distributed_do.py 12.9 KB
Newer Older
1
2
3
4
import numpy as np
from .random import Random
from mpi4py import MPI

Martin Reinecke's avatar
Martin Reinecke committed
5
6
7
8
9
10
11
__all__ = ["ntask", "rank", "master", "local_shape", "data_object", "full",
           "empty", "zeros", "ones", "empty_like", "vdot", "abs", "exp",
           "log", "sqrt", "bincount", "from_object", "from_random",
           "local_data", "ibegin", "np_allreduce_sum", "distaxis",
           "from_local_data", "from_global_data", "to_global_data",
           "redistribute", "default_distaxis"]

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
12
13
14
15
_comm = MPI.COMM_WORLD
ntask = _comm.Get_size()
rank = _comm.Get_rank()
master = rank==0
16
17


Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
18
def _shareSize(nwork, nshares, myshare):
19
20
    nbase = nwork//nshares
    return nbase if myshare>=nwork%nshares else nbase+1
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
21
22

def _shareRange(nwork, nshares, myshare):
Martin Reinecke's avatar
Martin Reinecke committed
23
24
25
26
27
    nbase = nwork//nshares;
    additional = nwork%nshares;
    lo = myshare*nbase + min(myshare, additional)
    hi = lo+nbase+ (1 if myshare<additional else 0)
    return lo,hi
28

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
29
def local_shape(shape, distaxis):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
30
31
    if len(shape)==0:
        distaxis = -1
32
33
34
    if distaxis==-1:
        return shape
    shape2=list(shape)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
35
    shape2[distaxis]=_shareSize(shape[distaxis],ntask,rank)
36
37
38
39
40
    return tuple(shape2)

class data_object(object):
    def __init__(self, shape, data, distaxis):
        """Must not be called directly by users"""
Martin Reinecke's avatar
Martin Reinecke committed
41
        self._shape = tuple(shape)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
42
43
        if len(self._shape)==0:
            distaxis = -1
44
        self._distaxis = distaxis
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
45
        lshape = local_shape(self._shape, self._distaxis)
46
47
        self._data = data

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
48
    def _sanity_checks(self):
49
50
51
52
53
        # check whether the distaxis is consistent
        if self._distaxis<-1 or self._distaxis>=len(self._shape):
            raise ValueError
        itmp=np.array(self._distaxis)
        otmp=np.empty(ntask,dtype=np.int)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
54
        _comm.Allgather(itmp,otmp)
55
56
57
58
        if np.any(otmp!=self._distaxis):
            raise ValueError
        # check whether the global shape is consistent
        itmp=np.array(self._shape)
Martin Reinecke's avatar
Martin Reinecke committed
59
        otmp=np.empty((ntask,len(self._shape)),dtype=np.int)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
60
        _comm.Allgather(itmp,otmp)
61
        for i in range(ntask):
Martin Reinecke's avatar
Martin Reinecke committed
62
            if np.any(otmp[i,:]!=self._shape):
63
64
65
66
67
68
69
                raise ValueError
        # check shape of local data
        if self._distaxis<0:
            if self._data.shape!=self._shape:
                raise ValueError
        else:
            itmp=np.array(self._shape)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
70
            itmp[self._distaxis] = _shareSize(self._shape[self._distaxis],ntask,rank)
Martin Reinecke's avatar
Martin Reinecke committed
71
            if np.any(self._data.shape!=itmp):
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
                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
88
        return data_object(self._shape, self._data.real, self._distaxis)
89
90
91

    @property
    def imag(self):
Martin Reinecke's avatar
Martin Reinecke committed
92
        return data_object(self._shape, self._data.imag, self._distaxis)
93

Martin Reinecke's avatar
Martin Reinecke committed
94
    def _contraction_helper(self, op, mpiop, axis):
95
96
97
98
        if axis is not None:
            if len(axis)==len(self._data.shape):
                axis = None
        if axis is None:
Martin Reinecke's avatar
Martin Reinecke committed
99
100
            res = np.array(getattr(self._data, op)())
            if (self._distaxis==-1):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
101
102
                return res[()]
            res2 = np.empty((),dtype=res.dtype)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
103
            _comm.Allreduce(res,res2,mpiop)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
104
            return res2[()]
105
106

        if self._distaxis in axis:
Martin Reinecke's avatar
Martin Reinecke committed
107
108
            res = getattr(self._data, op)(axis=axis)
            res2 = np.empty_like(res)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
109
            _comm.Allreduce(res,res2,mpiop)
Martin Reinecke's avatar
Martin Reinecke committed
110
            return from_global_data(res2, distaxis=0)
111
        else:
Martin Reinecke's avatar
Martin Reinecke committed
112
            # perform the contraction on the local data
Martin Reinecke's avatar
Martin Reinecke committed
113
114
115
116
117
118
119
120
121
122
            res = getattr(self._data, op)(axis=axis)
            if self._distaxis == -1:
                return from_global_data(res,distaxis=0)
            shp = list(res.shape)
            shift=0
            for ax in axis:
                if ax<self._distaxis:
                    shift+=1
            shp[self._distaxis-shift] = self.shape[self._distaxis]
            return from_local_data(shp, res, self._distaxis-shift)
123
124
125
126
127
128
129
130
131

        # check if the result is scalar or if a result_field must be constr.
        if np.isscalar(data):
            return data
        else:
            return data_object(data)

    def sum(self, axis=None):
        return self._contraction_helper("sum", MPI.SUM, axis)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
132
133
134
135
    def min(self, axis=None):
        return self._contraction_helper("min", MPI.MIN, axis)
    def max(self, axis=None):
        return self._contraction_helper("max", MPI.MAX, axis)
136

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
137
138
139
140
141
142
143
144
    # FIXME: to be improved!
    def mean(self):
        return self.sum()/self.size
    def std(self):
        return np.sqrt(self.var())
    def var(self):
        return (abs(self-self.mean())**2).mean()

145
    def _binary_helper(self, other, op):
Martin Reinecke's avatar
Martin Reinecke committed
146
        a = self
147
        if isinstance(other, data_object):
Martin Reinecke's avatar
Martin Reinecke committed
148
            b = other
149
150
151
152
            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
153
154
            a = a._data
            b = b._data
155
        else:
Martin Reinecke's avatar
Martin Reinecke committed
156
            a = a._data
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
            b = other

        tval = getattr(a, op)(b)
        return self if tval is a else data_object(self._shape, tval, self._distaxis)

    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
195
196
197
    def __idiv__(self, other):
        return self._binary_helper(other, op='__idiv__')

198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
    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
fixes    
Martin Reinecke committed
220
        return data_object(self._shape,-self._data,self._distaxis)
221
222

    def __abs__(self):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
223
        return data_object(self._shape,np.abs(self._data),self._distaxis)
224
225
226
227
228
229
230
231

    def all(self):
        return self._data.all()

    def any(self):
        return self._data.any()


Martin Reinecke's avatar
Martin Reinecke committed
232
def full(shape, fill_value, dtype=None, distaxis=0):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
233
    return data_object(shape, np.full(local_shape(shape, distaxis), fill_value, dtype), distaxis)
234
235


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
236
def empty(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
237
    return data_object(shape, np.empty(local_shape(shape, distaxis), dtype), distaxis)
238
239


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
240
def zeros(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
241
    return data_object(shape, np.zeros(local_shape(shape, distaxis), dtype), distaxis)
242
243


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
244
def ones(shape, dtype=None, distaxis=0):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
245
    return data_object(shape, np.ones(local_shape(shape, distaxis), dtype), distaxis)
246
247
248
249
250
251
252


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
253
254
    tmp = np.array(np.vdot(a._data, b._data))
    res = np.empty((),dtype=tmp.dtype)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
255
    _comm.Allreduce(tmp,res,MPI.SUM)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
256
    return res[()]
257
258
259
260
261
262
263


def _math_helper(x, function, out):
    if out is not None:
        function(x._data, out=out._data)
        return out
    else:
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
264
        return data_object(x.shape,function(x._data),x._distaxis)
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290


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)


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


def bincount(x, weights=None, minlength=None):
    if weights is not None:
        weights = weights._data
    res = np.bincount(x._data, weights, minlength)
    return data_object(res)


def from_object(object, dtype=None, copy=True):
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
291
    return data_object(object._shape, np.array(object._data, dtype=dtype, copy=copy), distaxis=object._distaxis)
292
293


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
294
def from_random(random_type, shape, dtype=np.float64, distaxis=0, **kwargs):
295
    generator_function = getattr(Random, random_type)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
296
297
298
    #lshape = local_shape(shape, distaxis)
    #return data_object(shape, generator_function(dtype=dtype, shape=lshape, **kwargs), distaxis=distaxis)
    return from_global_data(generator_function(dtype=dtype, shape=shape, **kwargs), distaxis=distaxis)
299

Martin Reinecke's avatar
Martin Reinecke committed
300
301
302
303
def local_data(arr):
    return arr._data


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
304
305
def ibegin(arr):
    res = [0] * arr._data.ndim
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
306
    res[arr._distaxis] = _shareRange(arr._shape[arr._distaxis],ntask,rank)[0]
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
307
    return tuple(res)
Martin Reinecke's avatar
Martin Reinecke committed
308
309


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
310
311
def np_allreduce_sum(arr):
    res = np.empty_like(arr)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
312
    _comm.Allreduce(arr,res,MPI.SUM)
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
313
    return res
Martin Reinecke's avatar
Martin Reinecke committed
314
315
316
317
318
319
320
321
322
323
324
325
326


def distaxis(arr):
    return arr._distaxis


def from_local_data (shape, arr, distaxis):
    return data_object(shape, arr, distaxis)


def from_global_data (arr, distaxis=0):
    if distaxis==-1:
        return data_object(arr.shape, arr, distaxis)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
327
    lo, hi = _shareRange(arr.shape[distaxis],ntask,rank)
Martin Reinecke's avatar
Martin Reinecke committed
328
329
330
331
332
    sl = [slice(None)]*len(arr.shape)
    sl[distaxis]=slice(lo,hi)
    return data_object(arr.shape, arr[sl], distaxis)


Martin Reinecke's avatar
fixes    
Martin Reinecke committed
333
334
335
336
337
338
339
def to_global_data (arr):
    if arr._distaxis==-1:
        return arr._data
    tmp = redistribute(arr, dist=-1)
    return tmp._data


Martin Reinecke's avatar
Martin Reinecke committed
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
def redistribute (arr, dist=None, nodist=None):
    if dist is not None:
        if nodist is not None:
            raise ValueError
        if dist==arr._distaxis:
            return arr
    else:
        if nodist is None:
            raise ValueError
        if arr._distaxis not in nodist:
            return arr
        dist=-1
        for i in range(len(arr.shape)):
            if i not in nodist:
                dist=i
                break
Martin Reinecke's avatar
Martin Reinecke committed
356

Martin Reinecke's avatar
Martin Reinecke committed
357
358
359
360
361
362
363
    if arr._distaxis==-1:  # just pick the proper subset
        return from_global_data(arr._data, dist)
    if dist==-1: # gather data
        tmp = np.moveaxis(arr._data, arr._distaxis, 0)
        slabsize=np.prod(tmp.shape[1:])*tmp.itemsize
        sz=np.empty(ntask,dtype=np.int)
        for i in range(ntask):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
364
            sz[i]=slabsize*_shareSize(arr.shape[arr._distaxis],ntask,i)
Martin Reinecke's avatar
Martin Reinecke committed
365
366
367
368
369
        disp=np.empty(ntask,dtype=np.int)
        disp[0]=0
        disp[1:]=np.cumsum(sz[:-1])
        tmp=tmp.flatten()
        out = np.empty(arr.size,dtype=arr.dtype)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
370
        _comm.Allgatherv(tmp,[out,sz,disp,MPI.BYTE])
Martin Reinecke's avatar
Martin Reinecke committed
371
372
373
374
        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
375
376
377
        out = np.moveaxis(out, 0, arr._distaxis)
        return from_global_data (out, distaxis=-1)
    # real redistribution via Alltoallv
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
378
    # temporary slow, but simple solution for comparison purposes:
379
    #return redistribute(redistribute(arr,dist=-1),dist=dist)
Martin Reinecke's avatar
Martin Reinecke committed
380

Martin Reinecke's avatar
Martin Reinecke committed
381
382
383
384
385
386
    tmp = np.moveaxis(arr._data, (dist, arr._distaxis), (0, 1))
    tshape = tmp.shape
    slabsize=np.prod(tmp.shape[2:])*tmp.itemsize
    ssz=np.empty(ntask,dtype=np.int)
    rsz=np.empty(ntask,dtype=np.int)
    for i in range(ntask):
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
387
388
        ssz[i]=_shareSize(arr.shape[dist],ntask,i)*tmp.shape[1]*slabsize
        rsz[i]=_shareSize(arr.shape[dist],ntask,rank)*_shareSize(arr.shape[arr._distaxis],ntask,i)*slabsize
Martin Reinecke's avatar
Martin Reinecke committed
389
390
391
392
393
394
395
    sdisp=np.empty(ntask,dtype=np.int)
    rdisp=np.empty(ntask,dtype=np.int)
    sdisp[0]=0
    rdisp[0]=0
    sdisp[1:]=np.cumsum(ssz[:-1])
    rdisp[1:]=np.cumsum(rsz[:-1])
    tmp=tmp.flatten()
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
396
    out = np.empty(np.prod(local_shape(arr.shape,dist)),dtype=arr.dtype)
Martin Reinecke's avatar
Martin Reinecke committed
397
398
    s_msg = [tmp, (ssz, sdisp), MPI.BYTE]
    r_msg = [out, (rsz, rdisp), MPI.BYTE]
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
399
400
    _comm.Alltoallv(s_msg, r_msg)
    out2 = np.empty([_shareSize(arr.shape[dist],ntask,rank), arr.shape[arr._distaxis]] +list(tshape[2:]), dtype=arr.dtype)
401
402
403
    ofs=0
    for i in range(ntask):
        lsize = rsz[i]//tmp.itemsize
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
404
405
        lo,hi = _shareRange(arr.shape[arr._distaxis],ntask,i)
        out2[slice(None),slice(lo,hi)] = out[ofs:ofs+lsize].reshape([_shareSize(arr.shape[dist],ntask,rank),_shareSize(arr.shape[arr._distaxis],ntask,i)]+list(tshape[2:]))
406
        ofs += lsize
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
407
    new_shape = [_shareSize(arr.shape[dist],ntask,rank), arr.shape[arr._distaxis]] +list(tshape[2:])
408
409
410
    out2=out2.reshape(new_shape)
    out2 = np.moveaxis(out2, (0, 1), (dist, arr._distaxis))
    return from_local_data (arr.shape, out2, dist)
Martin Reinecke's avatar
Martin Reinecke committed
411
412
413
414


def default_distaxis():
    return 0