-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathmodule.cpp
More file actions
473 lines (413 loc) · 29 KB
/
Copy pathmodule.cpp
File metadata and controls
473 lines (413 loc) · 29 KB
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
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
// pybind11 module definition for numpycpp.
// Registers double + float32 overloads for every template function.
// All bindings use static_cast to disambiguate wrapper from native overloads in numpy::.
#include <pybind11/pybind11.h>
#include <pybind11/stl.h>
#include <dlfcn.h>
#include "numpycpp/numpy_py.h"
namespace py = pybind11;
// -- signature-disambiguated binding helpers ----------------------------------
// Unary element-wise: py::array_t<T>(*)(const py::array_t<T>&)
#define BIND_F1(name) \
m.def(#name, static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::name)); \
m.def(#name, static_cast<py::array_t<float>(*)(const py::array_t<float>&)>(&numpy::name))
// Reduction: T(*)(const py::array_t<T>&)
#define BIND_F_REDUCE(name) \
m.def(#name, static_cast<double(*)(const py::array_t<double>&)>(&numpy::name)); \
m.def(#name, static_cast<float(*)(const py::array_t<float>&)>(&numpy::name))
// Reduction returning py::ssize_t
#define BIND_F_REDUCE_PYSSIZE(name) \
m.def(#name, static_cast<py::ssize_t(*)(const py::array_t<double>&)>(&numpy::name)); \
m.def(#name, static_cast<py::ssize_t(*)(const py::array_t<float>&)>(&numpy::name))
// 1-arg array transform: py::array_t<T>(*)(const py::array_t<T>&, py::ssize_t)
#define BIND_F_MANIP1(name) \
m.def(#name, static_cast<py::array_t<double>(*)(const py::array_t<double>&, py::ssize_t)>(&numpy::name)); \
m.def(#name, static_cast<py::array_t<float>(*)(const py::array_t<float>&, py::ssize_t)>(&numpy::name))
// Stack/concat: py::array_t<T>(*)(const std::vector<py::array_t<T>>&)
#define BIND_F_STACK(name) \
m.def(#name, static_cast<py::array_t<double>(*)(const std::vector<py::array_t<double>>&)>(&numpy::name)); \
m.def(#name, static_cast<py::array_t<float>(*)(const std::vector<py::array_t<float>>&)>(&numpy::name))
// ============================================================================
PYBIND11_MODULE(numpycpp, m) {
m.doc() = "C++ pixel-level alignment of Python numpy, powered by Eigen";
// -- eager init: 预加载 g_svml_handle，加速首次 dlsym 调用。
// resolve_svml() 自身也是 fork 安全的（pid 检测），这里只是预热。--
{
std::string path = numpy::detail::find_umath_path();
if (!path.empty()) {
numpy::detail::g_svml_handle = dlopen(path.c_str(), RTLD_NOLOAD | RTLD_LAZY);
numpy::detail::g_svml_pid = getpid();
}
}
// -- 编译模式报告 ----------------------------------------------------------
m.def("compile_mode", []() -> const char* {
#ifdef NUMPYCPP_STD_ONLY
return "std";
#else
return "bit_exact";
#endif
});
// -- astype dtype 匹配诊断 -------------------------------------------------
// 返回 arr 的 dtype 匹配详情。仅用 buf.format + itemsize（C API），
// 不调用 dt.kind()（避免 Python↔C++ 递归）。
m.def("_diag_astype_dtype", [](const py::array& arr) -> py::dict {
auto buf = arr.request();
auto dt = arr.dtype();
char fc = buf.format.empty() ? '\0' :
(buf.format[0] == '<' || buf.format[0] == '>' || buf.format[0] == '=')
? buf.format[1] : buf.format[0];
bool is_f32 = (fc == 'f' && buf.itemsize == 4);
bool is_f64 = (fc == 'd' && buf.itemsize == 8) ||
(fc == 'f' && buf.itemsize == 8);
bool dt_of_f32 = dt.is(py::dtype::of<float>());
bool dt_of_f64 = dt.is(py::dtype::of<double>());
py::dict result;
result["dtype_str"] = py::str(dt);
result["format_char"] = std::string(1, fc);
result["itemsize"] = buf.itemsize;
result["format"] = buf.format;
result["is_f32(fmt)"] = is_f32;
result["is_f64(fmt)"] = is_f64;
result["dtype.is(float)"] = dt_of_f32;
result["dtype.is(double)"]= dt_of_f64;
result["fallback_works"] = (is_f32 || is_f64 || dt_of_f32 || dt_of_f64);
return result;
});
// -- atan2 dlsym 诊断 ----------------------------------------------------
m.def("_diag_atan2_dlsym", []() -> py::dict {
py::dict r;
std::string umath_path = numpy::detail::find_umath_path();
r["umath_path"] = umath_path.empty() ? "(null)" : umath_path;
if (!umath_path.empty()) {
void* h = dlopen(umath_path.c_str(), RTLD_NOLOAD | RTLD_LAZY);
r["dlopen_manual_ok"] = (h != nullptr);
if (!h) r["dlerror"] = dlerror();
numpy::detail::g_svml_handle = h;
}
r["handle_ok"] = (numpy::detail::g_svml_handle != nullptr);
// 尝试解析各种符号
void* sym_npy = numpy::detail::g_svml_handle ?
dlsym(numpy::detail::g_svml_handle, "npy_atan2") : nullptr;
void* sym_svml = numpy::detail::g_svml_handle ?
dlsym(numpy::detail::g_svml_handle, "__svml_atan28") : nullptr;
void* sym_std = dlsym(RTLD_DEFAULT, "atan2");
r["npy_atan2_ok"] = (sym_npy != nullptr);
r["svml_atan28_ok"] = (sym_svml != nullptr);
r["std_atan2_ok"] = (sym_std != nullptr);
// 实际调用 npy_atan2
if (sym_npy) {
auto fn = (double (*)(double,double))sym_npy;
r["npy_test_val"] = fn(4.7294, 3.5340);
}
if (sym_std) {
auto fn = (double (*)(double,double))sym_std;
r["std_test_val"] = fn(4.7294, 3.5340);
}
return r;
});
// -- linalg submodule --------------------------------------------------
py::module_ la = m.def_submodule("linalg", "numpy.linalg equivalents");
la.def("norm", static_cast<float(*)(const py::array_t<float>&)>(&numpy::linalg::norm));
la.def("norm", static_cast<double(*)(const py::array_t<double>&)>(&numpy::linalg::norm));
la.def("norm", static_cast<py::array_t<float>(*)(const py::array_t<float>&, int)>(&numpy::linalg::norm), py::arg("arr"), py::arg("axis") = -1);
la.def("norm", static_cast<py::array_t<double>(*)(const py::array_t<double>&, int)>(&numpy::linalg::norm), py::arg("arr"), py::arg("axis") = -1);
la.def("inv", static_cast<py::array_t<float>(*)(const py::array_t<float>&)>(&numpy::linalg::inv));
la.def("inv", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::linalg::inv));
// -- Array creation ----------------------------------------------------
BIND_F1(zeros_like); BIND_F1(ones_like); BIND_F1(empty_like);
m.def("full_like", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::full_like));
m.def("full_like", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::full_like));
// NOTE: _bool suffix — dtype-specific wrappers needed because pybind11
// cannot deduce template argument from a Python dtype keyword.
m.def("full_like", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, bool)>(&numpy::full_like));
m.def("zeros_like", static_cast<py::array(*)(const py::array&, const std::string&)>(&numpy::zeros_like),
py::arg("arr"), py::arg("dtype"));
m.def("ones_like", static_cast<py::array(*)(const py::array&, const std::string&)>(&numpy::ones_like),
py::arg("arr"), py::arg("dtype"));
m.def("zeros", &numpy::zeros);
m.def("ones", &numpy::ones);
m.def("full", static_cast<py::array_t<double>(*)(const std::vector<py::ssize_t>&, double)>(&numpy::full));
m.def("empty", static_cast<py::array_t<double>(*)(const std::vector<py::ssize_t>&)>(&numpy::empty));
// -- arange / linspace / logspace / geomspace --------------------------
// Single py::object dispatch: dtype inferred from input numpy scalar type
// (np.float32 → float32 output, Python float / np.float64 → float64 output)
// exactly mirroring numpy's own dtype inference behaviour.
m.def("arange",
static_cast<py::array(*)(py::object, py::object, py::object)>(&numpy::arange),
py::arg("a"), py::arg("b")=py::none(), py::arg("c")=py::none());
m.def("linspace",
static_cast<py::array(*)(py::object, py::object, py::ssize_t, bool)>(&numpy::linspace),
py::arg("start"), py::arg("stop"), py::arg("num")=50, py::arg("endpoint")=true);
m.def("logspace",
static_cast<py::array(*)(py::object, py::object, py::ssize_t, bool, double)>(&numpy::logspace),
py::arg("start"), py::arg("stop"), py::arg("num")=50, py::arg("endpoint")=true, py::arg("base")=10.0);
m.def("geomspace",
static_cast<py::array(*)(py::object, py::object, py::ssize_t, bool)>(&numpy::geomspace),
py::arg("start"), py::arg("stop"), py::arg("num")=50, py::arg("endpoint")=true);
// -- eye / identity / diag ---------------------------------------------
// eye / identity — M=None 表示方阵，严格对齐 numpy API
m.def("eye", static_cast<py::array_t<float> (*)(py::ssize_t,py::object,int)>(&numpy::eye<float>),
py::arg("N"), py::arg("M")=py::none(), py::arg("k")=0);
m.def("eye", static_cast<py::array_t<double>(*)(py::ssize_t,py::object,int)>(&numpy::eye),
py::arg("N"), py::arg("M")=py::none(), py::arg("k")=0);
m.def("identity", static_cast<py::array_t<float> (*)(py::ssize_t)>(&numpy::identity<float>));
m.def("identity", static_cast<py::array_t<double>(*)(py::ssize_t)>(&numpy::identity));
m.def("diag", static_cast<py::array_t<float> (*)(const py::array_t<float>&, int)>(&numpy::diag<float>),
py::arg("v"), py::arg("k")=0);
m.def("diag", static_cast<py::array_t<double>(*)(const py::array_t<double>&,int)>(&numpy::diag),
py::arg("v"), py::arg("k")=0);
// -- astype ------------------------------------------------------------
// NOTE: astype_int / astype_bool / astype_bool_from_int instead of a
// single "astype" — pybind11 cannot resolve overloads that differ only
// by return type (e.g. astype<double> vs astype<bool> both take
// py::array_t<double>). Each dtype combo needs a distinct Python name.
m.def("astype", static_cast<py::array(*)(const py::array&, const std::string&)>(&numpy::astype),
py::arg("arr"), py::arg("dtype"));
m.def("truncate_to_float32", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::truncate_to_float32));
// -- Element-wise math -------------------------------------------------
BIND_F1(sqrt); BIND_F1(abs); BIND_F1(exp); BIND_F1(log);
BIND_F1(sin); BIND_F1(cos); BIND_F1(tan);
BIND_F1(cbrt); BIND_F1(expm1); BIND_F1(log1p);
BIND_F1(log10); BIND_F1(log2); BIND_F1(arcsin); BIND_F1(arccos); BIND_F1(arctan);
BIND_F1(round); BIND_F1(floor); BIND_F1(ceil);
BIND_F1(degrees); BIND_F1(radians); BIND_F1(sign);
BIND_F1(reciprocal);
m.def("power", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::power));
m.def("power", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::power));
m.def("clip", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double, double)>(&numpy::clip));
m.def("clip", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float, float)>(&numpy::clip));
// -- Reduction (single-arg) --------------------------------------------
BIND_F_REDUCE(sum);
m.def("mean", static_cast<double(*)(const py::array_t<double>&)>(&numpy::mean));
m.def("mean", static_cast<float(*)(const py::array_t<float>&)>(&numpy::mean));
BIND_F_REDUCE(max); BIND_F_REDUCE(min);
m.def("any", static_cast<bool(*)(const py::array_t<bool>&)>(&numpy::any));
m.def("all", static_cast<bool(*)(const py::array_t<bool>&)>(&numpy::all));
// -- mean with axis (preserves input dtype, matching numpy) -------------
m.def("mean", static_cast<py::array_t<double>(*)(const py::array_t<double>&, int)>(&numpy::mean),
py::arg("arr"), py::arg("axis"));
m.def("mean", static_cast<py::array_t<float>(*)(const py::array_t<float>&, int)>(&numpy::mean),
py::arg("arr"), py::arg("axis"));
// -- Comparison --------------------------------------------------------
m.def("greater", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::greater));
m.def("greater", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::greater));
m.def("less", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::less));
m.def("less", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::less));
m.def("equal", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::equal));
m.def("equal", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::equal));
m.def("greater_equal", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::greater_equal));
m.def("greater_equal", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::greater_equal));
m.def("less_equal", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::less_equal));
m.def("less_equal", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::less_equal));
// -- Logical -----------------------------------------------------------
m.def("logical_and", static_cast<py::array_t<bool>(*)(const py::array_t<bool>&, const py::array_t<bool>&)>(&numpy::logical_and));
m.def("logical_or", static_cast<py::array_t<bool>(*)(const py::array_t<bool>&, const py::array_t<bool>&)>(&numpy::logical_or));
m.def("logical_not", static_cast<py::array_t<bool>(*)(const py::array_t<bool>&)>(&numpy::logical_not));
// -- Array access ------------------------------------------------------
m.def("array_get", static_cast<double(*)(const py::array_t<double>&, py::ssize_t)>(&numpy::array_get));
m.def("array_get", static_cast<double(*)(const py::array_t<double>&, py::ssize_t, py::ssize_t)>(&numpy::array_get));
m.def("array_get", static_cast<bool(*)(const py::array_t<bool>&, py::ssize_t)>(&numpy::array_get));
m.def("array_get", static_cast<double(*)(const py::array&, py::ssize_t)>(&numpy::array_get));
// -- Dict helpers ------------------------------------------------------
m.def("get_array", &numpy::get_array);
m.def("set_array", &numpy::set_array);
// -- Conversion --------------------------------------------------------
m.def("asarray", static_cast<py::array_t<double>(*)(const std::vector<double>&)>(&numpy::asarray));
m.def("asarray", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::asarray));
m.def("array", static_cast<py::array_t<double>(*)(const std::vector<double>&)>(&numpy::array));
m.def("array", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::array));
// -- transpose / flatten -----------------------------------------------
BIND_F1(transpose); BIND_F1(flatten);
// -- to_vector ---------------------------------------------------------
m.def("to_vector", static_cast<std::vector<bool>(*)(const py::array_t<bool>&)>(&numpy::to_vector));
m.def("to_vector", static_cast<std::vector<double>(*)(const py::array_t<double>&)>(&numpy::to_vector));
// -- Slice helpers (template + non-template overloads) ------------------
m.def("slice", static_cast<py::array_t<double>(*)(const py::array_t<double>&, py::ssize_t, py::ssize_t)>(&numpy::slice));
m.def("slice", static_cast<py::array_t<float>(*)(const py::array_t<float>&, py::ssize_t, py::ssize_t)>(&numpy::slice));
m.def("slice", static_cast<py::array(*)(const py::array&, py::ssize_t, py::ssize_t)>(&numpy::slice));
// -- Column slice ------------------------------------------------------
// NOTE: named take_cols, not take — numpy.take(a, indices, axis) is
// fully general; this is a specialized subset (first N columns, axis=1,
// 2D only) exposing a simpler (arr, n) signature.
BIND_F_MANIP1(take_cols);
// -- Slice assignment ---------------------------------------------------
// NOTE: float overload registered before double — ensures pybind11
// dispatches float32 arrays to the correct (in-place) overload instead
// of silently casting to double and mutating a temporary copy.
m.def("slice_assign", static_cast<void(*)(py::array_t<float>, py::ssize_t, float)>(&numpy::slice_assign));
m.def("slice_assign", static_cast<void(*)(py::array_t<double>, py::ssize_t, double)>(&numpy::slice_assign));
m.def("slice_assign", static_cast<void(*)(py::array_t<int>, py::ssize_t, int)>(&numpy::slice_assign));
m.def("slice_assign", static_cast<void(*)(py::array_t<bool>, py::ssize_t, bool)>(&numpy::slice_assign));
// -- Binary element-wise: scalar overloads BEFORE array-array ----------
m.def("hypot", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::hypot));
m.def("hypot", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::hypot));
m.def("arctan2", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::arctan2));
m.def("arctan2", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::arctan2));
m.def("arctan2", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::arctan2));
m.def("arctan2", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::arctan2));
m.def("maximum", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::maximum));
m.def("maximum", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::maximum));
m.def("maximum", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::maximum));
m.def("maximum", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::maximum));
m.def("minimum", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::minimum));
m.def("minimum", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::minimum));
m.def("minimum", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::minimum));
m.def("minimum", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::minimum));
// -- Comparison helpers ------------------------------------------------
m.def("not_equal", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, double)>(&numpy::not_equal));
m.def("not_equal", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, float)>(&numpy::not_equal));
m.def("not_equal", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::not_equal));
m.def("not_equal", static_cast<py::array_t<bool>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::not_equal));
// -- Special value helpers ---------------------------------------------
m.def("isnan", static_cast<py::array_t<bool>(*)(const py::array_t<double>&)>(&numpy::isnan));
m.def("isnan", static_cast<py::array_t<bool>(*)(const py::array_t<float>&)>(&numpy::isnan));
m.def("isinf", static_cast<py::array_t<bool>(*)(const py::array_t<double>&)>(&numpy::isinf));
m.def("isinf", static_cast<py::array_t<bool>(*)(const py::array_t<float>&)>(&numpy::isinf));
m.def("isfinite", static_cast<py::array_t<bool>(*)(const py::array_t<double>&)>(&numpy::isfinite));
m.def("isfinite", static_cast<py::array_t<bool>(*)(const py::array_t<float>&)>(&numpy::isfinite));
// -- Array manipulation ------------------------------------------------
m.def("diff", static_cast<py::array_t<double>(*)(const py::array_t<double>&, int, int)>(&numpy::diff),
py::arg("arr"), py::arg("n") = 1, py::arg("axis") = -1);
m.def("diff", static_cast<py::array_t<float>(*)(const py::array_t<float>&, int, int)>(&numpy::diff),
py::arg("arr"), py::arg("n") = 1, py::arg("axis") = -1);
BIND_F_STACK(stack); BIND_F_STACK(vstack); BIND_F_STACK(hstack);
m.def("concatenate", static_cast<py::array_t<double>(*)(const std::vector<py::array_t<double>>&, int)>(&numpy::concatenate),
py::arg("arrays"), py::arg("axis") = 0);
m.def("concatenate", static_cast<py::array_t<float>(*)(const std::vector<py::array_t<float>>&, int)>(&numpy::concatenate),
py::arg("arrays"), py::arg("axis") = 0);
m.def("where", static_cast<py::array_t<double>(*)(const py::array_t<bool>&, double, double)>(&numpy::where));
m.def("where", static_cast<py::array_t<float>(*)(const py::array_t<bool>&, float, float)>(&numpy::where));
m.def("where", static_cast<py::array_t<double>(*)(const py::array_t<bool>&, const py::array_t<double>&, const py::array_t<double>&)>(&numpy::where));
m.def("where", static_cast<py::array_t<float>(*)(const py::array_t<bool>&, const py::array_t<float>&, const py::array_t<float>&)>(&numpy::where));
// -- Statistical -------------------------------------------------------
BIND_F_REDUCE(std); BIND_F_REDUCE(var);
// -- Set operations ----------------------------------------------------
m.def("isin", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, const std::vector<double>&)>(&numpy::isin));
m.def("isin", static_cast<py::array_t<bool>(*)(const py::array_t<double>&, const std::vector<int>&)>(&numpy::isin));
m.def("intersect1d", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::intersect1d));
m.def("intersect1d", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::intersect1d));
m.def("flatnonzero", static_cast<py::array_t<py::ssize_t>(*)(const py::array_t<double>&)>(&numpy::flatnonzero));
m.def("unwrap", static_cast<py::array_t<double>(*)(const py::array_t<double>&, double)>(&numpy::unwrap), py::arg("arr"), py::arg("discont") = M_PI);
m.def("unwrap", static_cast<py::array_t<float>(*)(const py::array_t<float>&, float)>(&numpy::unwrap), py::arg("arr"), py::arg("discont") = (float)M_PI);
m.def("cumsum", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::cumsum));
m.def("cumsum", static_cast<py::array_t<float>(*)(const py::array_t<float>&)>(&numpy::cumsum));
m.def("squeeze", static_cast<py::array_t<double>(*)(const py::array_t<double>&)>(&numpy::squeeze));
m.def("squeeze", static_cast<py::array_t<float>(*)(const py::array_t<float>&)>(&numpy::squeeze));
// -- Interpolation -----------------------------------------------------
m.def("interp", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&, const py::array_t<double>&)>(&numpy::interp));
// -- Logical XOR -------------------------------------------------------
m.def("logical_xor", static_cast<py::array_t<bool>(*)(const py::array_t<bool>&, const py::array_t<bool>&)>(&numpy::logical_xor));
// -- Sorting and indexing ----------------------------------------------
m.def("argsort", static_cast<py::array_t<py::ssize_t>(*)(const py::array_t<double>&)>(&numpy::argsort));
m.def("argsort", static_cast<py::array_t<py::ssize_t>(*)(const py::array_t<float>&)>(&numpy::argsort));
BIND_F_REDUCE_PYSSIZE(argmax); BIND_F_REDUCE_PYSSIZE(argmin);
// -- Array transformation ----------------------------------------------
BIND_F_MANIP1(roll); BIND_F1(flip); BIND_F_MANIP1(repeat); BIND_F_MANIP1(tile);
// -- Safe division -----------------------------------------------------
m.def("safe_divide", &numpy::safe_divide, py::arg("a"), py::arg("b"), py::arg("default_val") = 0.0);
// -- Advanced / Fancy indexing -----------------------------------------
// numpy.take(a, indices, axis=None)
m.def("take",
static_cast<py::array_t<double>(*)(
const py::array_t<double>&,
const py::array_t<py::ssize_t>&, int)>(&numpy::take),
py::arg("arr"), py::arg("indices"), py::arg("axis") = -1);
m.def("take",
static_cast<py::array_t<float>(*)(
const py::array_t<float>&,
const py::array_t<py::ssize_t>&, int)>(&numpy::take),
py::arg("arr"), py::arg("indices"), py::arg("axis") = -1);
// numpy.compress(condition, a) — boolean mask gather
m.def("compress",
static_cast<py::array_t<double>(*)(
const py::array_t<bool>&,
const py::array_t<double>&)>(&numpy::compress));
m.def("compress",
static_cast<py::array_t<float>(*)(
const py::array_t<bool>&,
const py::array_t<float>&)>(&numpy::compress));
// slice(a, starts, stops, steps) — N-D overload of existing slice()
m.def("slice",
static_cast<py::array_t<double>(*)(
const py::array_t<double>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&)>(&numpy::slice));
m.def("slice",
static_cast<py::array_t<float>(*)(
const py::array_t<float>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&)>(&numpy::slice));
// numpy.put(a, indices, values) — in-place scatter
m.def("put",
static_cast<void(*)(
py::array_t<double>,
const py::array_t<py::ssize_t>&,
const py::array_t<double>&)>(&numpy::put));
m.def("put",
static_cast<void(*)(
py::array_t<float>,
const py::array_t<py::ssize_t>&,
const py::array_t<float>&)>(&numpy::put));
// numpy.putmask(a, mask, scalar) — in-place a[mask] = scalar
// NOTE: float BEFORE double — pybind11 forcecast would silently upcast
// a float32 array to a float64 temporary copy and modify that copy instead
// of the original array. Registering the narrower type first avoids this.
m.def("putmask",
static_cast<void(*)(py::array_t<float>,
const py::array_t<bool>&, float)>(
&numpy::putmask));
m.def("putmask",
static_cast<void(*)(py::array_t<double>,
const py::array_t<bool>&, double)>(
&numpy::putmask));
// numpy.putmask(a, mask, values) — in-place a[mask] = array (float first)
m.def("putmask",
static_cast<void(*)(py::array_t<float>,
const py::array_t<bool>&,
const py::array_t<float>&)>(
&numpy::putmask));
m.def("putmask",
static_cast<void(*)(py::array_t<double>,
const py::array_t<bool>&,
const py::array_t<double>&)>(
&numpy::putmask));
// slice_assign(a, starts, stops, steps, scalar) — N-D overload (float first)
m.def("slice_assign",
static_cast<void(*)(py::array_t<float>,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&, float)>(
&numpy::slice_assign));
m.def("slice_assign",
static_cast<void(*)(py::array_t<double>,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&, double)>(
&numpy::slice_assign));
// slice_assign(a, starts, stops, steps, values) — N-D array overload (float first)
m.def("slice_assign",
static_cast<void(*)(py::array_t<float>,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const py::array_t<float>&)>(
&numpy::slice_assign));
m.def("slice_assign",
static_cast<void(*)(py::array_t<double>,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const std::vector<py::ssize_t>&,
const py::array_t<double>&)>(
&numpy::slice_assign));
// -- Dot product -------------------------------------------------------
m.def("dot", static_cast<double(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::dot));
m.def("dot", static_cast<float(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::dot));
// -- Matmul ------------------------------------------------------------
m.def("matmul", static_cast<py::array_t<double>(*)(const py::array_t<double>&, const py::array_t<double>&)>(&numpy::matmul));
m.def("matmul", static_cast<py::array_t<float>(*)(const py::array_t<float>&, const py::array_t<float>&)>(&numpy::matmul));
// -- Einsum ------------------------------------------------------------
m.def("einsum", static_cast<py::array_t<double>(*)(const std::string&, const py::array_t<double>&, const py::array_t<double>&)>(&numpy::einsum));
m.def("einsum", static_cast<py::array_t<float>(*)(const std::string&, const py::array_t<float>&, const py::array_t<float>&)>(&numpy::einsum));
}