-
Notifications
You must be signed in to change notification settings - Fork 1
/
Changes
168 lines (99 loc) · 4.74 KB
/
Changes
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
numpysane (0.41)
* Updated build to mrbuild 1.10
-- Dima Kogan <dkogan@debian.org> Sun, 26 May 2024 17:04:32 -0700
numpysane (0.40)
* numpysane_pywrap: no more nested functions. So the generated code
builds with clang
-- Dima Kogan <dkogan@debian.org> Sat, 23 Dec 2023 15:49:25 -0800
numpysane (0.39)
* numpysane_pywrap: extra broadcasted slices in the output are allowed
-- Dima Kogan <dkogan@debian.org> Wed, 28 Jun 2023 00:03:56 -0700
numpysane (0.38)
* Reworked the python install (pip stuff only)
-- Dima Kogan <dkogan@debian.org> Tue, 20 Jun 2023 12:40:49 -0700
numpysane (0.37)
* C broadcasting logic fix: implicit dimensions of length-1 are added
properly. Prior to this patch the dimension deficient arrays fail
CHECK_CONTIGUOUS()
-- Dima Kogan <dkogan@debian.org> Tue, 25 Apr 2023 00:52:31 -0700
numpysane (0.36)
* Uninteresting housekeeping
-- Dima Kogan <dkogan@debian.org> Sat, 14 Jan 2023 17:52:28 -0800
numpysane (0.35)
* added broadcast_extra_dims()
-- Dima Kogan <dkogan@debian.org> Wed, 14 Jul 2021 02:41:25 -0700
numpysane (0.34)
* glue(np.array(()), x) and glue(x, np.array(())) both work
-- Dima Kogan <dkogan@debian.org> Thu, 27 May 2021 17:03:33 -0700
numpysane (0.33)
* matmult() supports in-place output via the "out" keyword argument
-- Dima Kogan <dkogan@debian.org> Thu, 15 Apr 2021 18:15:46 -0700
numpysane (0.32)
* glue(): minor bug fix
Prior to this fix this would happen:
print( nps.glue(np.array(()),np.arange(5,),
axis=-2). shape )
---> (5,)
This is unintuitive because I glued something along dimension -2, but
the result doesn't even have such a dimension. This patch calls
atleast_dims() right before glue() returns, so that we get a shape
(1,5) instead in this case
-- Dima Kogan <dkogan@debian.org> Thu, 25 Mar 2021 18:52:19 -0700
numpysane (0.31)
* broadcast_define(): better out_kwarg logic. If the output is written
in-place, the inner function doesn't need to return anything. And the
broadcast_define()-wrapped function automatically returns the right
thing
-- Dima Kogan <dkogan@debian.org> Wed, 03 Feb 2021 12:51:54 -0800
numpysane (0.30)
* numpysane_pywrap can find its C templates even if installed via pip
* numpysane.mag() can take a dtype keyword argument
* mag(), inner(), norm2() now use the given dtype for all their
computations, so selecting an appropriate dtype can prevent overflows
-- Dima Kogan <dkogan@debian.org> Mon, 01 Feb 2021 15:31:03 -0800
numpysane (0.29)
* numpysane_pywrap: module docstrings can span multiple lines
-- Dima Kogan <dkogan@debian.org> Tue, 17 Nov 2020 12:09:00 -0800
numpysane (0.28)
* more precise logic for size-0 concatenation
I can once again accumulate arrays from np.array(())
-- Dima Kogan <dkogan@debian.org> Wed, 23 Sep 2020 13:06:08 -0700
numpysane (0.27)
* numpysane_pywrap item__...() macro works with non-trivial arguments
-- Dima Kogan <dkogan@debian.org> Mon, 21 Sep 2020 14:09:17 -0700
numpysane (0.26)
* glue() and cat() handles size-0 arrays better
* numpysane_pywrap: size-0 arrays are always deemed contiguous
-- Dima Kogan <dkogan@debian.org> Sat, 19 Sep 2020 20:13:05 -0700
numpysane (0.25)
* nps.dummy() supports multiple axes given at once. So I can do
something like nps.dummy(x, -2, -2)
* numpysane_pywrap: generated code can use ctype__NAME and item__NAME to
simplify handling of non-contiguous data
-- Dima Kogan <dkogan@debian.org> Sat, 05 Sep 2020 13:52:19 -0700
numpysane (0.24)
* C broadcasting: I can pass strings in the extra, non-broadcastable
arguments
* C broadcasting: added support for precomputed cookies to do as much of
the work as possible outside of the slice loop
-- Dima Kogan <dkogan@debian.org> Fri, 19 Jun 2020 10:55:47 -0700
numpysane (0.23)
* Bug fix: C broadcasting doesn't write to uninitialized memory when
given a size-0 matrix
-- Dima Kogan <dkogan@debian.org> Fri, 12 Jun 2020 19:16:25 -0700
numpysane (0.22)
* broadcast_define() and the generated function checks its arguments for
validity more thoroughly
* outer() doesn't require identically-dimensioned input
* mass rewrite of the documentation
* Added C-level broadcasting
* License change: any version of the LGPL instead of LGPL-3+
-- Dima Kogan <dkogan@debian.org> Sat, 14 Mar 2020 23:40:29 -0700
numpysane (0.20)
* nps.matmult(..., out=out) produces in-place results when one of the
arguments is 1D
-- Dima Kogan <dkogan@debian.org> Sat, 30 Nov 2019 18:20:49 -0800
numpysane (0.19)
* Added mag() convenience function. mag(x) = sqrt(norm2(x))
* Initial support for C-level broadcasting
-- Dima Kogan <dima@secretsauce.net> Thu, 28 Nov 2019 18:50:02 -0800