for i in range(ntask):
Martin Reinecke's avatar
Martin Reinecke committed
77
            if np.any(otmp[i, :] != self._shape):
78
79
                raise ValueError
        # check shape of local data
Martin Reinecke's avatar
Martin Reinecke committed
80
81
        if self._distaxis < 0:
            if self._data.shape != self._shape:
82
83
                raise ValueError
        else:
Martin Reinecke's avatar
Martin Reinecke committed
84
85
86
87
            itmp = np.array(self._shape)
            itmp[self._distaxis] = _shareSize(self._shape[self._distaxis],
                                              ntask, rank)
            if np.any(self._data.shape != itmp):
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
                raise ValueError

    @property
    def dtype(self):
        return self._data.dtype

    @property
    def shape(self):
        return self._shape

    @property
    def size(self):
        return np.prod(self._shape)

    @property
    def real(self):
Martin Reinecke's avatar
Martin Reinecke committed
104
        return data_object(self._shape, self._data.real, self._distaxis)
105
106
107

    @property
    def imag(self):
Martin Reinecke's avatar
Martin Reinecke committed
108
        return data_object(self._shape, self._data.imag, self._distaxis)
109

Martin Reinecke's avatar
Martin Reinecke committed
110
111
112
113
114
115
    def conj(self):
        return data_object(self._shape, self._data.conj(), self._distaxis)

    def conjugate(self):
        return data_object(self._shape, self._data.conjugate(), self._distaxis)

Martin Reinecke's avatar
Martin Reinecke committed
116
    def _contraction_helper(self, op, mpiop, axis):
117
        if axis is not None:
Martin Reinecke's avatar
Martin Reinecke committed
118
            if len(axis) == len(self._data.shape):
119
120
                axis = None
        if axis is None:
Martin Reinecke's avatar
Martin Reinecke committed
121
            res = np.array(getattr(self._data, op)())
Martin Reinecke's avatar
Martin Reinecke committed
122
            if (self._distaxis == -1):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
123
                return res[()]
Martin Reinecke's avatar
Martin Reinecke committed
124
125
            res2 = np.empty((), dtype=res.dtype)
            _comm.Allreduce(res, res2, mpiop)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
126
            return res2[()]
127
128

        if self._distaxis in axis:
Martin Reinecke's avatar
Martin Reinecke committed
129
130
            res = getattr(self._data, op)(axis=axis)
            res2 = np.empty_like(res)
Martin Reinecke's avatar
Martin Reinecke committed
131
            _comm.Allreduce(res, res2, mpiop)
Martin Reinecke's avatar
Martin Reinecke committed
132
            return from_global_data(res2, distaxis=0)
133
        else:
Martin Reinecke's avatar
Martin Reinecke committed
134
            # perform the contraction on the local data
Martin Reinecke's avatar
Martin Reinecke committed
135
136
            res = getattr(self._data, op)(axis=axis)
            if self._distaxis == -1:
Martin Reinecke's avatar
Martin Reinecke committed
137
                return from_global_data(res, distaxis=0)
Martin Reinecke's avatar
Martin Reinecke committed
138
            shp = list(res.shape)
Martin Reinecke's avatar
Martin Reinecke committed
139
            shift = 0
Martin Reinecke's avatar
Martin Reinecke committed
140
            for ax in axis:
Martin Reinecke's avatar
Martin Reinecke committed
141
142
                if ax < self._distaxis:
                    shift += 1
Martin Reinecke's avatar
Martin Reinecke committed
143
144
            shp[self._distaxis-shift] = self.shape[self._distaxis]
            return from_local_data(shp, res, self._distaxis-shift)
145
146
147

    def sum(self, axis=None):
        return self._contraction_helper("sum", MPI.SUM, axis)
Martin Reinecke's avatar
Martin Reinecke committed
148

Martin Reinecke's avatar
fixes    
Martin Reinecke committed
149
150
    def min(self, axis=None):
        return self._contraction_helper("min", MPI.MIN, axis)
Martin Reinecke's avatar
Martin Reinecke committed
151

Martin Reinecke's avatar
fixes    
Martin Reinecke committed
152
153
    def max(self, axis=None):
        return self._contraction_helper("max", MPI.MAX, axis)
154

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
155
156
    def mean(self):
        return self.sum()/self.size
Martin Reinecke's avatar
Martin Reinecke committed
157

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
158
159
    def std(self):
        return np.sqrt(self.var())
Martin Reinecke's avatar
Martin Reinecke committed
160

Martin Reinecke's avatar
Martin Reinecke committed
161
    # FIXME: to be improved!
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
162
163
164
    def var(self):
        return (abs(self-self.mean())**2).mean()

165
    def _binary_helper(self, other, op):
Martin Reinecke's avatar
Martin Reinecke committed
166
        a = self
167
        if isinstance(other, data_object):
Martin Reinecke's avatar
Martin Reinecke committed
168
            b = other
169
170
171
172
            if a._shape != b._shape:
                raise ValueError("shapes are incompatible.")
            if a._distaxis != b._distaxis:
                raise ValueError("distributions are incompatible.")
Martin Reinecke's avatar
Martin Reinecke committed
173
174
            a = a._data
            b = b._data
Martin Reinecke's avatar
Martin Reinecke committed
175
176
177
178
        elif np.isscalar(other):
            a = a._data
            b = other
        elif isinstance(other, np.ndarray):
Martin Reinecke's avatar
Martin Reinecke committed
179
            a = a._data
180
            b = other
Martin Reinecke's avatar
Martin Reinecke committed
181
182
        else:
            return NotImplemented
183
184

        tval = getattr(a, op)(b)
Martin Reinecke's avatar
Martin Reinecke committed
185
186
187
188
        if tval is a:
            return self
        else:
            return data_object(self._shape, tval, self._distaxis)
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222

    def __add__(self, other):
        return self._binary_helper(other, op='__add__')

    def __radd__(self, other):
        return self._binary_helper(other, op='__radd__')

    def __iadd__(self, other):
        return self._binary_helper(other, op='__iadd__')

    def __sub__(self, other):
        return self._binary_helper(other, op='__sub__')

    def __rsub__(self, other):
        return self._binary_helper(other, op='__rsub__')

    def __isub__(self, other):
        return self._binary_helper(other, op='__isub__')

    def __mul__(self, other):
        return self._binary_helper(other, op='__mul__')

    def __rmul__(self, other):
        return self._binary_helper(other, op='__rmul__')

    def __imul__(self, other):
        return self._binary_helper(other, op='__imul__')

    def __div__(self, other):
        return self._binary_helper(other, op='__div__')

    def __rdiv__(self, other):
        return self._binary_helper(other, op='__rdiv__')

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
223
224
225
    def __idiv__(self, other):
        return self._binary_helper(other, op='__idiv__')

226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
    def __truediv__(self, other):
        return self._binary_helper(other, op='__truediv__')

    def __rtruediv__(self, other):
        return self._binary_helper(other, op='__rtruediv__')

    def __pow__(self, other):
        return self._binary_helper(other, op='__pow__')

    def __rpow__(self, other):
        return self._binary_helper(other, op='__rpow__')

    def __ipow__(self, other):
        return self._binary_helper(other, op='__ipow__')

    def __eq__(self, other):
        return self._binary_helper(other, op='__eq__')

    def __ne__(self, other):
        return self._binary_helper(other, op='__ne__')

    def __neg__(self):
Martin Reinecke's avatar
Martin Reinecke committed
248
        return data_object(self._shape, -self._data, self._distaxis)
249
250

    def __abs__(self):
Martin Reinecke's avatar
Martin Reinecke committed
251
        return data_object(self._shape, np.abs(self._data), self._distaxis)
252
253

    def all(self):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
254
        return self.sum() == self.size
255
256

    def any(self):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
257
        return self.sum() != 0
258

Martin Reinecke's avatar
fixes    
Martin Reinecke committed
259
260
    def fill(self, value):
        self._data.fill(value)
261

Martin Reinecke's avatar
Martin Reinecke committed
262
def full(shape, fill_value, dtype=None, distaxis=0):
Martin Reinecke's avatar
Martin Reinecke committed
263
264
    return data_object(shape, np.full(local_shape(shape, distaxis),
                                      fill_value, dtype), distaxis)
265
266


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
267
def empty(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
Martin Reinecke committed
268
269
    return data_object(shape, np.empty(local_shape(shape, distaxis),
                                       dtype), distaxis)
270
271


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
272
def zeros(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
Martin Reinecke committed
273
274
    return data_object(shape, np.zeros(local_shape(shape, distaxis), dtype),
                       distaxis)
275
276


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
277
def ones(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
Martin Reinecke committed
278
279
    return data_object(shape, np.ones(local_shape(shape, distaxis), dtype),
                       distaxis)
280
281
282
283
284
285
286


def empty_like(a, dtype=None):
    return data_object(np.empty_like(a._data, dtype))


def vdot(a, b):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
287
    tmp = np.array(np.vdot(a._data, b._data))
Martin Reinecke's avatar
Martin Reinecke committed
288
289
    res = np.empty((), dtype=tmp.dtype)
    _comm.Allreduce(tmp, res, MPI.SUM)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
290
    return res[()]
291
292
293
294
295
296
297


def _math_helper(x, function, out):
    if out is not None:
        function(x._data, out=out._data)
        return out
    else:
Martin Reinecke's avatar
Martin Reinecke committed
298
        return data_object(x.shape, function(x._data), x._distaxis)
299
300
301
302
303
304
305
306
307
308
309
310
311
312


def abs(a, out=None):
    return _math_helper(a, np.abs, out)


def exp(a, out=None):
    return _math_helper(a, np.exp, out)


def log(a, out=None):
    return _math_helper(a, np.log, out)


Martin Reinecke's avatar
Martin Reinecke committed
313
314
315
316
def tanh(a, out=None):
    return _math_helper(a, np.tanh, out)


317
318
319
320
321
def sqrt(a, out=None):
    return _math_helper(a, np.sqrt, out)


def from_object(object, dtype=None, copy=True):
Martin Reinecke's avatar
Martin Reinecke committed
322
323
324
    return data_object(object._shape, np.array(object._data, dtype=dtype,
                                               copy=copy),
                       distaxis=object._distaxis)
325
326


Martin Reinecke's avatar
Martin Reinecke committed
327
328
# This function draws all random numbers on all tasks, to produce the same
# array independent on the number of tasks
Martin Reinecke's avatar
Martin Reinecke committed
329
330
331
# MR FIXME: depending on what is really wanted/needed (i.e. same result
# independent of number of tasks, performance etc.) we need to adjust the
# algorithm.
Martin Reinecke's avatar
Martin Reinecke committed
332
def from_random(random_type, shape, dtype=np.float64, **kwargs):
333
    generator_function = getattr(Random, random_type)
Martin Reinecke's avatar
Martin Reinecke committed
334
335
336
337
338
339
340
    for i in range(ntask):
        lshape = list(shape)
        lshape[0] = _shareSize(shape[0], ntask, i)
        ldat = generator_function(dtype=dtype, shape=lshape, **kwargs)
        if i == rank:
            outdat = ldat
    return from_local_data(shape, outdat, distaxis=0)
341

Martin Reinecke's avatar
Martin Reinecke committed
342

Martin Reinecke's avatar
Martin Reinecke committed
343
344
345
346
def local_data(arr):
    return arr._data


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
347
348
def ibegin(arr):
    res = [0] * arr._data.ndim
Martin Reinecke's avatar
Martin Reinecke committed
349
    res[arr._distaxis] = _shareRange(arr._shape[arr._distaxis], ntask, rank)[0]
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
350
    return tuple(res)
Martin Reinecke's avatar
Martin Reinecke committed
351
352


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
353
354
def np_allreduce_sum(arr):
    res = np.empty_like(arr)
Martin Reinecke's avatar
Martin Reinecke committed
355
    _comm.Allreduce(arr, res, MPI.SUM)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
356
    return res
Martin Reinecke's avatar
Martin Reinecke committed
357
358
359
360
361
362


def distaxis(arr):
    return arr._distaxis


Martin Reinecke's avatar
Martin Reinecke committed
363
def from_local_data(shape, arr, distaxis):
Martin Reinecke's avatar
Martin Reinecke committed
364
365
366
    return data_object(shape, arr, distaxis)


Martin Reinecke's avatar
Martin Reinecke committed
367
368
def from_global_data(arr, distaxis=0):
    if distaxis == -1:
Martin Reinecke's avatar
Martin Reinecke committed
369
        return data_object(arr.shape, arr, distaxis)
Martin Reinecke's avatar
Martin Reinecke committed
370
    lo, hi = _shareRange(arr.shape[distaxis], ntask, rank)
Martin Reinecke's avatar
Martin Reinecke committed
371
    sl = [slice(None)]*len(arr.shape)
Martin Reinecke's avatar
Martin Reinecke committed
372
    sl[distaxis] = slice(lo, hi)
Martin Reinecke's avatar
Martin Reinecke committed
373
374
375
    return data_object(arr.shape, arr[sl], distaxis)


Martin Reinecke's avatar
Martin Reinecke committed
376
377
def to_global_data(arr):
    if arr._distaxis == -1:
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
378
379
380
381
382
        return arr._data
    tmp = redistribute(arr, dist=-1)
    return tmp._data


Martin Reinecke's avatar
Martin Reinecke committed
383
def redistribute(arr, dist=None, nodist=None):
Martin Reinecke's avatar
Martin Reinecke committed
384
385
386
    if dist is not None:
        if nodist is not None:
            raise ValueError
Martin Reinecke's avatar
Martin Reinecke committed
387
        if dist == arr._distaxis:
Martin Reinecke's avatar
Martin Reinecke committed
388
389
390
391
392
393
            return arr
    else:
        if nodist is None:
            raise ValueError
        if arr._distaxis not in nodist:
            return arr
Martin Reinecke's avatar
Martin Reinecke committed
394
        dist = -1
Martin Reinecke's avatar
Martin Reinecke committed
395
396
        for i in range(len(arr.shape)):
            if i not in nodist:
Martin Reinecke's avatar
Martin Reinecke committed
397
                dist = i
Martin Reinecke's avatar
Martin Reinecke committed
398
                break
Martin Reinecke's avatar
Martin Reinecke committed
399

Martin Reinecke's avatar
Martin Reinecke committed
400
    if arr._distaxis == -1:  # all data available, just pick the proper subset
Martin Reinecke's avatar
Martin Reinecke committed
401
        return from_global_data(arr._data, dist)
Martin Reinecke's avatar
Martin Reinecke committed
402
    if dist == -1:  # gather all data on all tasks
Martin Reinecke's avatar
Martin Reinecke committed
403
        tmp = np.moveaxis(arr._data, arr._distaxis, 0)
Martin Reinecke's avatar
Martin Reinecke committed
404
405
        slabsize = np.prod(tmp.shape[1:])*tmp.itemsize
        sz = np.empty(ntask, dtype=np.int)
Martin Reinecke's avatar
Martin Reinecke committed
406
        for i in range(ntask):
Martin Reinecke's avatar
Martin Reinecke committed
407
408
409
410
            sz[i] = slabsize*_shareSize(arr.shape[arr._distaxis], ntask, i)
        disp = np.empty(ntask, dtype=np.int)
        disp[0] = 0
        disp[1:] = np.cumsum(sz[:-1])
Martin Reinecke's avatar
Martin Reinecke committed
411
        tmp = np.require(tmp, requirements="C")
Martin Reinecke's avatar
Martin Reinecke committed
412
413
        out = np.empty(arr.size, dtype=arr.dtype)
        _comm.Allgatherv(tmp, [out, sz, disp, MPI.BYTE])
Martin Reinecke's avatar
Martin Reinecke committed
414
415
416
417
        shp = np.array(arr._shape)
        shp[1:arr._distaxis+1] = shp[0:arr._distaxis]
        shp[0] = arr.shape[arr._distaxis]
        out = out.reshape(shp)
Martin Reinecke's avatar
Martin Reinecke committed
418
        out = np.moveaxis(out, 0, arr._distaxis)
Martin Reinecke's avatar
Martin Reinecke committed
419
        return from_global_data(out, distaxis=-1)
Martin Reinecke's avatar
Martin Reinecke committed
420

Martin Reinecke's avatar
Martin Reinecke committed
421
    # real redistribution via Alltoallv
Martin Reinecke's avatar
Martin Reinecke committed
422
    ssz0 = arr._data.size//arr.shape[dist]
Martin Reinecke's avatar
Martin Reinecke committed
423
    ssz = np.empty(ntask, dtype=np.int)
Martin Reinecke's avatar
Martin Reinecke committed
424
425
426
    rszall = arr.size//arr.shape[dist]*_shareSize(arr.shape[dist], ntask, rank)
    rbuf = np.empty(rszall, dtype=arr.dtype)
    rsz0 = rszall//arr.shape[arr._distaxis]
Martin Reinecke's avatar
Martin Reinecke committed
427
    rsz = np.empty(ntask, dtype=np.int)
Martin Reinecke's avatar
Martin Reinecke committed
428
429
430
431
432
433
434
435
436
437
438
439
    if dist == 0:  # shortcut possible
        sbuf = np.ascontiguousarray(arr._data)
        for i in range(ntask):
            lo, hi = _shareRange(arr.shape[dist], ntask, i)
            ssz[i] = ssz0*(hi-lo)
            rsz[i] = rsz0*_shareSize(arr.shape[arr._distaxis], ntask, i)
    else:
        sbuf = np.empty(arr._data.size, dtype=arr.dtype)
        sslice = [slice(None)]*arr._data.ndim
        ofs = 0
        for i in range(ntask):
            lo, hi = _shareRange(arr.shape[dist], ntask, i)
Martin Reinecke's avatar
Martin Reinecke committed
440
            sslice[dist] = slice(lo, hi)
Martin Reinecke's avatar
Martin Reinecke committed
441
442
443
444
445
446
            ssz[i] = ssz0*(hi-lo)
            sbuf[ofs:ofs+ssz[i]] = arr._data[sslice].flat
            ofs += ssz[i]
            rsz[i] = rsz0*_shareSize(arr.shape[arr._distaxis], ntask, i)
    ssz *= arr._data.itemsize
    rsz *= arr._data.itemsize
Martin Reinecke's avatar
Martin Reinecke committed
447
448
    sdisp = np.append(0, np.cumsum(ssz[:-1]))
    rdisp = np.append(0, np.cumsum(rsz[:-1]))
Martin Reinecke's avatar
Martin Reinecke committed
449
450
    s_msg = [sbuf, (ssz, sdisp), MPI.BYTE]
    r_msg = [rbuf, (rsz, rdisp), MPI.BYTE]
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
451
    _comm.Alltoallv(s_msg, r_msg)
Martin Reinecke's avatar
Martin Reinecke committed
452
    del sbuf  # free memory
Martin Reinecke's avatar
Martin Reinecke committed
453
454
455
456
457
458
459
460
461
    if arr._distaxis == 0:
        rbuf = rbuf.reshape(local_shape(arr.shape, dist))
        arrnew = from_local_data(arr.shape, rbuf, distaxis=dist)
    else:
        arrnew = empty(arr.shape, dtype=arr.dtype, distaxis=dist)
        rslice = [slice(None)]*arr._data.ndim
        ofs = 0
        for i in range(ntask):
            lo, hi = _shareRange(arr.shape[arr._distaxis], ntask, i)
Martin Reinecke's avatar
Martin Reinecke committed
462
            rslice[arr._distaxis] = slice(lo, hi)
Martin Reinecke's avatar
Martin Reinecke committed
463
464
465
466
            sz = rsz[i]//arr._data.itemsize
            arrnew._data[rslice].flat = rbuf[ofs:ofs+sz]
            ofs += sz
    return arrnew
Martin Reinecke's avatar
Martin Reinecke committed
467
468


Martin Reinecke's avatar
Martin Reinecke committed
469
470
def transpose(arr):
    if len(arr.shape) != 2 or arr._distaxis != 0:
Martin Reinecke's avatar
Martin Reinecke committed
471
        raise ValueError("bad input")
Martin Reinecke's avatar
Martin Reinecke committed
472
473
474
475
476
477
478
479
480
481
482
    ssz0 = arr._data.size//arr.shape[1]
    ssz = np.empty(ntask, dtype=np.int)
    rszall = arr.size//arr.shape[1]*_shareSize(arr.shape[1], ntask, rank)
    rbuf = np.empty(rszall, dtype=arr.dtype)
    rsz0 = rszall//arr.shape[0]
    rsz = np.empty(ntask, dtype=np.int)
    sbuf = np.empty(arr._data.size, dtype=arr.dtype)
    ofs = 0
    for i in range(ntask):
        lo, hi = _shareRange(arr.shape[1], ntask, i)
        ssz[i] = ssz0*(hi-lo)
Martin Reinecke's avatar
Martin Reinecke committed
483
        sbuf[ofs:ofs+ssz[i]] = arr._data[:, lo:hi].flat
Martin Reinecke's avatar
Martin Reinecke committed
484
485
486
487
488
489
490
491
492
493
494
495
        ofs += ssz[i]
        rsz[i] = rsz0*_shareSize(arr.shape[0], ntask, i)
    ssz *= arr._data.itemsize
    rsz *= arr._data.itemsize
    sdisp = np.append(0, np.cumsum(ssz[:-1]))
    rdisp = np.append(0, np.cumsum(rsz[:-1]))
    s_msg = [sbuf, (ssz, sdisp), MPI.BYTE]
    r_msg = [rbuf, (rsz, rdisp), MPI.BYTE]
    _comm.Alltoallv(s_msg, r_msg)
    del sbuf  # free memory
    arrnew = empty((arr.shape[1], arr.shape[0]), dtype=arr.dtype, distaxis=0)
    ofs = 0
Martin Reinecke's avatar
Martin Reinecke committed
496
    sz2 = _shareSize(arr.shape[1], ntask, rank)
Martin Reinecke's avatar
Martin Reinecke committed
497
498
499
    for i in range(ntask):
        lo, hi = _shareRange(arr.shape[0], ntask, i)
        sz = rsz[i]//arr._data.itemsize
Martin Reinecke's avatar
Martin Reinecke committed
500
        arrnew._data[:, lo:hi] = rbuf[ofs:ofs+sz].reshape(hi-lo, sz2).T
Martin Reinecke's avatar
Martin Reinecke committed
501
502
503
504
        ofs += sz
    return arrnew


Martin Reinecke's avatar
Martin Reinecke committed
505
506
def default_distaxis():
    return 0