-
Notifications
You must be signed in to change notification settings - Fork 216
/
device.py
2026 lines (1865 loc) · 81 KB
/
device.py
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
"""
Module implementing the C++ "standalone" device.
"""
import inspect
import itertools
import numbers
import os
import shutil
import subprocess
import sys
import tempfile
import time
import zlib
from collections import Counter, defaultdict
from collections.abc import Mapping
from distutils import ccompiler
from hashlib import md5
import numpy as np
import brian2
from brian2.codegen.codeobject import check_compiler_kwds
from brian2.codegen.cpp_prefs import get_compiler_and_args, get_msvc_env
from brian2.codegen.generators.cpp_generator import c_data_type
from brian2.core.functions import Function
from brian2.core.namespace import get_local_namespace
from brian2.core.preferences import BrianPreference, prefs
from brian2.core.variables import (
ArrayVariable,
Constant,
DynamicArrayVariable,
Variable,
VariableView,
)
from brian2.devices.device import Device, all_devices, reset_device, set_device
from brian2.groups.group import Group
from brian2.input import TimedArray
from brian2.parsing.rendering import CPPNodeRenderer
from brian2.synapses.synapses import Synapses
from brian2.units import second
from brian2.units.fundamentalunits import Quantity, fail_for_dimension_mismatch
from brian2.utils.filelock import FileLock
from brian2.utils.filetools import copy_directory, ensure_directory, in_directory
from brian2.utils.logger import get_logger, std_silent
from brian2.utils.stringtools import word_substitute
from .codeobject import CPPStandaloneCodeObject, openmp_pragma
__all__ = []
logger = get_logger(__name__)
# Preferences
prefs.register_preferences(
"devices.cpp_standalone",
"C++ standalone preferences ",
openmp_threads=BrianPreference(
default=0,
docs="""
The number of threads to use if OpenMP is turned on. By default, this value is set to 0 and the C++ code
is generated without any reference to OpenMP. If greater than 0, then the corresponding number of threads
are used to launch the simulation.
""",
),
openmp_spatialneuron_strategy=BrianPreference(
default=None,
validator=lambda val: val in [None, "branches", "systems"],
docs="""
DEPRECATED. Previously used to chose the strategy to parallelize the
solution of the three tridiagonal systems for multicompartmental
neurons. Now, its value is ignored.
""",
),
make_cmd_unix=BrianPreference(
default="make",
docs="""
The make command used to compile the standalone project. Defaults to the
standard GNU make commane "make".""",
),
run_cmd_unix=BrianPreference(
default="./main",
validator=lambda val: isinstance(val, str) or isinstance(val, list),
docs="""
The command used to run the compiled standalone project. Defaults to executing
the compiled binary with "./main". Must be a single binary as string or a list
of command arguments (e.g. ["./binary", "--key", "value"]).
""",
),
extra_make_args_unix=BrianPreference(
default=["-j"],
docs="""
Additional flags to pass to the GNU make command on Linux/OS-X.
Defaults to "-j" for parallel compilation.""",
),
extra_make_args_windows=BrianPreference(
default=[],
docs="""
Additional flags to pass to the nmake command on Windows. By default, no
additional flags are passed.
""",
),
run_environment_variables=BrianPreference(
default={"LD_BIND_NOW": "1"},
docs="""
Dictionary of environment variables and their values that will be set
during the execution of the standalone code.
""",
),
)
class CPPWriter:
def __init__(self, project_dir):
self.project_dir = project_dir
self.source_files = set()
self.header_files = set()
def write(self, filename, contents):
logger.diagnostic(f"Writing file {filename}:\n{contents}")
if filename.lower().endswith(".cpp") or filename.lower().endswith(".c"):
self.source_files.add(filename)
elif filename.lower().endswith(".h"):
self.header_files.add(filename)
elif filename.endswith(".*"):
self.write(f"{filename[:-1]}cpp", contents.cpp_file)
self.write(f"{filename[:-1]}h", contents.h_file)
return
fullfilename = os.path.join(self.project_dir, filename)
if os.path.exists(fullfilename):
with open(fullfilename, encoding="utf8") as f:
if f.read() == contents:
return
with open(fullfilename, "w", encoding="utf8") as f:
f.write(contents)
def invert_dict(x):
return {v: k for k, v in x.items()}
class CPPStandaloneDevice(Device):
"""
The `Device` used for C++ standalone simulations.
"""
def __init__(self):
super().__init__()
#: Dictionary mapping `ArrayVariable` objects to their globally
#: unique name
self.arrays = {}
#: Dictionary mapping `ArrayVariable` objects to their value or to
#: ``None`` if the value (potentially) depends on executed code. This
#: mechanism allows to access state variables in standalone mode if
#: their value is known at run time
self.array_cache = {}
#: List of all dynamic arrays
#: Dictionary mapping `DynamicArrayVariable` objects with 1 dimension to
#: their globally unique name
self.dynamic_arrays = {}
#: Dictionary mapping `DynamicArrayVariable` objects with 2 dimensions
#: to their globally unique name
self.dynamic_arrays_2d = {}
#: List of all arrays to be filled with zeros (list of (var, varname) )
self.zero_arrays = []
#: List of all arrays to be filled with numbers (list of
#: (var, varname, start) tuples
self.arange_arrays = []
#: Set of all existing synapses
self.synapses = set()
#: Whether the simulation has been run
self.has_been_run = False
#: Whether apply_run_args has been called
self.run_args_applied = False
#: Whether a run should trigger a build
self.build_on_run = False
#: build options
self.build_options = None
#: The directory which contains the generated code and results
self.project_dir = None
#: The directory which contains the results (relative to `project_dir``)
self.results_dir = None
#: Whether to generate profiling information (stored in an instance
#: variable to be accessible during CodeObject generation)
self.enable_profiling = False
#: CodeObjects that use profiling (users can potentially enable
#: profiling only for a subset of runs)
self.profiled_codeobjects = []
#: Dict of all static saved arrays
self.static_arrays = {}
#: Dict of all TimedArray objects
self.timed_arrays = {}
self.code_objects = {}
self.main_queue = []
self.runfuncs = {}
self.networks = set()
self.static_array_specs = []
self.report_func = ""
#: Code lines that have been manually added with `device.insert_code`
#: Dictionary mapping slot names to lists of lines.
#: Note that the main slot is handled separately as part of `main_queue`
self.code_lines = {
"before_start": [],
"after_start": [],
"before_network_run": [],
"after_network_run": [],
"before_end": [],
"after_end": [],
}
#: Dictionary storing compile and binary execution times
self.timers = {"run_binary": None, "compile": {"clean": None, "make": None}}
self.clocks = set()
self.extra_compile_args = []
self.define_macros = []
self.headers = []
self.include_dirs = ["brianlib/randomkit"]
self.library_dirs = ["brianlib/randomkit"]
self.runtime_library_dirs = []
self.run_environment_variables = {}
if sys.platform.startswith("darwin"):
if "DYLD_LIBRARY_PATH" in os.environ:
dyld_library_path = f"{os.environ['DYLD_LIBRARY_PATH']}:{os.path.join(sys.prefix, 'lib')}"
else:
dyld_library_path = os.path.join(sys.prefix, "lib")
self.run_environment_variables["DYLD_LIBRARY_PATH"] = dyld_library_path
self.libraries = []
if sys.platform == "win32":
self.libraries += ["advapi32"]
self.extra_link_args = []
self.writer = None
def reinit(self):
# Remember the build_on_run setting and its options -- important during
# testing
build_on_run = self.build_on_run
build_options = self.build_options
self.__init__()
super().reinit()
self.build_on_run = build_on_run
self.build_options = build_options
def spike_queue(self, source_start, source_end):
return None # handled differently
def freeze(self, code, ns):
# TODO: Remove this function at some point
logger.warn(
"The CPPStandaloneDevice.freeze function should no longer "
"be used, add constant definitions directly to the "
'code in the "CONSTANTS" section instead.',
name_suffix="deprecated_freeze_use",
once=True,
)
# this is a bit of a hack, it should be passed to the template somehow
for k, v in ns.items():
if isinstance(v, Variable) and v.scalar and v.constant and v.read_only:
try:
v = v.get_value()
except NotImplementedError:
continue
if isinstance(v, str):
code = word_substitute(code, {k: v})
elif isinstance(v, numbers.Number):
# Use a renderer to correctly transform constants such as True or inf
renderer = CPPNodeRenderer()
string_value = renderer.render_expr(repr(v))
if prefs.core.default_float_dtype == np.float32 and isinstance(
v, (float, np.float32, np.float64)
):
string_value += "f"
if v < 0:
string_value = f"({string_value})"
code = word_substitute(code, {k: string_value})
else:
pass # don't deal with this object
return code
def insert_code(self, slot, code):
"""
Insert code directly into main.cpp
"""
if slot == "main":
self.main_queue.append(("insert_code", code))
elif slot in self.code_lines:
self.code_lines[slot].append(code)
else:
logger.warn(f"Ignoring device code, unknown slot: {slot}, code: {code}")
def apply_run_args(self):
if self.run_args_applied:
raise RuntimeError(
"The 'apply_run_args()' function can only be called once."
)
self.insert_code("main", "set_from_command_line(args);")
self.run_args_applied = True
def static_array(self, name, arr):
arr = np.atleast_1d(arr)
assert len(arr), f"length for {name}: {len(arr)}"
name = f"_static_array_{name}"
basename = name
i = 0
while name in self.static_arrays:
i += 1
name = f"{basename}_{str(i)}"
self.static_arrays[name] = arr.copy()
return name
def get_array_name(self, var, access_data=True):
"""
Return a globally unique name for `var`.
Parameters
----------
access_data : bool, optional
For `DynamicArrayVariable` objects, specifying `True` here means the
name for the underlying data is returned. If specifying `False`,
the name of object itself is returned (e.g. to allow resizing).
"""
if isinstance(var, DynamicArrayVariable):
if access_data:
return self.arrays[var]
elif var.ndim == 1:
return self.dynamic_arrays[var]
else:
return self.dynamic_arrays_2d[var]
elif isinstance(var, ArrayVariable):
return self.arrays[var]
else:
raise TypeError(f"Do not have a name for variable of type {type(var)}.")
def get_array_filename(self, var, basedir=None):
"""
Return a file name for a variable.
Parameters
----------
var : `ArrayVariable`
The variable to get a filename for.
basedir : str
The base directory for the filename, defaults to ``'results'``.
DEPRECATED: Will raise an error if specified.
Returns
-------
filename : str
A filename of the form
``varname+'_'+str(zlib.crc32(varname))``, where varname
is the name returned by `get_array_name`.
Notes
-----
The reason that the filename is not simply ``varname`` is
that this could lead to file names that are not unique in file systems
that are not case sensitive (e.g. on Windows).
"""
if basedir is not None:
raise ValueError("Specifying 'basedir' is no longer supported.")
varname = self.get_array_name(var, access_data=False)
return f"{varname}_{str(zlib.crc32(varname.encode('utf-8')))}"
def add_array(self, var):
# Note that a dynamic array variable is added to both the arrays and
# the _dynamic_array dictionary
if isinstance(var, DynamicArrayVariable):
# The code below is slightly more complicated than just looking
# for a unique name as above for static_array, the name has
# potentially to be unique for more than one dictionary, with
# different prefixes. This is because dynamic arrays are added to
# a ``dynamic_arrays`` dictionary (with a `_dynamic` prefix) and to
# the general ``arrays`` dictionary. We want to make sure that we
# use the same name in the two dictionaries, not for example
# ``_dynamic_array_source_name_2`` and ``_array_source_name_1``
# (this would work fine, but it would make the code harder to read).
orig_dynamic_name = dynamic_name = (
f"_dynamic_array_{var.owner.name}_{var.name}"
)
orig_array_name = array_name = f"_array_{var.owner.name}_{var.name}"
suffix = 0
if var.ndim == 1:
dynamic_dict = self.dynamic_arrays
elif var.ndim == 2:
dynamic_dict = self.dynamic_arrays_2d
else:
raise AssertionError(
"Did not expect a dynamic array with {var.ndim} dimensions."
)
while (
dynamic_name in dynamic_dict.values()
or array_name in self.arrays.values()
):
suffix += 1
dynamic_name = f"{orig_dynamic_name}_{int(suffix)}"
array_name = f"{orig_array_name}_{int(suffix)}"
dynamic_dict[var] = dynamic_name
self.arrays[var] = array_name
else:
orig_array_name = array_name = f"_array_{var.owner.name}_{var.name}"
suffix = 0
while array_name in self.arrays.values():
suffix += 1
array_name = f"{orig_array_name}_{int(suffix)}"
self.arrays[var] = array_name
def init_with_zeros(self, var, dtype):
if isinstance(var, DynamicArrayVariable):
varname = f"_dynamic{self.arrays[var]}"
else:
varname = self.arrays[var]
self.zero_arrays.append((var, varname))
self.array_cache[var] = np.zeros(var.size, dtype=dtype)
def init_with_arange(self, var, start, dtype):
if isinstance(var, DynamicArrayVariable):
varname = f"_dynamic{self.arrays[var]}"
else:
varname = self.arrays[var]
self.arange_arrays.append((var, varname, start))
self.array_cache[var] = np.arange(0, var.size, dtype=dtype) + start
def fill_with_array(self, var, arr):
arr = np.asarray(arr)
if arr.size == 0:
return # nothing to do
array_name = self.get_array_name(var, access_data=False)
if isinstance(var, DynamicArrayVariable):
# We can never be sure about the size of a dynamic array, so
# we can't do correct broadcasting. Therefore, we do not cache
# them at all for now.
self.array_cache[var] = None
else:
new_arr = np.empty(var.size, dtype=var.dtype)
new_arr[:] = arr
self.array_cache[var] = new_arr
if arr.size == 1:
if var.size == 1:
value = CPPNodeRenderer().render_expr(repr(arr.item(0)))
# For a single assignment, generate a code line instead of storing the array
self.main_queue.append(("set_by_single_value", (array_name, 0, value)))
else:
self.main_queue.append(
(
"set_by_constant",
(array_name, arr.item(), isinstance(var, DynamicArrayVariable)),
)
)
else:
# Using the std::vector instead of a pointer to the underlying
# data for dynamic arrays is fast enough here and it saves us some
# additional work to set up the pointer
static_array_name = self.static_array(array_name, arr)
self.main_queue.append(
(
"set_by_array",
(
array_name,
static_array_name,
isinstance(var, DynamicArrayVariable),
),
)
)
def resize(self, var, new_size):
array_name = self.get_array_name(var, access_data=False)
self.main_queue.append(("resize_array", (array_name, new_size)))
def variableview_set_with_index_array(self, variableview, item, value, check_units):
if isinstance(item, slice) and item == slice(None):
item = "True"
value = Quantity(value)
if (
isinstance(item, int) or (isinstance(item, np.ndarray) and item.shape == ())
) and value.size == 1:
array_name = self.get_array_name(variableview.variable, access_data=False)
value_str = CPPNodeRenderer().render_expr(repr(np.asarray(value).item(0)))
if self.array_cache.get(variableview.variable, None) is not None:
self.array_cache[variableview.variable][item] = value
# For a single assignment, generate a code line instead of storing the array
self.main_queue.append(
("set_by_single_value", (array_name, item, value_str))
)
# Simple case where we don't have to do any indexing
elif item == "True" and variableview.index_var in ("_idx", "0"):
self.fill_with_array(variableview.variable, value)
else:
# We have to calculate indices. This will not work for synaptic
# variables
try:
indices = np.asarray(
variableview.indexing(item, index_var=variableview.index_var)
)
except NotImplementedError:
raise NotImplementedError(
f"Cannot set variable '{variableview.name}' "
"this way in standalone, try using "
"string expressions."
)
# Using the std::vector instead of a pointer to the underlying
# data for dynamic arrays is fast enough here and it saves us some
# additional work to set up the pointer
arrayname = self.get_array_name(variableview.variable, access_data=False)
if indices.shape != () and (
value.shape == () or (value.size == 1 and indices.size > 1)
):
value = np.repeat(value, indices.size)
elif value.shape != indices.shape and len(value) != len(indices):
raise ValueError(
"Provided values do not match the size "
"of the indices, "
f"{len(value)} != len(indices)."
)
staticarrayname_index = self.static_array(f"_index_{arrayname}", indices)
staticarrayname_value = self.static_array(f"_value_{arrayname}", value)
self.array_cache[variableview.variable] = None
self.main_queue.append(
(
"set_array_by_array",
(arrayname, staticarrayname_index, staticarrayname_value),
)
)
def get_value(self, var, access_data=True):
# Usually, we cannot retrieve the values of state variables in
# standalone scripts since their values might depend on the evaluation
# of expressions at runtime. For some variables we do know the value
# however (values that have been set with explicit values and not
# changed in code objects)
if self.array_cache.get(var, None) is not None:
return self.array_cache[var]
else:
# After the network has been run, we can retrieve the values from
# disk
if self.has_been_run:
dtype = var.dtype
fname = os.path.join(self.results_dir, self.get_array_filename(var))
with open(fname, "rb") as f:
data = np.fromfile(f, dtype=dtype)
# This is a bit of an heuristic, but our 2d dynamic arrays are
# only expanding in one dimension, we assume here that the
# other dimension has size 0 at the beginning
if isinstance(var.size, tuple) and len(var.size) == 2:
if var.size[0] * var.size[1] == len(data):
size = var.size
elif var.size[0] == 0:
size = (len(data) // var.size[1], var.size[1])
elif var.size[1] == 0:
size = (var.size[0], len(data) // var.size[0])
else:
raise IndexError(
"Do not now how to deal with 2d "
f"array of size {var.size!s}, the array on "
f"disk has length {len(data)}."
)
var.size = size
return data.reshape(var.size)
var.size = len(data)
return data
raise NotImplementedError(
"Cannot retrieve the values of state "
"variables in standalone code before the "
"simulation has been run."
)
def variableview_get_subexpression_with_index_array(
self, variableview, item, run_namespace=None
):
if not self.has_been_run:
raise NotImplementedError(
"Cannot retrieve the values of state "
"variables in standalone code before the "
"simulation has been run."
)
# Temporarily switch to the runtime device to evaluate the subexpression
# (based on the values stored on disk)
set_device("runtime")
result = VariableView.get_subexpression_with_index_array(
variableview, item, run_namespace=run_namespace
)
reset_device()
return result
def variableview_get_with_expression(self, variableview, code, run_namespace=None):
raise NotImplementedError(
"Cannot retrieve the values of state "
"variables with string expressions in "
"standalone scripts."
)
def code_object_class(self, codeobj_class=None, fallback_pref=None):
"""
Return `CodeObject` class (either `CPPStandaloneCodeObject` class or input)
Parameters
----------
codeobj_class : a `CodeObject` class, optional
If this is keyword is set to None or no arguments are given, this method will return
the default (`CPPStandaloneCodeObject` class).
fallback_pref : str, optional
For the cpp_standalone device this option is ignored.
Returns
-------
codeobj_class : class
The `CodeObject` class that should be used
"""
# Ignore the requested pref (used for optimization in runtime)
if codeobj_class is None:
return CPPStandaloneCodeObject
else:
return codeobj_class
def code_object(
self,
owner,
name,
abstract_code,
variables,
template_name,
variable_indices,
codeobj_class=None,
template_kwds=None,
override_conditional_write=None,
compiler_kwds=None,
):
if compiler_kwds is None:
compiler_kwds = {}
check_compiler_kwds(
compiler_kwds,
[
"headers",
"sources",
"define_macros",
"libraries",
"include_dirs",
"library_dirs",
"runtime_library_dirs",
],
"C++ standalone",
)
if template_kwds is None:
template_kwds = dict()
else:
template_kwds = dict(template_kwds)
# In standalone mode, the only place where we use additional header
# files is by inserting them into the template
codeobj_headers = compiler_kwds.get("headers", [])
template_kwds["user_headers"] = (
self.headers + prefs["codegen.cpp.headers"] + codeobj_headers
)
template_kwds["profiled"] = self.enable_profiling
do_not_invalidate = set()
if template_name == "synapses_create_array":
cache = self.array_cache
if (
cache[variables["N"]] is None
): # synapses have been previously created with code
# Nothing we can do
logger.debug(
f"Synapses for '{owner.name}' have previously been created with "
"code, we therefore cannot cache the synapses created with arrays "
f"via '{name}'",
name_suffix="code_created_synapses_exist",
)
else: # first time we create synapses, or all previous connect calls were with arrays
cache[variables["N"]][0] += variables["sources"].size
do_not_invalidate.add(variables["N"])
for var, value in [
(
variables["_synaptic_pre"],
variables["sources"].get_value()
+ variables["_source_offset"].get_value(),
),
(
variables["_synaptic_post"],
variables["targets"].get_value()
+ variables["_target_offset"].get_value(),
),
]:
cache[var] = np.append(
cache.get(var, np.empty(0, dtype=int)), value
)
do_not_invalidate.add(var)
codeobj = super().code_object(
owner,
name,
abstract_code,
variables,
template_name,
variable_indices,
codeobj_class=codeobj_class,
template_kwds=template_kwds,
override_conditional_write=override_conditional_write,
compiler_kwds=compiler_kwds,
)
self.code_objects[codeobj.name] = codeobj
if self.enable_profiling:
self.profiled_codeobjects.append(codeobj.name)
for var in codeobj.variables.values():
if isinstance(var, TimedArray):
self.timed_arrays[var] = var.name
# We mark all writeable (i.e. not read-only) variables used by the code
# as "dirty" to avoid that the cache contains incorrect values. This
# might remove a number of variables from the cache unnecessarily,
# since many variables are only read in the code.
# On the other hand, there are also *read-only* variables that can be
# changed by code (the "read-only" attribute only refers to the user
# being able to change values directly). For example, synapse creation
# write source and target indices, and monitors write the shared values.
# To correctly mark these values as changed, templates can include a
# "WRITES_TO_READ_ONLY_VARIABLES" comment, stating the name of the
# changed variables. For a monitor, this would for example state that
# the number of recorded values "N" changes. For the recorded variables,
# however, this information cannot be included in the template because
# it is up to the user to define which variables are recorded. For such
# cases, the "owner" object (e.g. a SpikeMonitor) can define a
# "written_readonly_vars" attribute, storing a set of `Variable` objects
# that will be changed by the owner's code objects.
template = getattr(codeobj.templater, template_name)
written_readonly_vars = {
codeobj.variables[varname] for varname in template.writes_read_only
} | getattr(owner, "written_readonly_vars", set())
for var in codeobj.variables.values():
if (
isinstance(var, ArrayVariable)
and var not in do_not_invalidate
and (not var.read_only or var in written_readonly_vars)
):
self.array_cache[var] = None
return codeobj
def check_openmp_compatible(self, nb_threads):
if nb_threads > 0:
logger.warn(
"OpenMP code is not yet well tested, and may be inaccurate.",
"openmp",
once=True,
)
logger.diagnostic(f"Using OpenMP with {int(nb_threads)} threads ")
if prefs.devices.cpp_standalone.openmp_spatialneuron_strategy is not None:
logger.warn(
"The devices.cpp_standalone.openmp_spatialneuron_strategy "
"preference is no longer used and will be removed in "
"future versions of Brian.",
"openmp_spatialneuron_strategy",
once=True,
)
def generate_objects_source(
self,
writer,
arange_arrays,
synapses,
static_array_specs,
networks,
timed_arrays,
):
arr_tmp = self.code_object_class().templater.objects(
None,
None,
array_specs=self.arrays,
dynamic_array_specs=self.dynamic_arrays,
dynamic_array_2d_specs=self.dynamic_arrays_2d,
zero_arrays=self.zero_arrays,
arange_arrays=arange_arrays,
synapses=synapses,
clocks=self.clocks,
static_array_specs=static_array_specs,
networks=networks,
get_array_filename=self.get_array_filename,
get_array_name=self.get_array_name,
profiled_codeobjects=self.profiled_codeobjects,
code_objects=list(self.code_objects.values()),
timed_arrays=timed_arrays,
)
writer.write("objects.*", arr_tmp)
def generate_main_source(self, writer):
main_lines = []
procedures = [("", main_lines)]
runfuncs = {}
for func, args in self.main_queue:
if func == "before_run_code_object":
(codeobj,) = args
main_lines.append(f"_before_run_{codeobj.name}();")
elif func == "run_code_object":
(codeobj,) = args
main_lines.append(f"_run_{codeobj.name}();")
elif func == "after_run_code_object":
(codeobj,) = args
main_lines.append(f"_after_run_{codeobj.name}();")
elif func == "run_network":
net, netcode = args
main_lines.extend(netcode)
elif func == "set_by_constant":
arrayname, value, is_dynamic = args
size_str = f"{arrayname}.size()" if is_dynamic else f"_num_{arrayname}"
code = f"""
{openmp_pragma('static')}
for(int i=0; i<{size_str}; i++)
{{
{arrayname}[i] = {CPPNodeRenderer().render_expr(repr(value))};
}}
"""
main_lines.extend(code.split("\n"))
elif func == "set_by_array":
arrayname, staticarrayname, is_dynamic = args
size_str = f"{arrayname}.size()" if is_dynamic else f"_num_{arrayname}"
code = f"""
{openmp_pragma('static')}
for(int i=0; i<{size_str}; i++)
{{
{arrayname}[i] = {staticarrayname}[i];
}}
"""
main_lines.extend(code.split("\n"))
elif func == "set_by_single_value":
arrayname, item, value = args
code = f"{arrayname}[{item}] = {value};"
main_lines.extend([code])
elif func == "set_array_by_array":
arrayname, staticarrayname_index, staticarrayname_value = args
code = f"""
{openmp_pragma('static')}
for(int i=0; i<_num_{staticarrayname_index}; i++)
{{
{arrayname}[{staticarrayname_index}[i]] = {staticarrayname_value}[i];
}}
"""
main_lines.extend(code.split("\n"))
elif func == "resize_array":
array_name, new_size = args
main_lines.append(f"{array_name}.resize({new_size});")
elif func == "insert_code":
main_lines.append(args)
elif func == "start_run_func":
name, include_in_parent = args
if include_in_parent:
main_lines.append(f"{name}();")
main_lines = []
procedures.append((name, main_lines))
elif func == "end_run_func":
name, include_in_parent = args
name, main_lines = procedures.pop(-1)
runfuncs[name] = main_lines
name, main_lines = procedures[-1]
elif func == "seed":
seed = args
nb_threads = prefs.devices.cpp_standalone.openmp_threads
if nb_threads == 0: # no OpenMP
nb_threads = 1
main_lines.append(f"for (int _i=0; _i<{nb_threads}; _i++)")
if seed is None: # random
main_lines.append(
" rk_randomseed(brian::_mersenne_twister_states[_i]);"
)
else:
main_lines.append(
f" rk_seed({seed!r}L + _i,"
" brian::_mersenne_twister_states[_i]);"
)
else:
raise NotImplementedError(f"Unknown main queue function type {func}")
self.runfuncs = runfuncs
# generate the finalisations
for codeobj in self.code_objects.values():
if hasattr(codeobj.code, "main_finalise"):
main_lines.append(codeobj.code.main_finalise)
user_headers = self.headers + prefs["codegen.cpp.headers"]
main_tmp = self.code_object_class().templater.main(
None,
None,
main_lines=main_lines,
code_lines=self.code_lines,
code_objects=list(self.code_objects.values()),
report_func=self.report_func,
dt=float(self.defaultclock.dt),
user_headers=user_headers,
)
writer.write("main.cpp", main_tmp)
def generate_codeobj_source(self, writer):
# Generate data for non-constant values
renderer = CPPNodeRenderer()
code_object_defs = defaultdict(list)
for codeobj in self.code_objects.values():
lines = []
for k, v in codeobj.variables.items():
if isinstance(v, ArrayVariable):
try:
if isinstance(v, DynamicArrayVariable):
if v.ndim == 1:
dyn_array_name = self.dynamic_arrays[v]
array_name = self.arrays[v]
c_type = c_data_type(v.dtype)
line = (
f"{c_type}* const {array_name} ="
f" {dyn_array_name}.empty()? 0 :"
f" &{dyn_array_name}[0];"
)
lines.append(line)
line = (
f"const size_t _num{k} = {dyn_array_name}.size();"
)
lines.append(line)
else:
lines.append(f"const size_t _num{k} = {v.size};")
except TypeError:
pass
elif isinstance(v, Constant):
value = renderer.render_expr(repr(v.value))
c_type = c_data_type(v.dtype)
line = f"const {c_type} {k} = {value};"
lines.append(line)
for line in lines:
# Sometimes an array is referred to by to different keys in our
# dictionary -- make sure to never add a line twice
if line not in code_object_defs[codeobj.name]:
code_object_defs[codeobj.name].append(line)
# Generate the code objects
for codeobj in self.code_objects.values():
# Before/after run code
for block in codeobj.before_after_blocks:
cpp_code = getattr(codeobj.code, f"{block}_cpp_file")
cpp_code = cpp_code.replace(
"%CONSTANTS%", "\n".join(code_object_defs[codeobj.name])
)
h_code = getattr(codeobj.code, f"{block}_h_file")
writer.write(f"code_objects/{block}_{codeobj.name}.cpp", cpp_code)
writer.write(f"code_objects/{block}_{codeobj.name}.h", h_code)
# Main code
code = codeobj.code.cpp_file
code = code.replace(
"%CONSTANTS%", "\n".join(code_object_defs[codeobj.name])
)
writer.write(f"code_objects/{codeobj.name}.cpp", code)
writer.write(f"code_objects/{codeobj.name}.h", codeobj.code.h_file)
def generate_network_source(self, writer, compiler):
maximum_run_time = self._maximum_run_time
if maximum_run_time is not None:
maximum_run_time = float(maximum_run_time)
network_tmp = self.code_object_class().templater.network(
None, None, maximum_run_time=maximum_run_time
)
writer.write("network.*", network_tmp)
def generate_synapses_classes_source(self, writer):
synapses_classes_tmp = self.code_object_class().templater.synapses_classes(
None, None
)
writer.write("synapses_classes.*", synapses_classes_tmp)
def generate_run_source(self, writer):
run_tmp = self.code_object_class().templater.run(
None,
None,
run_funcs=self.runfuncs,
code_objects=list(self.code_objects.values()),
user_headers=self.headers,
array_specs=self.arrays,
clocks=self.clocks,
)
writer.write("run.*", run_tmp)
def generate_makefile(
self, writer, compiler, compiler_flags, linker_flags, nb_threads, debug
):
if compiler == "msvc":
if nb_threads > 1:
openmp_flag = "/openmp"
else: