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
249
250
251
252
253
254
255
256
257
258
259
260
|
# -*- coding: utf-8 -*-
#cython: embedsignature=True, language_level=3
#cython: boundscheck=False, wraparound=False, cdivision=True, initializedcheck=False,
## This is for developping:
##cython: profile=True, warn.undeclared=True, warn.unused=True, warn.unused_result=False, warn.unused_arg=True
#
# Project: Azimuthal integration
# https://github.com/silx-kit/pyFAI
#
# Copyright (C) 2015-2020 European Synchrotron Radiation Facility, France
#
# Principal author: Jérôme Kieffer (Jerome.Kieffer@ESRF.eu)
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
# THE SOFTWARE.
"""Common Look-Up table/CSR object creation tools and conversion"""
__author__ = "Jerome Kieffer"
__contact__ = "Jerome.kieffer@esrf.fr"
__date__ = "26/06/2020"
__status__ = "stable"
__license__ = "MIT"
include "regrid_common.pxi"
include "CSR_common.pxi"
include "LUT_common.pxi"
def LUT_to_CSR(lut):
"""Conversion between sparse matrix representations
:param lut: Look-up table as 2D array of (int idx, float coef)
:return: the same matrix as CSR representation
:rtype: 3-tuple of numpy array (data, indices, indptr)
"""
cdef:
int nrow, ncol
ncol = lut.shape[1]
nrow = lut.shape[0]
cdef:
lut_t[:, ::1] lut_ = numpy.ascontiguousarray(lut, lut_d)
data_t[::1] data = numpy.zeros(nrow * ncol, data_d)
index_t[::1] indices = numpy.zeros(nrow * ncol, index_d)
index_t[::1] indptr = numpy.zeros(nrow + 1, index_d)
int i, j, nelt
lut_t point
with nogil:
nelt = 0
for i in range(nrow):
indptr[i] = nelt
for j in range(ncol):
point = lut_[i, j]
if point.coef <= 0.0:
continue
else:
data[nelt] = point.coef
indices[nelt] = point.idx
nelt += 1
indptr[nrow] = nelt
return numpy.asarray(data[:nelt]), numpy.asarray(indices[:nelt]), numpy.asarray(indptr)
def CSR_to_LUT(data, indices, indptr):
"""Conversion between sparse matrix representations
:param data: coef of the sparse matrix as 1D array
:param indices: index of the col position in input array as 1D array
:param indptr: index of the start of the row in the indices array
:return: the same matrix as LUT representation
:rtype: record array of (int idx, float coef)
"""
cdef:
int nrow, ncol
nrow = indptr.shape[0] - 1
ncol = (indptr[1:] - indptr[:nrow]).max()
assert nrow > 0, "nrow >0"
assert ncol > 0, "ncol >0"
cdef:
data_t[::1] data_ = numpy.ascontiguousarray(data, dtype=data_d)
index_t[::1] indices_ = numpy.ascontiguousarray(indices, dtype=index_d)
index_t[::1] indptr_ = numpy.ascontiguousarray(indptr, dtype=index_d)
lut_t[:, ::1] lut = numpy.zeros((nrow, ncol), dtype=lut_d)
lut_t point
int i, j, nelt
float coef
with nogil:
for i in range(nrow):
nelt = 0
for j in range(indptr_[i], indptr_[i + 1]):
coef = data_[j]
if coef <= 0.0:
continue
point.coef = coef
point.idx = indices_[j]
lut[i, nelt] = point
nelt += 1
return numpy.asarray(lut)
cdef class Vector:
"""Variable size vector"""
cdef:
readonly int size, allocated
data_t[::1] coef
index_t[::1] idx
def __cinit__(self, int min_size=4):
self.allocated = min_size
self.coef = numpy.empty(self.allocated, dtype=data_d)
self.idx = numpy.empty(self.allocated, dtype=index_d)
self.size = 0
def __dealloc__(self):
self.coef = self.idx = None
def __len__(self):
return self.size
def __repr__(self):
return "Vector of size %i (%i elements allocated)" % (self.size, self.allocated)
@property
def nbytes(self):
"Calculate the actual size of the object (in bytes)"
return (self.allocated + 1) * 8
def get_data(self):
return numpy.asarray(self.idx[:self.size]), numpy.asarray(self.coef[:self.size])
cdef inline void _append(self, int idx, float coef):
cdef:
int pos, new_allocated
index_t[::1] newidx
data_t[::1] newcoef
pos = self.size
self.size = pos + 1
if pos >= self.allocated - 1:
new_allocated = self.allocated * 2
newcoef = numpy.empty(new_allocated, dtype=data_d)
newcoef[:pos] = self.coef[:pos]
self.coef = newcoef
newidx = numpy.empty(new_allocated, dtype=index_d)
newidx[:pos] = self.idx[:pos]
self.idx = newidx
self.allocated = new_allocated
self.coef[pos] = coef
self.idx[pos] = idx
def append(self, idx, coef):
"Python implementation of _append in cython"
self._append(<int> idx, <float> coef)
cdef class ArrayBuilder:
cdef:
readonly int size
Vector[:] lines
def __cinit__(self, int nlines, min_size=4):
cdef int i
self.size = nlines
nullarray = numpy.array([None] * nlines)
self.lines = nullarray
for i in range(nlines):
self.lines[i] = Vector(min_size=min_size)
def __dealloc__(self):
cdef int i
for i in range(self.size):
self.lines[i] = None
self.lines = None
def __len__(self):
return self.size
def __repr__(self):
cdef int i, max_line = 0
for i in range(self.size):
max_line = max(max_line, self.lines[i].size)
return "ArrayBuilder of %i lines, the longest is %i" % (self.size, max_line)
@property
def nbytes(self):
"Calculate the actual size of the object (in bytes)"
cdef int i, sum = 0
for i in range(self.size):
sum += self.lines[i].nbytes
return sum
cdef inline void _append(self, int line, int col, float value):
cdef:
Vector vector
vector = self.lines[line]
vector._append(col, value)
def append(self, line, col, value):
'Python wrapper for _append in cython'
self._append(<int> line, <int> col, <float> value)
def as_LUT(self):
cdef:
int i, j, max_size = 0
index_t[::1] local_idx
data_t[:] local_coef
lut_t[:, :] lut
Vector vector
for i in range(len(self.lines)):
if len(self.lines[i]) > max_size:
max_size = len(self.lines[i])
lut = numpy.zeros((len(self.lines), max_size), dtype=lut_d)
for i in range(len(self.lines)):
vector = self.lines[i]
local_idx, local_coef = vector.get_data()
for j in range(len(vector)):
lut[i, j] = lut_t(local_idx[j], local_coef[j])
return numpy.asarray(lut, dtype=lut_d)
def as_CSR(self):
cdef:
int i, start, end, total_size = 0
Vector vector
index_t[:] idptr, idx, local_idx
data_t[:] coef, local_coef
idptr = numpy.zeros(len(self.lines) + 1, dtype=index_d)
for i in range(len(self.lines)):
total_size += len(self.lines[i])
idptr[i + 1] = total_size
coef = numpy.zeros(total_size, dtype=numpy.float32)
idx = numpy.zeros(total_size, dtype=numpy.int32)
for i in range(len(self.lines)):
vector = self.lines[i]
local_idx, local_coef = vector.get_data()
start = idptr[i]
end = start + len(vector)
idx[start:end] = local_idx
coef[start:end] = local_coef
return numpy.asarray(idptr), numpy.asarray(idx), numpy.asarray(coef)
|
