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
|
import functools
import six
from six.moves import reduce
from heapq import nsmallest
from operator import itemgetter, mul
import numpy
from .dtypes import dtype_to_ctype, _fill_dtype_registry
from .gpuarray import GpuArray
_fill_dtype_registry()
def as_argument(obj, name):
if isinstance(obj, GpuArray):
return ArrayArg(obj.dtype, name)
else:
return ScalarArg(numpy.asarray(obj).dtype, name)
class Argument(object):
def __init__(self, dtype, name):
self.dtype = dtype
self.name = name
def ctype(self):
return dtype_to_ctype(self.dtype)
def __hash__(self):
return hash(type(self)) ^ hash(self.dtype) ^ hash(self.name)
def __eq__(self, other):
return (type(self) == type(other) and
self.dtype == other.dtype and
self.name == other.name)
class ArrayArg(Argument):
def decltype(self):
return "GLOBAL_MEM {} *".format(self.ctype())
def expr(self):
return "{}[i]".format(self.name)
def isarray(self):
return True
def spec(self):
return GpuArray
class ScalarArg(Argument):
def decltype(self):
return self.ctype()
def expr(self):
return self.name
def isarray(self):
return False
def spec(self):
return self.dtype
def check_args(args, collapse=False, broadcast=False):
"""
Returns the properties of arguments and checks if they all match
(are all the same shape)
If `collapse` is True dimension collapsing will be performed.
If `collapse` is False dimension collapsing will not be performed.
If `broadcast` is True array broadcasting will be performed which
means that dimensions which are of size 1 in some arrays but not
others will be repeated to match the size of the other arrays.
If `broadcast` is False no broadcasting takes place.
"""
# For compatibility with old collapse=None option
if collapse is None:
collapse = True
strs = []
offsets = []
dims = None
for arg in args:
if isinstance(arg, GpuArray):
strs.append(arg.strides)
offsets.append(arg.offset)
if dims is None:
n, nd, dims = arg.size, arg.ndim, arg.shape
else:
if arg.ndim != nd:
raise ValueError("Array order differs")
if not broadcast and arg.shape != dims:
raise ValueError("Array shape differs")
else:
strs.append(None)
offsets.append(None)
if dims is None:
raise TypeError("No arrays in kernel arguments, "
"something is wrong")
tdims = dims
if broadcast or collapse:
# make the strides and dims editable
dims = list(dims)
strs = [list(str) if str is not None else str for str in strs]
if broadcast:
# Set strides to 0s when needed.
# Get the full shape in dims (no ones unless all arrays have it).
if 1 in dims:
for i, ary in enumerate(args):
if strs[i] is None:
continue
shp = ary.shape
for i, d in enumerate(shp):
if dims[i] != d and dims[i] == 1:
dims[i] = d
n *= d
tdims = tuple(dims)
for i, ary in enumerate(args):
if strs[i] is None:
continue
shp = ary.shape
if tdims != shp:
for j, d in enumerate(shp):
if dims[j] != d:
# Might want to add a per-dimension enable mechanism
if d == 1:
strs[i][j] = 0
else:
raise ValueError("Array shape differs")
if collapse and nd > 1:
# remove dimensions that are of size 1
for i in range(nd - 1, -1, -1):
if nd > 1 and dims[i] == 1:
del dims[i]
for str in strs:
if str is not None:
del str[i]
nd -= 1
# collapse contiguous dimensions
for i in range(nd - 1, 0, -1):
if all(str is None or str[i] * dims[i] == str[i - 1]
for str in strs):
dims[i - 1] *= dims[i]
del dims[i]
for str in strs:
if str is not None:
str[i - 1] = str[i]
del str[i]
nd -= 1
if broadcast or collapse:
# re-wrap dims and tuples
dims = tuple(dims)
strs = [tuple(str) if str is not None else None for str in strs]
return n, nd, dims, tuple(strs), tuple(offsets)
def lru_cache(maxsize=20):
def decorating_function(user_function):
cache = {}
last_use = {}
time = [0] # workaround for Python 2, which doesn't have nonlocal
@functools.wraps(user_function)
def wrapper(*key):
time[0] += 1
try:
result = cache[key]
wrapper.hits += 1
except KeyError:
result = user_function(*key)
cache[key] = result
wrapper.misses += 1
# purge least recently used cache entries
if len(cache) > wrapper.maxsize:
for key0, _ in nsmallest(wrapper.maxsize // 10,
six.iteritems(last_use),
key=itemgetter(1)):
del cache[key0], last_use[key0]
last_use[key] = time[0]
return result
def clear():
cache.clear()
last_use.clear()
wrapper.hits = wrapper.misses = 0
time[0] = 0
@functools.wraps(user_function)
def get(*key):
result = cache[key]
time[0] += 1
last_use[key] = time[0]
wrapper.hits += 1
return result
wrapper.hits = wrapper.misses = 0
wrapper.maxsize = maxsize
wrapper.clear = clear
wrapper.get = get
return wrapper
return decorating_function
def prod(iterable):
return reduce(mul, iterable, 1)
|
