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
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
|
/*@targets
** $maxopt $autovec baseline
** sse2 avx2
** neon
** vsx2
** vx
**/
#define _UMATHMODULE
#define _MULTIARRAYMODULE
#define NPY_NO_DEPRECATED_API NPY_API_VERSION
#include "simd/simd.h"
#include "loops_utils.h"
#include "loops.h"
// Provides the various *_LOOP macros
#include "fast_loop_macros.h"
/*
*****************************************************************************
** INTEGER LOOPS
*****************************************************************************
*/
/*
* Arithmetic bit shift operations.
*
* Intel hardware masks bit shift values, so large shifts wrap around
* and can produce surprising results. The special handling ensures that
* behavior is independent of compiler or hardware.
* TODO: We could implement consistent behavior for negative shifts,
* which is undefined in C.
*/
#define INT_left_shift_needs_clear_floatstatus
#define UINT_left_shift_needs_clear_floatstatus
/**begin repeat
* #TYPE = BYTE, UBYTE, SHORT, USHORT, INT, UINT,
* LONG, ULONG, LONGLONG, ULONGLONG#
* #type = npy_byte, npy_ubyte, npy_short, npy_ushort, npy_int, npy_uint,
* npy_long, npy_ulong, npy_longlong, npy_ulonglong#
* #ftype = npy_float, npy_float, npy_float, npy_float, npy_double, npy_double,
* npy_double, npy_double, npy_double, npy_double#
* #SIGNED = 1, 0, 1, 0, 1, 0, 1, 0, 1, 0#
* #c = hh,uhh,h,uh,,u,l,ul,ll,ull#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_positive)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = +in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_square)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(data))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in * in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_reciprocal)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(data))
{
UNARY_LOOP_FAST(@type@, @type@, *out = 1.0 / in);
}
/**begin repeat1
* Arithmetic
* #kind = add, subtract, multiply#
* #OP = +, -, *#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
if (IS_BINARY_REDUCE) {
BINARY_REDUCE_LOOP_FAST(@type@, io1 @OP@= in2);
}
else {
BINARY_LOOP_FAST(@type@, @type@, *out = in1 @OP@ in2);
}
}
/**end repeat1**/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_left_shift)
(char **args, npy_intp const *dimensions, npy_intp const *steps,
void *NPY_UNUSED(func))
{
BINARY_LOOP_FAST(@type@, @type@, *out = npy_lshift@c@(in1, in2));
#ifdef @TYPE@_left_shift_needs_clear_floatstatus
// For some reason, our macOS CI sets an "invalid" flag here, but only
// for some types.
npy_clear_floatstatus_barrier((char*)dimensions);
#endif
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_right_shift)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
#ifndef NPY_DO_NOT_OPTIMIZE_@TYPE@_right_shift
BINARY_LOOP_FAST(@type@, @type@, *out = npy_rshift@c@(in1, in2));
#else
BINARY_LOOP {
@type@ in1 = *(@type@ *)ip1;
@type@ in2 = *(@type@ *)ip2;
*(@type@ *)op1 = npy_rshift@c@(in1, in2);
}
#endif
}
/**end repeat**/
/*
*****************************************************************************
** UNSIGNED INTEGER LOOPS
*****************************************************************************
*/
/**begin repeat
* #TYPE = UBYTE, USHORT, UINT, ULONG, ULONGLONG#
* #type = npy_ubyte, npy_ushort, npy_uint, npy_ulong, npy_ulonglong#
* #c = u,u,u,ul,ull#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_absolute)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_sign)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in > 0 ? 1 : 0);
}
/**begin repeat1
* Arithmetic
* #kind = bitwise_and, bitwise_or, bitwise_xor#
* #OP = &, |, ^#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
if (IS_BINARY_REDUCE) {
BINARY_REDUCE_LOOP_FAST(@type@, io1 @OP@= in2);
}
else {
BINARY_LOOP_FAST(@type@, @type@, *out = in1 @OP@ in2);
}
}
/**end repeat1**/
/**begin repeat1
* #kind = logical_and, logical_or#
* #OP = &&, ||#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
/*
* gcc vectorization of this is not good (PR60575) but manual integer
* vectorization is too tedious to be worthwhile
*/
BINARY_LOOP_FAST(@type@, npy_bool, *out = in1 @OP@ in2);
}
/**end repeat1**/
NPY_FINLINE npy_bool @TYPE@_logical_xor_(@type@ in1, @type@ in2)
{ return (!!in1) != (!!in2); }
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_logical_xor)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
BINARY_LOOP_FAST(@type@, npy_bool, *out = @TYPE@_logical_xor_(in1, in2));
}
/**begin repeat1
* #kind = isnan, isinf, isfinite#
* #func = npy_isnan, npy_isinf, npy_isfinite#
* #val = NPY_FALSE, NPY_FALSE, NPY_TRUE#
**/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
/*
* The (void)in; suppresses an unused variable warning raised by gcc and allows
* us to re-use this macro even though we do not depend on in
*/
UNARY_LOOP_FAST(@type@, npy_bool, (void)in; *out = @val@);
}
/**end repeat1**/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_conjugate)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_logical_not)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, npy_bool, *out = !in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_invert)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = ~in);
}
/**end repeat**/
/*
*****************************************************************************
** SIGNED! INTEGER LOOPS
*****************************************************************************
*/
/**begin repeat
* #TYPE = BYTE, SHORT, INT, LONG, LONGLONG#
* #type = npy_byte, npy_short, npy_int, npy_long, npy_longlong#
* #c = ,,,l,ll#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_absolute)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = (in >= 0) ? in : -in);
}
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_sign)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in > 0 ? 1 : (in < 0 ? -1 : 0));
}
/**begin repeat1
* #kind = conjugate, invert, isnan, isinf, isfinite,
* logical_and, logical_or, logical_xor, logical_not,
* bitwise_and, bitwise_or, bitwise_xor#
**/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *func)
{
NPY_CPU_DISPATCH_CURFX(U@TYPE@_@kind@)(args, dimensions, steps, func);
}
/**end repeat1**/
/**end repeat**/
/*
*****************************************************************************
** BOOLEAN LOOPS **
*****************************************************************************
*/
/**begin repeat
* #kind = isnan, isinf, isfinite#
* #func = npy_isnan, npy_isinf, npy_isfinite#
* #val = NPY_FALSE, NPY_FALSE, NPY_TRUE#
**/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(BOOL_@kind@)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *func)
{
NPY_CPU_DISPATCH_CURFX(UBYTE_@kind@)(args, dimensions, steps, func);
}
/**end repeat**/
/*
*****************************************************************************
** HALF-FLOAT LOOPS **
*****************************************************************************
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(HALF_absolute)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(npy_half, npy_half, *out = in&0x7fffu);
}
/*
*****************************************************************************
** DATETIME LOOPS **
*****************************************************************************
*/
/**begin repeat
* #type = npy_datetime, npy_timedelta#
* #TYPE = DATETIME, TIMEDELTA#
*/
NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_isinf)
(char **args, npy_intp const *dimensions, npy_intp const *steps, void *func)
{
NPY_CPU_DISPATCH_CURFX(ULONGLONG_isinf)(args, dimensions, steps, func);
}
/**end repeat**/
|
