aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/tools/cython/Cython/Utility/MemoryView.pyx
blob: 277c0bd87a36cefd906d54d8f04aa7ed4fdd756f (plain) (blame)
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
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
#################### View.MemoryView ####################

# This utility provides cython.array and cython.view.memoryview

from __future__ import absolute_import

cimport cython

# from cpython cimport ...
cdef extern from "Python.h":
    int PyIndex_Check(object)
    object PyLong_FromVoidPtr(void *)

cdef extern from "pythread.h":
    ctypedef void *PyThread_type_lock

    PyThread_type_lock PyThread_allocate_lock()
    void PyThread_free_lock(PyThread_type_lock)
    int PyThread_acquire_lock(PyThread_type_lock, int mode) nogil
    void PyThread_release_lock(PyThread_type_lock) nogil

cdef extern from "<string.h>":
    void *memset(void *b, int c, size_t len)

cdef extern from *:
    bint __PYX_CYTHON_ATOMICS_ENABLED() noexcept
    int __Pyx_GetBuffer(object, Py_buffer *, int) except -1
    void __Pyx_ReleaseBuffer(Py_buffer *)

    ctypedef struct PyObject
    ctypedef Py_ssize_t Py_intptr_t
    void Py_INCREF(PyObject *)
    void Py_DECREF(PyObject *)

    void* PyMem_Malloc(size_t n)
    void PyMem_Free(void *p)
    void* PyObject_Malloc(size_t n)
    void PyObject_Free(void *p)

    cdef struct __pyx_memoryview "__pyx_memoryview_obj":
        Py_buffer view
        PyObject *obj
        __Pyx_TypeInfo *typeinfo

    ctypedef struct {{memviewslice_name}}:
        __pyx_memoryview *memview
        char *data
        Py_ssize_t shape[{{max_dims}}]
        Py_ssize_t strides[{{max_dims}}]
        Py_ssize_t suboffsets[{{max_dims}}]

    void __PYX_INC_MEMVIEW({{memviewslice_name}} *memslice, int have_gil)
    void __PYX_XDEC_MEMVIEW({{memviewslice_name}} *memslice, int have_gil)

    ctypedef struct __pyx_buffer "Py_buffer":
        PyObject *obj

    PyObject *Py_None

    cdef enum:
        PyBUF_C_CONTIGUOUS,
        PyBUF_F_CONTIGUOUS,
        PyBUF_ANY_CONTIGUOUS
        PyBUF_FORMAT
        PyBUF_WRITABLE
        PyBUF_STRIDES
        PyBUF_INDIRECT
        PyBUF_ND
        PyBUF_RECORDS
        PyBUF_RECORDS_RO

    ctypedef struct __Pyx_TypeInfo:
        pass

    cdef object capsule "__pyx_capsule_create" (void *p, char *sig)
    cdef int __pyx_array_getbuffer(PyObject *obj, Py_buffer view, int flags)
    cdef int __pyx_memoryview_getbuffer(PyObject *obj, Py_buffer view, int flags)

cdef extern from *:
    ctypedef int __pyx_atomic_int
    {{memviewslice_name}} slice_copy_contig "__pyx_memoryview_copy_new_contig"(
                                 __Pyx_memviewslice *from_mvs,
                                 char *mode, int ndim,
                                 size_t sizeof_dtype, int contig_flag,
                                 bint dtype_is_object) nogil except *
    bint slice_is_contig "__pyx_memviewslice_is_contig" (
                            {{memviewslice_name}} mvs, char order, int ndim) nogil
    bint slices_overlap "__pyx_slices_overlap" ({{memviewslice_name}} *slice1,
                                                {{memviewslice_name}} *slice2,
                                                int ndim, size_t itemsize) nogil


cdef extern from "<stdlib.h>":
    void *malloc(size_t) nogil
    void free(void *) nogil
    void *memcpy(void *dest, void *src, size_t n) nogil




#
### cython.array class
#

@cname("__pyx_array")
cdef class array:

    cdef:
        char *data
        Py_ssize_t len
        char *format
        int ndim
        Py_ssize_t *_shape
        Py_ssize_t *_strides
        Py_ssize_t itemsize
        unicode mode  # FIXME: this should have been a simple 'char'
        bytes _format
        void (*callback_free_data)(void *data)
        # cdef object _memview
        cdef bint free_data
        cdef bint dtype_is_object

    def __cinit__(array self, tuple shape, Py_ssize_t itemsize, format not None,
                  mode="c", bint allocate_buffer=True):

        cdef int idx
        cdef Py_ssize_t i, dim
        cdef PyObject **p

        self.ndim = <int> len(shape)
        self.itemsize = itemsize

        if not self.ndim:
            raise ValueError("Empty shape tuple for cython.array")

        if itemsize <= 0:
            raise ValueError("itemsize <= 0 for cython.array")

        if not isinstance(format, bytes):
            format = format.encode('ASCII')
        self._format = format  # keep a reference to the byte string
        self.format = self._format

        # use single malloc() for both shape and strides
        self._shape = <Py_ssize_t *> PyObject_Malloc(sizeof(Py_ssize_t)*self.ndim*2)
        self._strides = self._shape + self.ndim

        if not self._shape:
            raise MemoryError("unable to allocate shape and strides.")

        # cdef Py_ssize_t dim, stride
        for idx, dim in enumerate(shape):
            if dim <= 0:
                raise ValueError("Invalid shape in axis %d: %d." % (idx, dim))
            self._shape[idx] = dim

        cdef char order
        if mode == 'fortran':
            order = b'F'
            self.mode = u'fortran'
        elif mode == 'c':
            order = b'C'
            self.mode = u'c'
        else:
            raise ValueError("Invalid mode, expected 'c' or 'fortran', got %s" % mode)

        self.len = fill_contig_strides_array(self._shape, self._strides,
                                             itemsize, self.ndim, order)

        self.free_data = allocate_buffer
        self.dtype_is_object = format == b'O'
        if allocate_buffer:
            # use malloc() for backwards compatibility
            # in case external code wants to change the data pointer
            self.data = <char *>malloc(self.len)
            if not self.data:
                raise MemoryError("unable to allocate array data.")

            if self.dtype_is_object:
                p = <PyObject **> self.data
                for i in range(self.len / itemsize):
                    p[i] = Py_None
                    Py_INCREF(Py_None)

    @cname('getbuffer')
    def __getbuffer__(self, Py_buffer *info, int flags):
        cdef int bufmode = -1
        if self.mode == u"c":
            bufmode = PyBUF_C_CONTIGUOUS | PyBUF_ANY_CONTIGUOUS
        elif self.mode == u"fortran":
            bufmode = PyBUF_F_CONTIGUOUS | PyBUF_ANY_CONTIGUOUS
        if not (flags & bufmode):
            raise ValueError("Can only create a buffer that is contiguous in memory.")
        info.buf = self.data
        info.len = self.len
        info.ndim = self.ndim
        info.shape = self._shape
        info.strides = self._strides
        info.suboffsets = NULL
        info.itemsize = self.itemsize
        info.readonly = 0

        if flags & PyBUF_FORMAT:
            info.format = self.format
        else:
            info.format = NULL

        info.obj = self

    __pyx_getbuffer = capsule(<void *> &__pyx_array_getbuffer, "getbuffer(obj, view, flags)")

    def __dealloc__(array self):
        if self.callback_free_data != NULL:
            self.callback_free_data(self.data)
        elif self.free_data:
            if self.dtype_is_object:
                refcount_objects_in_slice(self.data, self._shape,
                                          self._strides, self.ndim, False)
            free(self.data)
        PyObject_Free(self._shape)

    @property
    def memview(self):
        return self.get_memview()

    @cname('get_memview')
    cdef get_memview(self):
        flags =  PyBUF_ANY_CONTIGUOUS|PyBUF_FORMAT|PyBUF_WRITABLE
        return  memoryview(self, flags, self.dtype_is_object)

    def __len__(self):
        return self._shape[0]

    def __getattr__(self, attr):
        return getattr(self.memview, attr)

    def __getitem__(self, item):
        return self.memview[item]

    def __setitem__(self, item, value):
        self.memview[item] = value


@cname("__pyx_array_new")
cdef array array_cwrapper(tuple shape, Py_ssize_t itemsize, char *format,
                          char *mode, char *buf):
    cdef array result

    if buf == NULL:
        result = array(shape, itemsize, format, mode.decode('ASCII'))
    else:
        result = array(shape, itemsize, format, mode.decode('ASCII'),
                       allocate_buffer=False)
        result.data = buf

    return result


#
### Memoryview constants and cython.view.memoryview class
#

# Disable generic_contiguous, as it makes trouble verifying contiguity:
#   - 'contiguous' or '::1' means the dimension is contiguous with dtype
#   - 'indirect_contiguous' means a contiguous list of pointers
#   - dtype contiguous must be contiguous in the first or last dimension
#     from the start, or from the dimension following the last indirect dimension
#
#   e.g.
#           int[::indirect_contiguous, ::contiguous, :]
#
#   is valid (list of pointers to 2d fortran-contiguous array), but
#
#           int[::generic_contiguous, ::contiguous, :]
#
#   would mean you'd have assert dimension 0 to be indirect (and pointer contiguous) at runtime.
#   So it doesn't bring any performance benefit, and it's only confusing.

@cname('__pyx_MemviewEnum')
cdef class Enum(object):
    cdef object name
    def __init__(self, name):
        self.name = name
    def __repr__(self):
        return self.name

cdef generic = Enum("<strided and direct or indirect>")
cdef strided = Enum("<strided and direct>") # default
cdef indirect = Enum("<strided and indirect>")
# Disable generic_contiguous, as it is a troublemaker
#cdef generic_contiguous = Enum("<contiguous and direct or indirect>")
cdef contiguous = Enum("<contiguous and direct>")
cdef indirect_contiguous = Enum("<contiguous and indirect>")

# 'follow' is implied when the first or last axis is ::1


@cname('__pyx_align_pointer')
cdef void *align_pointer(void *memory, size_t alignment) nogil:
    "Align pointer memory on a given boundary"
    cdef Py_intptr_t aligned_p = <Py_intptr_t> memory
    cdef size_t offset

    with cython.cdivision(True):
        offset = aligned_p % alignment

    if offset > 0:
        aligned_p += alignment - offset

    return <void *> aligned_p


# pre-allocate thread locks for reuse
## note that this could be implemented in a more beautiful way in "normal" Cython,
## but this code gets merged into the user module and not everything works there.
DEF THREAD_LOCKS_PREALLOCATED = 8
cdef int __pyx_memoryview_thread_locks_used = 0
cdef PyThread_type_lock[THREAD_LOCKS_PREALLOCATED] __pyx_memoryview_thread_locks = [
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
    PyThread_allocate_lock(),
]


@cname('__pyx_memoryview')
cdef class memoryview(object):

    cdef object obj
    cdef object _size
    cdef object _array_interface
    cdef PyThread_type_lock lock
    # the following array will contain a single __pyx_atomic int with
    # suitable alignment
    cdef __pyx_atomic_int acquisition_count[2]
    cdef __pyx_atomic_int *acquisition_count_aligned_p
    cdef Py_buffer view
    cdef int flags
    cdef bint dtype_is_object
    cdef __Pyx_TypeInfo *typeinfo

    def __cinit__(memoryview self, object obj, int flags, bint dtype_is_object=False):
        self.obj = obj
        self.flags = flags
        if type(self) is memoryview or obj is not None:
            __Pyx_GetBuffer(obj, &self.view, flags)
            if <PyObject *> self.view.obj == NULL:
                (<__pyx_buffer *> &self.view).obj = Py_None
                Py_INCREF(Py_None)

        if not __PYX_CYTHON_ATOMICS_ENABLED():
            global __pyx_memoryview_thread_locks_used
            if __pyx_memoryview_thread_locks_used < THREAD_LOCKS_PREALLOCATED:
                self.lock = __pyx_memoryview_thread_locks[__pyx_memoryview_thread_locks_used]
                __pyx_memoryview_thread_locks_used += 1
            if self.lock is NULL:
                self.lock = PyThread_allocate_lock()
                if self.lock is NULL:
                    raise MemoryError

        if flags & PyBUF_FORMAT:
            self.dtype_is_object = (self.view.format[0] == b'O' and self.view.format[1] == b'\0')
        else:
            self.dtype_is_object = dtype_is_object

        self.acquisition_count_aligned_p = <__pyx_atomic_int *> align_pointer(
                  <void *> &self.acquisition_count[0], sizeof(__pyx_atomic_int))
        self.typeinfo = NULL

    def __dealloc__(memoryview self):
        if self.obj is not None:
            __Pyx_ReleaseBuffer(&self.view)
        elif (<__pyx_buffer *> &self.view).obj == Py_None:
            # Undo the incref in __cinit__() above.
            (<__pyx_buffer *> &self.view).obj = NULL
            Py_DECREF(Py_None)

        cdef int i
        global __pyx_memoryview_thread_locks_used
        if self.lock != NULL:
            for i in range(__pyx_memoryview_thread_locks_used):
                if __pyx_memoryview_thread_locks[i] is self.lock:
                    __pyx_memoryview_thread_locks_used -= 1
                    if i != __pyx_memoryview_thread_locks_used:
                        __pyx_memoryview_thread_locks[i], __pyx_memoryview_thread_locks[__pyx_memoryview_thread_locks_used] = (
                            __pyx_memoryview_thread_locks[__pyx_memoryview_thread_locks_used], __pyx_memoryview_thread_locks[i])
                    break
            else:
                PyThread_free_lock(self.lock)

    cdef char *get_item_pointer(memoryview self, object index) except NULL:
        cdef Py_ssize_t dim
        cdef char *itemp = <char *> self.view.buf

        for dim, idx in enumerate(index):
            itemp = pybuffer_index(&self.view, itemp, idx, dim)

        return itemp

    #@cname('__pyx_memoryview_getitem')
    def __getitem__(memoryview self, object index):
        if index is Ellipsis:
            return self

        have_slices, indices = _unellipsify(index, self.view.ndim)

        cdef char *itemp
        if have_slices:
            return memview_slice(self, indices)
        else:
            itemp = self.get_item_pointer(indices)
            return self.convert_item_to_object(itemp)

    def __setitem__(memoryview self, object index, object value):
        if self.view.readonly:
            raise TypeError("Cannot assign to read-only memoryview")

        have_slices, index = _unellipsify(index, self.view.ndim)

        if have_slices:
            obj = self.is_slice(value)
            if obj:
                self.setitem_slice_assignment(self[index], obj)
            else:
                self.setitem_slice_assign_scalar(self[index], value)
        else:
            self.setitem_indexed(index, value)

    cdef is_slice(self, obj):
        if not isinstance(obj, memoryview):
            try:
                obj = memoryview(obj, self.flags & ~PyBUF_WRITABLE | PyBUF_ANY_CONTIGUOUS,
                                 self.dtype_is_object)
            except TypeError:
                return None

        return obj

    cdef setitem_slice_assignment(self, dst, src):
        cdef {{memviewslice_name}} dst_slice
        cdef {{memviewslice_name}} src_slice

        memoryview_copy_contents(get_slice_from_memview(src, &src_slice)[0],
                                 get_slice_from_memview(dst, &dst_slice)[0],
                                 src.ndim, dst.ndim, self.dtype_is_object)

    cdef setitem_slice_assign_scalar(self, memoryview dst, value):
        cdef int array[128]
        cdef void *tmp = NULL
        cdef void *item

        cdef {{memviewslice_name}} *dst_slice
        cdef {{memviewslice_name}} tmp_slice
        dst_slice = get_slice_from_memview(dst, &tmp_slice)

        if <size_t>self.view.itemsize > sizeof(array):
            tmp = PyMem_Malloc(self.view.itemsize)
            if tmp == NULL:
                raise MemoryError
            item = tmp
        else:
            item = <void *> array

        try:
            if self.dtype_is_object:
                (<PyObject **> item)[0] = <PyObject *> value
            else:
                self.assign_item_from_object(<char *> item, value)

            # It would be easy to support indirect dimensions, but it's easier
            # to disallow :)
            if self.view.suboffsets != NULL:
                assert_direct_dimensions(self.view.suboffsets, self.view.ndim)
            slice_assign_scalar(dst_slice, dst.view.ndim, self.view.itemsize,
                                item, self.dtype_is_object)
        finally:
            PyMem_Free(tmp)

    cdef setitem_indexed(self, index, value):
        cdef char *itemp = self.get_item_pointer(index)
        self.assign_item_from_object(itemp, value)

    cdef convert_item_to_object(self, char *itemp):
        """Only used if instantiated manually by the user, or if Cython doesn't
        know how to convert the type"""
        import struct
        cdef bytes bytesitem
        # Do a manual and complete check here instead of this easy hack
        bytesitem = itemp[:self.view.itemsize]
        try:
            result = struct.unpack(self.view.format, bytesitem)
        except struct.error:
            raise ValueError("Unable to convert item to object")
        else:
            if len(self.view.format) == 1:
                return result[0]
            return result

    cdef assign_item_from_object(self, char *itemp, object value):
        """Only used if instantiated manually by the user, or if Cython doesn't
        know how to convert the type"""
        import struct
        cdef char c
        cdef bytes bytesvalue
        cdef Py_ssize_t i

        if isinstance(value, tuple):
            bytesvalue = struct.pack(self.view.format, *value)
        else:
            bytesvalue = struct.pack(self.view.format, value)

        for i, c in enumerate(bytesvalue):
            itemp[i] = c

    @cname('getbuffer')
    def __getbuffer__(self, Py_buffer *info, int flags):
        if flags & PyBUF_WRITABLE and self.view.readonly:
            raise ValueError("Cannot create writable memory view from read-only memoryview")

        if flags & PyBUF_ND:
            info.shape = self.view.shape
        else:
            info.shape = NULL

        if flags & PyBUF_STRIDES:
            info.strides = self.view.strides
        else:
            info.strides = NULL

        if flags & PyBUF_INDIRECT:
            info.suboffsets = self.view.suboffsets
        else:
            info.suboffsets = NULL

        if flags & PyBUF_FORMAT:
            info.format = self.view.format
        else:
            info.format = NULL

        info.buf = self.view.buf
        info.ndim = self.view.ndim
        info.itemsize = self.view.itemsize
        info.len = self.view.len
        info.readonly = self.view.readonly
        info.obj = self

    __pyx_getbuffer = capsule(<void *> &__pyx_memoryview_getbuffer, "getbuffer(obj, view, flags)")

    # Some properties that have the same semantics as in NumPy
    @property
    def T(self):
        cdef _memoryviewslice result = memoryview_copy(self)
        transpose_memslice(&result.from_slice)
        return result

    @property
    def base(self):
        return self.obj

    @property
    def shape(self):
        return tuple([length for length in self.view.shape[:self.view.ndim]])

    @property
    def strides(self):
        if self.view.strides == NULL:
            # Note: we always ask for strides, so if this is not set it's a bug
            raise ValueError("Buffer view does not expose strides")

        return tuple([stride for stride in self.view.strides[:self.view.ndim]])

    @property
    def suboffsets(self):
        if self.view.suboffsets == NULL:
            return (-1,) * self.view.ndim

        return tuple([suboffset for suboffset in self.view.suboffsets[:self.view.ndim]])

    @property
    def ndim(self):
        return self.view.ndim

    @property
    def itemsize(self):
        return self.view.itemsize

    @property
    def nbytes(self):
        return self.size * self.view.itemsize

    @property
    def size(self):
        if self._size is None:
            result = 1

            for length in self.view.shape[:self.view.ndim]:
                result *= length

            self._size = result

        return self._size

    def __len__(self):
        if self.view.ndim >= 1:
            return self.view.shape[0]

        return 0

    def __repr__(self):
        return "<MemoryView of %r at 0x%x>" % (self.base.__class__.__name__,
                                               id(self))

    def __str__(self):
        return "<MemoryView of %r object>" % (self.base.__class__.__name__,)

    # Support the same attributes as memoryview slices
    def is_c_contig(self):
        cdef {{memviewslice_name}} *mslice
        cdef {{memviewslice_name}} tmp
        mslice = get_slice_from_memview(self, &tmp)
        return slice_is_contig(mslice[0], 'C', self.view.ndim)

    def is_f_contig(self):
        cdef {{memviewslice_name}} *mslice
        cdef {{memviewslice_name}} tmp
        mslice = get_slice_from_memview(self, &tmp)
        return slice_is_contig(mslice[0], 'F', self.view.ndim)

    def copy(self):
        cdef {{memviewslice_name}} mslice
        cdef int flags = self.flags & ~PyBUF_F_CONTIGUOUS

        slice_copy(self, &mslice)
        mslice = slice_copy_contig(&mslice, "c", self.view.ndim,
                                   self.view.itemsize,
                                   flags|PyBUF_C_CONTIGUOUS,
                                   self.dtype_is_object)

        return memoryview_copy_from_slice(self, &mslice)

    def copy_fortran(self):
        cdef {{memviewslice_name}} src, dst
        cdef int flags = self.flags & ~PyBUF_C_CONTIGUOUS

        slice_copy(self, &src)
        dst = slice_copy_contig(&src, "fortran", self.view.ndim,
                                self.view.itemsize,
                                flags|PyBUF_F_CONTIGUOUS,
                                self.dtype_is_object)

        return memoryview_copy_from_slice(self, &dst)


@cname('__pyx_memoryview_new')
cdef memoryview_cwrapper(object o, int flags, bint dtype_is_object, __Pyx_TypeInfo *typeinfo):
    cdef memoryview result = memoryview(o, flags, dtype_is_object)
    result.typeinfo = typeinfo
    return result

@cname('__pyx_memoryview_check')
cdef inline bint memoryview_check(object o):
    return isinstance(o, memoryview)

cdef tuple _unellipsify(object index, int ndim):
    """
    Replace all ellipses with full slices and fill incomplete indices with
    full slices.
    """
    if not isinstance(index, tuple):
        tup = (index,)
    else:
        tup = index

    result = []
    have_slices = False
    seen_ellipsis = False
    for idx, item in enumerate(tup):
        if item is Ellipsis:
            if not seen_ellipsis:
                result.extend([slice(None)] * (ndim - len(tup) + 1))
                seen_ellipsis = True
            else:
                result.append(slice(None))
            have_slices = True
        else:
            if not isinstance(item, slice) and not PyIndex_Check(item):
                raise TypeError("Cannot index with type '%s'" % type(item))

            have_slices = have_slices or isinstance(item, slice)
            result.append(item)

    nslices = ndim - len(result)
    if nslices:
        result.extend([slice(None)] * nslices)

    return have_slices or nslices, tuple(result)

cdef assert_direct_dimensions(Py_ssize_t *suboffsets, int ndim):
    for suboffset in suboffsets[:ndim]:
        if suboffset >= 0:
            raise ValueError("Indirect dimensions not supported")

#
### Slicing a memoryview
#

@cname('__pyx_memview_slice')
cdef memoryview memview_slice(memoryview memview, object indices):
    cdef int new_ndim = 0, suboffset_dim = -1, dim
    cdef bint negative_step
    cdef {{memviewslice_name}} src, dst
    cdef {{memviewslice_name}} *p_src

    # dst is copied by value in memoryview_fromslice -- initialize it
    # src is never copied
    memset(&dst, 0, sizeof(dst))

    cdef _memoryviewslice memviewsliceobj

    assert memview.view.ndim > 0

    if isinstance(memview, _memoryviewslice):
        memviewsliceobj = memview
        p_src = &memviewsliceobj.from_slice
    else:
        slice_copy(memview, &src)
        p_src = &src

    # Note: don't use variable src at this point
    # SubNote: we should be able to declare variables in blocks...

    # memoryview_fromslice() will inc our dst slice
    dst.memview = p_src.memview
    dst.data = p_src.data

    # Put everything in temps to avoid this bloody warning:
    # "Argument evaluation order in C function call is undefined and
    #  may not be as expected"
    cdef {{memviewslice_name}} *p_dst = &dst
    cdef int *p_suboffset_dim = &suboffset_dim
    cdef Py_ssize_t start, stop, step
    cdef bint have_start, have_stop, have_step

    for dim, index in enumerate(indices):
        if PyIndex_Check(index):
            slice_memviewslice(
                p_dst, p_src.shape[dim], p_src.strides[dim], p_src.suboffsets[dim],
                dim, new_ndim, p_suboffset_dim,
                index, 0, 0, # start, stop, step
                0, 0, 0, # have_{start,stop,step}
                False)
        elif index is None:
            p_dst.shape[new_ndim] = 1
            p_dst.strides[new_ndim] = 0
            p_dst.suboffsets[new_ndim] = -1
            new_ndim += 1
        else:
            start = index.start or 0
            stop = index.stop or 0
            step = index.step or 0

            have_start = index.start is not None
            have_stop = index.stop is not None
            have_step = index.step is not None

            slice_memviewslice(
                p_dst, p_src.shape[dim], p_src.strides[dim], p_src.suboffsets[dim],
                dim, new_ndim, p_suboffset_dim,
                start, stop, step,
                have_start, have_stop, have_step,
                True)
            new_ndim += 1

    if isinstance(memview, _memoryviewslice):
        return memoryview_fromslice(dst, new_ndim,
                                    memviewsliceobj.to_object_func,
                                    memviewsliceobj.to_dtype_func,
                                    memview.dtype_is_object)
    else:
        return memoryview_fromslice(dst, new_ndim, NULL, NULL,
                                    memview.dtype_is_object)


#
### Slicing in a single dimension of a memoryviewslice
#

cdef extern from "<stdlib.h>":
    void abort() nogil
    void printf(char *s, ...) nogil

cdef extern from "<stdio.h>":
    ctypedef struct FILE
    FILE *stderr
    int fputs(char *s, FILE *stream)

cdef extern from "pystate.h":
    void PyThreadState_Get() nogil

    # These are not actually nogil, but we check for the GIL before calling them
    void PyErr_SetString(PyObject *type, char *msg) nogil
    PyObject *PyErr_Format(PyObject *exc, char *msg, ...) nogil

@cname('__pyx_memoryview_slice_memviewslice')
cdef int slice_memviewslice(
        {{memviewslice_name}} *dst,
        Py_ssize_t shape, Py_ssize_t stride, Py_ssize_t suboffset,
        int dim, int new_ndim, int *suboffset_dim,
        Py_ssize_t start, Py_ssize_t stop, Py_ssize_t step,
        int have_start, int have_stop, int have_step,
        bint is_slice) nogil except -1:
    """
    Create a new slice dst given slice src.

    dim             - the current src dimension (indexing will make dimensions
                                                 disappear)
    new_dim         - the new dst dimension
    suboffset_dim   - pointer to a single int initialized to -1 to keep track of
                      where slicing offsets should be added
    """

    cdef Py_ssize_t new_shape
    cdef bint negative_step

    if not is_slice:
        # index is a normal integer-like index
        if start < 0:
            start += shape
        if not 0 <= start < shape:
            _err_dim(IndexError, "Index out of bounds (axis %d)", dim)
    else:
        # index is a slice
        negative_step = have_step != 0 and step < 0

        if have_step and step == 0:
            _err_dim(ValueError, "Step may not be zero (axis %d)", dim)

        # check our bounds and set defaults
        if have_start:
            if start < 0:
                start += shape
                if start < 0:
                    start = 0
            elif start >= shape:
                if negative_step:
                    start = shape - 1
                else:
                    start = shape
        else:
            if negative_step:
                start = shape - 1
            else:
                start = 0

        if have_stop:
            if stop < 0:
                stop += shape
                if stop < 0:
                    stop = 0
            elif stop > shape:
                stop = shape
        else:
            if negative_step:
                stop = -1
            else:
                stop = shape

        if not have_step:
            step = 1

        # len = ceil( (stop - start) / step )
        with cython.cdivision(True):
            new_shape = (stop - start) // step

            if (stop - start) - step * new_shape:
                new_shape += 1

        if new_shape < 0:
            new_shape = 0

        # shape/strides/suboffsets
        dst.strides[new_ndim] = stride * step
        dst.shape[new_ndim] = new_shape
        dst.suboffsets[new_ndim] = suboffset

    # Add the slicing or idexing offsets to the right suboffset or base data *
    if suboffset_dim[0] < 0:
        dst.data += start * stride
    else:
        dst.suboffsets[suboffset_dim[0]] += start * stride

    if suboffset >= 0:
        if not is_slice:
            if new_ndim == 0:
                dst.data = (<char **> dst.data)[0] + suboffset
            else:
                _err_dim(IndexError, "All dimensions preceding dimension %d "
                                     "must be indexed and not sliced", dim)
        else:
            suboffset_dim[0] = new_ndim

    return 0

#
### Index a memoryview
#
@cname('__pyx_pybuffer_index')
cdef char *pybuffer_index(Py_buffer *view, char *bufp, Py_ssize_t index,
                          Py_ssize_t dim) except NULL:
    cdef Py_ssize_t shape, stride, suboffset = -1
    cdef Py_ssize_t itemsize = view.itemsize
    cdef char *resultp

    if view.ndim == 0:
        shape = view.len / itemsize
        stride = itemsize
    else:
        shape = view.shape[dim]
        stride = view.strides[dim]
        if view.suboffsets != NULL:
            suboffset = view.suboffsets[dim]

    if index < 0:
        index += view.shape[dim]
        if index < 0:
            raise IndexError("Out of bounds on buffer access (axis %d)" % dim)

    if index >= shape:
        raise IndexError("Out of bounds on buffer access (axis %d)" % dim)

    resultp = bufp + index * stride
    if suboffset >= 0:
        resultp = (<char **> resultp)[0] + suboffset

    return resultp

#
### Transposing a memoryviewslice
#
@cname('__pyx_memslice_transpose')
cdef int transpose_memslice({{memviewslice_name}} *memslice) nogil except 0:
    cdef int ndim = memslice.memview.view.ndim

    cdef Py_ssize_t *shape = memslice.shape
    cdef Py_ssize_t *strides = memslice.strides

    # reverse strides and shape
    cdef int i, j
    for i in range(ndim / 2):
        j = ndim - 1 - i
        strides[i], strides[j] = strides[j], strides[i]
        shape[i], shape[j] = shape[j], shape[i]

        if memslice.suboffsets[i] >= 0 or memslice.suboffsets[j] >= 0:
            _err(ValueError, "Cannot transpose memoryview with indirect dimensions")

    return 1

#
### Creating new memoryview objects from slices and memoryviews
#
@cname('__pyx_memoryviewslice')
cdef class _memoryviewslice(memoryview):
    "Internal class for passing memoryview slices to Python"

    # We need this to keep our shape/strides/suboffset pointers valid
    cdef {{memviewslice_name}} from_slice
    # We need this only to print it's class' name
    cdef object from_object

    cdef object (*to_object_func)(char *)
    cdef int (*to_dtype_func)(char *, object) except 0

    def __dealloc__(self):
        __PYX_XDEC_MEMVIEW(&self.from_slice, 1)

    cdef convert_item_to_object(self, char *itemp):
        if self.to_object_func != NULL:
            return self.to_object_func(itemp)
        else:
            return memoryview.convert_item_to_object(self, itemp)

    cdef assign_item_from_object(self, char *itemp, object value):
        if self.to_dtype_func != NULL:
            self.to_dtype_func(itemp, value)
        else:
            memoryview.assign_item_from_object(self, itemp, value)

    @property
    def base(self):
        return self.from_object

    __pyx_getbuffer = capsule(<void *> &__pyx_memoryview_getbuffer, "getbuffer(obj, view, flags)")


@cname('__pyx_memoryview_fromslice')
cdef memoryview_fromslice({{memviewslice_name}} memviewslice,
                          int ndim,
                          object (*to_object_func)(char *),
                          int (*to_dtype_func)(char *, object) except 0,
                          bint dtype_is_object):

    cdef _memoryviewslice result

    if <PyObject *> memviewslice.memview == Py_None:
        return None

    # assert 0 < ndim <= memviewslice.memview.view.ndim, (
    #                 ndim, memviewslice.memview.view.ndim)

    result = _memoryviewslice(None, 0, dtype_is_object)

    result.from_slice = memviewslice
    __PYX_INC_MEMVIEW(&memviewslice, 1)

    result.from_object = (<memoryview> memviewslice.memview).base
    result.typeinfo = memviewslice.memview.typeinfo

    result.view = memviewslice.memview.view
    result.view.buf = <void *> memviewslice.data
    result.view.ndim = ndim
    (<__pyx_buffer *> &result.view).obj = Py_None
    Py_INCREF(Py_None)

    if (<memoryview>memviewslice.memview).flags & PyBUF_WRITABLE:
        result.flags = PyBUF_RECORDS
    else:
        result.flags = PyBUF_RECORDS_RO

    result.view.shape = <Py_ssize_t *> result.from_slice.shape
    result.view.strides = <Py_ssize_t *> result.from_slice.strides

    # only set suboffsets if actually used, otherwise set to NULL to improve compatibility
    result.view.suboffsets = NULL
    for suboffset in result.from_slice.suboffsets[:ndim]:
        if suboffset >= 0:
            result.view.suboffsets = <Py_ssize_t *> result.from_slice.suboffsets
            break

    result.view.len = result.view.itemsize
    for length in result.view.shape[:ndim]:
        result.view.len *= length

    result.to_object_func = to_object_func
    result.to_dtype_func = to_dtype_func

    return result

@cname('__pyx_memoryview_get_slice_from_memoryview')
cdef {{memviewslice_name}} *get_slice_from_memview(memoryview memview,
                                                   {{memviewslice_name}} *mslice) except NULL:
    cdef _memoryviewslice obj
    if isinstance(memview, _memoryviewslice):
        obj = memview
        return &obj.from_slice
    else:
        slice_copy(memview, mslice)
        return mslice

@cname('__pyx_memoryview_slice_copy')
cdef void slice_copy(memoryview memview, {{memviewslice_name}} *dst):
    cdef int dim
    cdef (Py_ssize_t*) shape, strides, suboffsets

    shape = memview.view.shape
    strides = memview.view.strides
    suboffsets = memview.view.suboffsets

    dst.memview = <__pyx_memoryview *> memview
    dst.data = <char *> memview.view.buf

    for dim in range(memview.view.ndim):
        dst.shape[dim] = shape[dim]
        dst.strides[dim] = strides[dim]
        dst.suboffsets[dim] = suboffsets[dim] if suboffsets else -1

@cname('__pyx_memoryview_copy_object')
cdef memoryview_copy(memoryview memview):
    "Create a new memoryview object"
    cdef {{memviewslice_name}} memviewslice
    slice_copy(memview, &memviewslice)
    return memoryview_copy_from_slice(memview, &memviewslice)

@cname('__pyx_memoryview_copy_object_from_slice')
cdef memoryview_copy_from_slice(memoryview memview, {{memviewslice_name}} *memviewslice):
    """
    Create a new memoryview object from a given memoryview object and slice.
    """
    cdef object (*to_object_func)(char *)
    cdef int (*to_dtype_func)(char *, object) except 0

    if isinstance(memview, _memoryviewslice):
        to_object_func = (<_memoryviewslice> memview).to_object_func
        to_dtype_func = (<_memoryviewslice> memview).to_dtype_func
    else:
        to_object_func = NULL
        to_dtype_func = NULL

    return memoryview_fromslice(memviewslice[0], memview.view.ndim,
                                to_object_func, to_dtype_func,
                                memview.dtype_is_object)


#
### Copy the contents of a memoryview slices
#
cdef Py_ssize_t abs_py_ssize_t(Py_ssize_t arg) nogil:
    if arg < 0:
        return -arg
    else:
        return arg

@cname('__pyx_get_best_slice_order')
cdef char get_best_order({{memviewslice_name}} *mslice, int ndim) nogil:
    """
    Figure out the best memory access order for a given slice.
    """
    cdef int i
    cdef Py_ssize_t c_stride = 0
    cdef Py_ssize_t f_stride = 0

    for i in range(ndim - 1, -1, -1):
        if mslice.shape[i] > 1:
            c_stride = mslice.strides[i]
            break

    for i in range(ndim):
        if mslice.shape[i] > 1:
            f_stride = mslice.strides[i]
            break

    if abs_py_ssize_t(c_stride) <= abs_py_ssize_t(f_stride):
        return 'C'
    else:
        return 'F'

@cython.cdivision(True)
cdef void _copy_strided_to_strided(char *src_data, Py_ssize_t *src_strides,
                                   char *dst_data, Py_ssize_t *dst_strides,
                                   Py_ssize_t *src_shape, Py_ssize_t *dst_shape,
                                   int ndim, size_t itemsize) nogil:
    # Note: src_extent is 1 if we're broadcasting
    # dst_extent always >= src_extent as we don't do reductions
    cdef Py_ssize_t i
    cdef Py_ssize_t src_extent = src_shape[0]
    cdef Py_ssize_t dst_extent = dst_shape[0]
    cdef Py_ssize_t src_stride = src_strides[0]
    cdef Py_ssize_t dst_stride = dst_strides[0]

    if ndim == 1:
       if (src_stride > 0 and dst_stride > 0 and
           <size_t> src_stride == itemsize == <size_t> dst_stride):
           memcpy(dst_data, src_data, itemsize * dst_extent)
       else:
           for i in range(dst_extent):
               memcpy(dst_data, src_data, itemsize)
               src_data += src_stride
               dst_data += dst_stride
    else:
        for i in range(dst_extent):
            _copy_strided_to_strided(src_data, src_strides + 1,
                                     dst_data, dst_strides + 1,
                                     src_shape + 1, dst_shape + 1,
                                     ndim - 1, itemsize)
            src_data += src_stride
            dst_data += dst_stride

cdef void copy_strided_to_strided({{memviewslice_name}} *src,
                                  {{memviewslice_name}} *dst,
                                  int ndim, size_t itemsize) nogil:
    _copy_strided_to_strided(src.data, src.strides, dst.data, dst.strides,
                             src.shape, dst.shape, ndim, itemsize)

@cname('__pyx_memoryview_slice_get_size')
cdef Py_ssize_t slice_get_size({{memviewslice_name}} *src, int ndim) nogil:
    "Return the size of the memory occupied by the slice in number of bytes"
    cdef Py_ssize_t shape, size = src.memview.view.itemsize

    for shape in src.shape[:ndim]:
        size *= shape

    return size

@cname('__pyx_fill_contig_strides_array')
cdef Py_ssize_t fill_contig_strides_array(
                Py_ssize_t *shape, Py_ssize_t *strides, Py_ssize_t stride,
                int ndim, char order) nogil:
    """
    Fill the strides array for a slice with C or F contiguous strides.
    This is like PyBuffer_FillContiguousStrides, but compatible with py < 2.6
    """
    cdef int idx

    if order == 'F':
        for idx in range(ndim):
            strides[idx] = stride
            stride *= shape[idx]
    else:
        for idx in range(ndim - 1, -1, -1):
            strides[idx] = stride
            stride *= shape[idx]

    return stride

@cname('__pyx_memoryview_copy_data_to_temp')
cdef void *copy_data_to_temp({{memviewslice_name}} *src,
                             {{memviewslice_name}} *tmpslice,
                             char order,
                             int ndim) nogil except NULL:
    """
    Copy a direct slice to temporary contiguous memory. The caller should free
    the result when done.
    """
    cdef int i
    cdef void *result

    cdef size_t itemsize = src.memview.view.itemsize
    cdef size_t size = slice_get_size(src, ndim)

    result = malloc(size)
    if not result:
        _err(MemoryError, NULL)

    # tmpslice[0] = src
    tmpslice.data = <char *> result
    tmpslice.memview = src.memview
    for i in range(ndim):
        tmpslice.shape[i] = src.shape[i]
        tmpslice.suboffsets[i] = -1

    fill_contig_strides_array(&tmpslice.shape[0], &tmpslice.strides[0], itemsize,
                              ndim, order)

    # We need to broadcast strides again
    for i in range(ndim):
        if tmpslice.shape[i] == 1:
            tmpslice.strides[i] = 0

    if slice_is_contig(src[0], order, ndim):
        memcpy(result, src.data, size)
    else:
        copy_strided_to_strided(src, tmpslice, ndim, itemsize)

    return result

# Use 'with gil' functions and avoid 'with gil' blocks, as the code within the blocks
# has temporaries that need the GIL to clean up
@cname('__pyx_memoryview_err_extents')
cdef int _err_extents(int i, Py_ssize_t extent1,
                             Py_ssize_t extent2) except -1 with gil:
    raise ValueError("got differing extents in dimension %d (got %d and %d)" %
                                                        (i, extent1, extent2))

@cname('__pyx_memoryview_err_dim')
cdef int _err_dim(object error, char *msg, int dim) except -1 with gil:
    raise error(msg.decode('ascii') % dim)

@cname('__pyx_memoryview_err')
cdef int _err(object error, char *msg) except -1 with gil:
    if msg != NULL:
        raise error(msg.decode('ascii'))
    else:
        raise error

@cname('__pyx_memoryview_copy_contents')
cdef int memoryview_copy_contents({{memviewslice_name}} src,
                                  {{memviewslice_name}} dst,
                                  int src_ndim, int dst_ndim,
                                  bint dtype_is_object) nogil except -1:
    """
    Copy memory from slice src to slice dst.
    Check for overlapping memory and verify the shapes.
    """
    cdef void *tmpdata = NULL
    cdef size_t itemsize = src.memview.view.itemsize
    cdef int i
    cdef char order = get_best_order(&src, src_ndim)
    cdef bint broadcasting = False
    cdef bint direct_copy = False
    cdef {{memviewslice_name}} tmp

    if src_ndim < dst_ndim:
        broadcast_leading(&src, src_ndim, dst_ndim)
    elif dst_ndim < src_ndim:
        broadcast_leading(&dst, dst_ndim, src_ndim)

    cdef int ndim = max(src_ndim, dst_ndim)

    for i in range(ndim):
        if src.shape[i] != dst.shape[i]:
            if src.shape[i] == 1:
                broadcasting = True
                src.strides[i] = 0
            else:
                _err_extents(i, dst.shape[i], src.shape[i])

        if src.suboffsets[i] >= 0:
            _err_dim(ValueError, "Dimension %d is not direct", i)

    if slices_overlap(&src, &dst, ndim, itemsize):
        # slices overlap, copy to temp, copy temp to dst
        if not slice_is_contig(src, order, ndim):
            order = get_best_order(&dst, ndim)

        tmpdata = copy_data_to_temp(&src, &tmp, order, ndim)
        src = tmp

    if not broadcasting:
        # See if both slices have equal contiguity, in that case perform a
        # direct copy. This only works when we are not broadcasting.
        if slice_is_contig(src, 'C', ndim):
            direct_copy = slice_is_contig(dst, 'C', ndim)
        elif slice_is_contig(src, 'F', ndim):
            direct_copy = slice_is_contig(dst, 'F', ndim)

        if direct_copy:
            # Contiguous slices with same order
            refcount_copying(&dst, dtype_is_object, ndim, False)
            memcpy(dst.data, src.data, slice_get_size(&src, ndim))
            refcount_copying(&dst, dtype_is_object, ndim, True)
            free(tmpdata)
            return 0

    if order == 'F' == get_best_order(&dst, ndim):
        # see if both slices have Fortran order, transpose them to match our
        # C-style indexing order
        transpose_memslice(&src)
        transpose_memslice(&dst)

    refcount_copying(&dst, dtype_is_object, ndim, False)
    copy_strided_to_strided(&src, &dst, ndim, itemsize)
    refcount_copying(&dst, dtype_is_object, ndim, True)

    free(tmpdata)
    return 0

@cname('__pyx_memoryview_broadcast_leading')
cdef void broadcast_leading({{memviewslice_name}} *mslice,
                            int ndim,
                            int ndim_other) nogil:
    cdef int i
    cdef int offset = ndim_other - ndim

    for i in range(ndim - 1, -1, -1):
        mslice.shape[i + offset] = mslice.shape[i]
        mslice.strides[i + offset] = mslice.strides[i]
        mslice.suboffsets[i + offset] = mslice.suboffsets[i]

    for i in range(offset):
        mslice.shape[i] = 1
        mslice.strides[i] = mslice.strides[0]
        mslice.suboffsets[i] = -1

#
### Take care of refcounting the objects in slices. Do this separately from any copying,
### to minimize acquiring the GIL
#

@cname('__pyx_memoryview_refcount_copying')
cdef void refcount_copying({{memviewslice_name}} *dst, bint dtype_is_object,
                           int ndim, bint inc) nogil:
    # incref or decref the objects in the destination slice if the dtype is
    # object
    if dtype_is_object:
        refcount_objects_in_slice_with_gil(dst.data, dst.shape,
                                           dst.strides, ndim, inc)

@cname('__pyx_memoryview_refcount_objects_in_slice_with_gil')
cdef void refcount_objects_in_slice_with_gil(char *data, Py_ssize_t *shape,
                                             Py_ssize_t *strides, int ndim,
                                             bint inc) with gil:
    refcount_objects_in_slice(data, shape, strides, ndim, inc)

@cname('__pyx_memoryview_refcount_objects_in_slice')
cdef void refcount_objects_in_slice(char *data, Py_ssize_t *shape,
                                    Py_ssize_t *strides, int ndim, bint inc):
    cdef Py_ssize_t i

    for i in range(shape[0]):
        if ndim == 1:
            if inc:
                Py_INCREF((<PyObject **> data)[0])
            else:
                Py_DECREF((<PyObject **> data)[0])
        else:
            refcount_objects_in_slice(data, shape + 1, strides + 1,
                                      ndim - 1, inc)

        data += strides[0]

#
### Scalar to slice assignment
#
@cname('__pyx_memoryview_slice_assign_scalar')
cdef void slice_assign_scalar({{memviewslice_name}} *dst, int ndim,
                              size_t itemsize, void *item,
                              bint dtype_is_object) nogil:
    refcount_copying(dst, dtype_is_object, ndim, False)
    _slice_assign_scalar(dst.data, dst.shape, dst.strides, ndim,
                         itemsize, item)
    refcount_copying(dst, dtype_is_object, ndim, True)


@cname('__pyx_memoryview__slice_assign_scalar')
cdef void _slice_assign_scalar(char *data, Py_ssize_t *shape,
                              Py_ssize_t *strides, int ndim,
                              size_t itemsize, void *item) nogil:
    cdef Py_ssize_t i
    cdef Py_ssize_t stride = strides[0]
    cdef Py_ssize_t extent = shape[0]

    if ndim == 1:
        for i in range(extent):
            memcpy(data, item, itemsize)
            data += stride
    else:
        for i in range(extent):
            _slice_assign_scalar(data, shape + 1, strides + 1,
                                ndim - 1, itemsize, item)
            data += stride


############### BufferFormatFromTypeInfo ###############
cdef extern from *:
    ctypedef struct __Pyx_StructField

    cdef enum:
        __PYX_BUF_FLAGS_PACKED_STRUCT
        __PYX_BUF_FLAGS_INTEGER_COMPLEX

    ctypedef struct __Pyx_TypeInfo:
      char* name
      __Pyx_StructField* fields
      size_t size
      size_t arraysize[8]
      int ndim
      char typegroup
      char is_unsigned
      int flags

    ctypedef struct __Pyx_StructField:
      __Pyx_TypeInfo* type
      char* name
      size_t offset

    ctypedef struct __Pyx_BufFmt_StackElem:
      __Pyx_StructField* field
      size_t parent_offset

    #ctypedef struct __Pyx_BufFmt_Context:
    #  __Pyx_StructField root
      __Pyx_BufFmt_StackElem* head

    struct __pyx_typeinfo_string:
        char string[3]

    __pyx_typeinfo_string __Pyx_TypeInfoToFormat(__Pyx_TypeInfo *)


@cname('__pyx_format_from_typeinfo')
cdef bytes format_from_typeinfo(__Pyx_TypeInfo *type):
    cdef __Pyx_StructField *field
    cdef __pyx_typeinfo_string fmt
    cdef bytes part, result

    if type.typegroup == 'S':
        assert type.fields != NULL
        assert type.fields.type != NULL

        if type.flags & __PYX_BUF_FLAGS_PACKED_STRUCT:
            alignment = b'^'
        else:
            alignment = b''

        parts = [b"T{"]
        field = type.fields

        while field.type:
            part = format_from_typeinfo(field.type)
            parts.append(part + b':' + field.name + b':')
            field += 1

        result = alignment.join(parts) + b'}'
    else:
        fmt = __Pyx_TypeInfoToFormat(type)
        if type.arraysize[0]:
            extents = [unicode(type.arraysize[i]) for i in range(type.ndim)]
            result = (u"(%s)" % u','.join(extents)).encode('ascii') + fmt.string
        else:
            result = fmt.string

    return result