1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
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
176
177
178
179
180
181
182
183
184
185
186
187
188
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
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
|
""" This is a mini-tutorial on iterators, strides, and
memory layout. It assumes you are familiar with the terms, see
http://docs.scipy.org/doc/numpy/reference/arrays.ndarray.html
for a more gentle introduction.
Given an array x: x.shape == [5,6], where each element occupies one byte
At which byte in x.data does the item x[3,4] begin?
if x.strides==[1,5]:
pData = x.pData + (x.start + 3*1 + 4*5)*sizeof(x.pData[0])
pData = x.pData + (x.start + 23) * sizeof(x.pData[0])
so the offset of the element is 23 elements after the first
What is the next element in x after coordinates [3,4]?
if x.order =='C':
next == [3,5] => offset is 28
if x.order =='F':
next == [4,4] => offset is 24
so for the strides [1,5] x is 'F' contiguous
likewise, for the strides [6,1] x would be 'C' contiguous.
Iterators have an internal representation of the current coordinates
(indices), the array, strides, and backstrides. A short digression to
explain backstrides: what is the coordinate and offset after [3,5] in
the example above?
if x.order == 'C':
next == [4,0] => offset is 4
if x.order == 'F':
next == [4,5] => offset is 25
Note that in 'C' order we stepped BACKWARDS 24 while 'overflowing' a
shape dimension
which is back 25 and forward 1,
which is x.strides[1] * (x.shape[1] - 1) + x.strides[0]
so if we precalculate the overflow backstride as
[x.strides[i] * (x.shape[i] - 1) for i in range(len(x.shape))]
we can do only addition while iterating
All the calculations happen in next()
"""
from rpython.rlib import jit
from pypy.module.micronumpy import support, constants as NPY
from pypy.module.micronumpy.base import W_NDimArray
class PureShapeIter(object):
def __init__(self, shape, idx_w):
self.shape = shape
self.shapelen = len(shape)
self.indexes = [0] * len(shape)
self._done = False
self.idx_w_i = [None] * len(idx_w)
self.idx_w_s = [None] * len(idx_w)
for i, w_idx in enumerate(idx_w):
if isinstance(w_idx, W_NDimArray):
self.idx_w_i[i], self.idx_w_s[i] = w_idx.create_iter(shape)
def done(self):
return self._done
@jit.unroll_safe
def next(self):
for i, idx_w_i in enumerate(self.idx_w_i):
if idx_w_i is not None:
self.idx_w_s[i] = idx_w_i.next(self.idx_w_s[i])
for i in range(self.shapelen - 1, -1, -1):
if self.indexes[i] < self.shape[i] - 1:
self.indexes[i] += 1
break
else:
self.indexes[i] = 0
else:
self._done = True
@jit.unroll_safe
def get_index(self, space, shapelen):
return [space.newint(self.indexes[i]) for i in range(shapelen)]
class IterState(object):
_immutable_fields_ = ['iterator', '_indices']
def __init__(self, iterator, index, indices, offset):
self.iterator = iterator
self.index = index
self._indices = indices
self.offset = offset
def same(self, other):
if self.offset == other.offset and \
self.index == other.index and \
self._indices == other._indices:
return self.iterator.same_shape(other.iterator)
return False
class ArrayIter(object):
_immutable_fields_ = ['contiguous', 'array', 'size', 'ndim_m1', 'shape_m1[*]',
'strides[*]', 'backstrides[*]', 'factors[*]',
'track_index']
track_index = True
@jit.unroll_safe
def __init__(self, array, size, shape, strides, backstrides):
assert len(shape) == len(strides) == len(backstrides)
self.contiguous = (array.flags & NPY.ARRAY_C_CONTIGUOUS and
array.shape == shape and array.strides == strides)
self.array = array
self.size = size
self.ndim_m1 = len(shape) - 1
#
self.shape_m1 = [s - 1 for s in shape]
self.strides = strides
self.backstrides = backstrides
ndim = len(shape)
factors = [0] * ndim
for i in xrange(ndim):
if i == 0:
factors[ndim-1] = 1
else:
factors[ndim-i-1] = factors[ndim-i] * shape[ndim-i]
self.factors = factors
def same_shape(self, other):
""" Iterating over the same element """
if not self.contiguous or not other.contiguous:
return False
return (self.contiguous == other.contiguous and
self.array.dtype is other.array.dtype and
self.shape_m1 == other.shape_m1 and
self.strides == other.strides and
self.backstrides == other.backstrides and
self.factors == other.factors)
@jit.unroll_safe
def reset(self, state=None, mutate=False):
index = 0
if state is None:
indices = [0] * len(self.shape_m1)
else:
assert state.iterator is self
indices = state._indices
for i in xrange(self.ndim_m1, -1, -1):
indices[i] = 0
offset = self.array.start
if not mutate:
return IterState(self, index, indices, offset)
state.index = index
state.offset = offset
@jit.unroll_safe
def next(self, state, mutate=False):
assert state.iterator is self
index = state.index
if self.track_index:
index += 1
indices = state._indices
offset = state.offset
if self.contiguous:
elsize = self.array.dtype.elsize
jit.promote(elsize)
offset += elsize
elif self.ndim_m1 == 0:
stride = self.strides[0]
jit.promote(stride)
offset += stride
else:
for i in xrange(self.ndim_m1, -1, -1):
idx = indices[i]
if idx < self.shape_m1[i]:
indices[i] = idx + 1
offset += self.strides[i]
break
else:
indices[i] = 0
offset -= self.backstrides[i]
if not mutate:
return IterState(self, index, indices, offset)
state.index = index
state.offset = offset
@jit.unroll_safe
def goto(self, index):
offset = self.array.start
if self.contiguous:
offset += index * self.array.dtype.elsize
elif self.ndim_m1 == 0:
offset += index * self.strides[0]
else:
current = index
for i in xrange(len(self.shape_m1)):
offset += (current / self.factors[i]) * self.strides[i]
current %= self.factors[i]
return IterState(self, index, None, offset)
@jit.unroll_safe
def indices(self, state):
assert state.iterator is self
assert self.track_index
indices = state._indices
if not (self.contiguous or self.ndim_m1 == 0):
return indices
current = state.index
for i in xrange(len(self.shape_m1)):
if self.factors[i] != 0:
indices[i] = current / self.factors[i]
current %= self.factors[i]
else:
indices[i] = 0
return indices
def done(self, state):
assert state.iterator is self
assert self.track_index
return state.index >= self.size
def getitem(self, state):
# assert state.iterator is self
return self.array.getitem(state.offset)
def getitem_bool(self, state):
assert state.iterator is self
return self.array.getitem_bool(state.offset)
def setitem(self, state, elem):
assert state.iterator is self
self.array.setitem(state.offset, elem)
def AxisIter(array, shape, axis):
strides = array.get_strides()
backstrides = array.get_backstrides()
if len(shape) == len(strides):
# keepdims = True
strides = strides[:axis] + [0] + strides[axis + 1:]
backstrides = backstrides[:axis] + [0] + backstrides[axis + 1:]
else:
strides = strides[:axis] + [0] + strides[axis:]
backstrides = backstrides[:axis] + [0] + backstrides[axis:]
return ArrayIter(array, support.product(shape), shape, strides, backstrides)
def AllButAxisIter(array, axis):
size = array.get_size()
shape = array.get_shape()[:]
backstrides = array.backstrides[:]
if size:
size /= shape[axis]
shape[axis] = backstrides[axis] = 0
return ArrayIter(array, size, shape, array.strides, backstrides)
|
