This file is indexed.

/usr/lib/python2.7/dist-packages/pyopencl/tools.py is in python-pyopencl 2017.2.2-1.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below.

  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
"""Various helpful bits and pieces without much of a common theme."""

from __future__ import division, absolute_import

__copyright__ = "Copyright (C) 2010 Andreas Kloeckner"

__license__ = """
Permission is hereby granted, free of charge, to any person
obtaining a copy of this software and associated documentation
files (the "Software"), to deal in the Software without
restriction, including without limitation the rights to use,
copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following
conditions:

The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS IN THE SOFTWARE.
"""


import six
from six.moves import zip, intern

import numpy as np
from decorator import decorator
import pyopencl as cl
from pytools import memoize, memoize_method
from pyopencl.cffi_cl import _lib
from pytools.persistent_dict import KeyBuilder as KeyBuilderBase

import re

from pyopencl.compyte.dtypes import (  # noqa
        get_or_register_dtype, TypeNameNotKnown,
        register_dtype, dtype_to_ctype)


def _register_types():
    from pyopencl.compyte.dtypes import (
            TYPE_REGISTRY, fill_registry_with_opencl_c_types)

    fill_registry_with_opencl_c_types(TYPE_REGISTRY)

    get_or_register_dtype("cfloat_t", np.complex64)
    get_or_register_dtype("cdouble_t", np.complex128)


_register_types()


# {{{ imported names

bitlog2 = _lib.bitlog2
from pyopencl.mempool import (  # noqa
        PooledBuffer, DeferredAllocator, ImmediateAllocator, MemoryPool)

# }}}


# {{{ first-arg caches

_first_arg_dependent_caches = []


@decorator
def first_arg_dependent_memoize(func, cl_object, *args):
    """Provides memoization for a function. Typically used to cache
    things that get created inside a :class:`pyopencl.Context`, e.g. programs
    and kernels. Assumes that the first argument of the decorated function is
    an OpenCL object that might go away, such as a :class:`pyopencl.Context` or
    a :class:`pyopencl.CommandQueue`, and based on which we might want to clear
    the cache.

    .. versionadded:: 2011.2
    """
    try:
        ctx_dict = func._pyopencl_first_arg_dep_memoize_dic
    except AttributeError:
        # FIXME: This may keep contexts alive longer than desired.
        # But I guess since the memory in them is freed, who cares.
        ctx_dict = func._pyopencl_first_arg_dep_memoize_dic = {}
        _first_arg_dependent_caches.append(ctx_dict)

    try:
        return ctx_dict[cl_object][args]
    except KeyError:
        arg_dict = ctx_dict.setdefault(cl_object, {})
        result = func(cl_object, *args)
        arg_dict[args] = result
        return result


context_dependent_memoize = first_arg_dependent_memoize


def first_arg_dependent_memoize_nested(nested_func):
    """Provides memoization for nested functions. Typically used to cache
    things that get created inside a :class:`pyopencl.Context`, e.g. programs
    and kernels. Assumes that the first argument of the decorated function is
    an OpenCL object that might go away, such as a :class:`pyopencl.Context` or
    a :class:`pyopencl.CommandQueue`, and will therefore respond to
    :func:`clear_first_arg_caches`.

    .. versionadded:: 2013.1

    Requires Python 2.5 or newer.
    """

    from functools import wraps
    cache_dict_name = intern("_memoize_inner_dic_%s_%s_%d"
            % (nested_func.__name__, nested_func.__code__.co_filename,
                nested_func.__code__.co_firstlineno))

    from inspect import currentframe
    # prevent ref cycle
    try:
        caller_frame = currentframe().f_back
        cache_context = caller_frame.f_globals[
                caller_frame.f_code.co_name]
    finally:
        #del caller_frame
        pass

    try:
        cache_dict = getattr(cache_context, cache_dict_name)
    except AttributeError:
        cache_dict = {}
        _first_arg_dependent_caches.append(cache_dict)
        setattr(cache_context, cache_dict_name, cache_dict)

    @wraps(nested_func)
    def new_nested_func(cl_object, *args):
        try:
            return cache_dict[cl_object][args]
        except KeyError:
            arg_dict = cache_dict.setdefault(cl_object, {})
            result = nested_func(cl_object, *args)
            arg_dict[args] = result
            return result

    return new_nested_func


def clear_first_arg_caches():
    """Empties all first-argument-dependent memoization caches. Also releases
    all held reference contexts. If it is important to you that the
    program detaches from its context, you might need to call this
    function to free all remaining references to your context.

    .. versionadded:: 2011.2
    """
    for cache in _first_arg_dependent_caches:
        cache.clear()


import atexit
atexit.register(clear_first_arg_caches)

# }}}


def get_test_platforms_and_devices(plat_dev_string=None):
    """Parse a string of the form 'PYOPENCL_TEST=0:0,1;intel:i5'.

    :return: list of tuples (platform, [device, device, ...])
    """

    if plat_dev_string is None:
        import os
        plat_dev_string = os.environ.get("PYOPENCL_TEST", None)

    def find_cl_obj(objs, identifier):
        try:
            num = int(identifier)
        except Exception:
            pass
        else:
            return objs[num]

        found = False
        for obj in objs:
            if identifier.lower() in (obj.name + ' ' + obj.vendor).lower():
                return obj
        if not found:
            raise RuntimeError("object '%s' not found" % identifier)

    if plat_dev_string:
        result = []

        for entry in plat_dev_string.split(";"):
            lhsrhs = entry.split(":")

            if len(lhsrhs) == 1:
                platform = find_cl_obj(cl.get_platforms(), lhsrhs[0])
                result.append((platform, platform.get_devices()))

            elif len(lhsrhs) != 2:
                raise RuntimeError("invalid syntax of PYOPENCL_TEST")
            else:
                plat_str, dev_strs = lhsrhs

                platform = find_cl_obj(cl.get_platforms(), plat_str)
                devs = platform.get_devices()
                result.append(
                        (platform,
                            [find_cl_obj(devs, dev_id)
                                for dev_id in dev_strs.split(",")]))

        return result

    else:
        return [
                (platform, platform.get_devices())
                for platform in cl.get_platforms()]


def pytest_generate_tests_for_pyopencl(metafunc):
    class ContextFactory:
        def __init__(self, device):
            self.device = device

        def __call__(self):
            # Get rid of leftovers from past tests.
            # CL implementations are surprisingly limited in how many
            # simultaneous contexts they allow...

            clear_first_arg_caches()

            from gc import collect
            collect()

            return cl.Context([self.device])

        def __str__(self):
            # Don't show address, so that parallel test collection works
            return ("<context factory for <pyopencl.Device '%s' on '%s'>" %
                    (self.device.name.strip(),
                     self.device.platform.name.strip()))

    test_plat_and_dev = get_test_platforms_and_devices()

    if ("device" in metafunc.funcargnames
            or "ctx_factory" in metafunc.funcargnames
            or "ctx_getter" in metafunc.funcargnames):
        arg_dict = {}

        for platform, plat_devs in test_plat_and_dev:
            if "platform" in metafunc.funcargnames:
                arg_dict["platform"] = platform

            for device in plat_devs:
                if "device" in metafunc.funcargnames:
                    arg_dict["device"] = device

                if "ctx_factory" in metafunc.funcargnames:
                    arg_dict["ctx_factory"] = ContextFactory(device)

                if "ctx_getter" in metafunc.funcargnames:
                    from warnings import warn
                    warn("The 'ctx_getter' arg is deprecated in "
                            "favor of 'ctx_factory'.",
                            DeprecationWarning)
                    arg_dict["ctx_getter"] = ContextFactory(device)

                metafunc.addcall(funcargs=arg_dict.copy(),
                        id=", ".join("%s=%s" % (arg, value)
                                for arg, value in six.iteritems(arg_dict)))

    elif "platform" in metafunc.funcargnames:
        for platform, plat_devs in test_plat_and_dev:
            metafunc.addcall(
                    funcargs=dict(platform=platform),
                    id=str(platform))


# {{{ C argument lists

class Argument(object):
    pass


class DtypedArgument(Argument):
    def __init__(self, dtype, name):
        self.dtype = np.dtype(dtype)
        self.name = name

    def __repr__(self):
        return "%s(%r, %s)" % (
                self.__class__.__name__,
                self.name,
                self.dtype)


class VectorArg(DtypedArgument):
    def __init__(self, dtype, name, with_offset=False):
        DtypedArgument.__init__(self, dtype, name)
        self.with_offset = with_offset

    def declarator(self):
        if self.with_offset:
            # Two underscores -> less likelihood of a name clash.
            return "__global %s *%s__base, long %s__offset" % (
                    dtype_to_ctype(self.dtype), self.name, self.name)
        else:
            result = "__global %s *%s" % (dtype_to_ctype(self.dtype), self.name)

        return result


class ScalarArg(DtypedArgument):
    def declarator(self):
        return "%s %s" % (dtype_to_ctype(self.dtype), self.name)


class OtherArg(Argument):
    def __init__(self, declarator, name):
        self.decl = declarator
        self.name = name

    def declarator(self):
        return self.decl


def parse_c_arg(c_arg, with_offset=False):
    for aspace in ["__local", "__constant"]:
        if aspace in c_arg:
            raise RuntimeError("cannot deal with local or constant "
                    "OpenCL address spaces in C argument lists ")

    c_arg = c_arg.replace("__global", "")

    if with_offset:
        def vec_arg_factory(dtype, name):
            return VectorArg(dtype, name, with_offset=True)
    else:
        vec_arg_factory = VectorArg

    from pyopencl.compyte.dtypes import parse_c_arg_backend
    return parse_c_arg_backend(c_arg, ScalarArg, vec_arg_factory)


def parse_arg_list(arguments, with_offset=False):
    """Parse a list of kernel arguments. *arguments* may be a comma-separate
    list of C declarators in a string, a list of strings representing C
    declarators, or :class:`Argument` objects.
    """

    if isinstance(arguments, str):
        arguments = arguments.split(",")

    def parse_single_arg(obj):
        if isinstance(obj, str):
            from pyopencl.tools import parse_c_arg
            return parse_c_arg(obj, with_offset=with_offset)
        else:
            return obj

    return [parse_single_arg(arg) for arg in arguments]


def get_arg_list_scalar_arg_dtypes(arg_types):
    result = []

    for arg_type in arg_types:
        if isinstance(arg_type, ScalarArg):
            result.append(arg_type.dtype)
        elif isinstance(arg_type, VectorArg):
            result.append(None)
            if arg_type.with_offset:
                result.append(np.int64)
        else:
            raise RuntimeError("arg type not understood: %s" % type(arg_type))

    return result


def get_arg_offset_adjuster_code(arg_types):
    result = []

    for arg_type in arg_types:
        if isinstance(arg_type, VectorArg) and arg_type.with_offset:
            result.append("__global %(type)s *%(name)s = "
                    "(__global %(type)s *) "
                    "((__global char *) %(name)s__base + %(name)s__offset);"
                    % dict(
                        type=dtype_to_ctype(arg_type.dtype),
                        name=arg_type.name))

    return "\n".join(result)


# }}}


def get_gl_sharing_context_properties():
    ctx_props = cl.context_properties

    from OpenGL import platform as gl_platform

    props = []

    import sys
    if sys.platform in ["linux", "linux2"]:
        from OpenGL import GLX
        props.append(
            (ctx_props.GL_CONTEXT_KHR, gl_platform.GetCurrentContext()))
        props.append(
                (ctx_props.GLX_DISPLAY_KHR,
                    GLX.glXGetCurrentDisplay()))
    elif sys.platform == "win32":
        from OpenGL import WGL
        props.append(
            (ctx_props.GL_CONTEXT_KHR, gl_platform.GetCurrentContext()))
        props.append(
                (ctx_props.WGL_HDC_KHR,
                    WGL.wglGetCurrentDC()))
    elif sys.platform == "darwin":
        props.append(
            (ctx_props.CONTEXT_PROPERTY_USE_CGL_SHAREGROUP_APPLE,
                cl.get_apple_cgl_share_group()))
    else:
        raise NotImplementedError("platform '%s' not yet supported"
                % sys.platform)

    return props


class _CDeclList:
    def __init__(self, device):
        self.device = device
        self.declared_dtypes = set()
        self.declarations = []
        self.saw_double = False
        self.saw_complex = False

    def add_dtype(self, dtype):
        dtype = np.dtype(dtype)

        if dtype in [np.float64 or np.complex128]:
            self.saw_double = True

        if dtype.kind == "c":
            self.saw_complex = True

        if dtype.kind != "V":
            return

        if dtype in self.declared_dtypes:
            return

        import pyopencl.cltypes
        if dtype in pyopencl.cltypes.vec_type_to_scalar_and_count:
            return

        for name, field_data in sorted(six.iteritems(dtype.fields)):
            field_dtype, offset = field_data[:2]
            self.add_dtype(field_dtype)

        _, cdecl = match_dtype_to_c_struct(
                self.device, dtype_to_ctype(dtype), dtype)

        self.declarations.append(cdecl)
        self.declared_dtypes.add(dtype)

    def visit_arguments(self, arguments):
        for arg in arguments:
            dtype = arg.dtype
            if dtype in [np.float64 or np.complex128]:
                self.saw_double = True

            if dtype.kind == "c":
                self.saw_complex = True

    def get_declarations(self):
        result = "\n\n".join(self.declarations)

        if self.saw_complex:
            result = (
                    "#include <pyopencl-complex.h>\n\n"
                    + result)

        if self.saw_double:
            result = (
                    """
                    #if __OPENCL_C_VERSION__ < 120
                    #pragma OPENCL EXTENSION cl_khr_fp64: enable
                    #endif
                    #define PYOPENCL_DEFINE_CDOUBLE
                    """
                    + result)

        return result


@memoize
def match_dtype_to_c_struct(device, name, dtype, context=None):
    """Return a tuple `(dtype, c_decl)` such that the C struct declaration
    in `c_decl` and the structure :class:`numpy.dtype` instance `dtype`
    have the same memory layout.

    Note that *dtype* may be modified from the value that was passed in,
    for example to insert padding.

    (As a remark on implementation, this routine runs a small kernel on
    the given *device* to ensure that :mod:`numpy` and C offsets and
    sizes match.)

    .. versionadded: 2013.1

    This example explains the use of this function::

        >>> import numpy as np
        >>> import pyopencl as cl
        >>> import pyopencl.tools
        >>> ctx = cl.create_some_context()
        >>> dtype = np.dtype([("id", np.uint32), ("value", np.float32)])
        >>> dtype, c_decl = pyopencl.tools.match_dtype_to_c_struct(
        ...     ctx.devices[0], 'id_val', dtype)
        >>> print c_decl
        typedef struct {
          unsigned id;
          float value;
        } id_val;
        >>> print dtype
        [('id', '<u4'), ('value', '<f4')]
        >>> cl.tools.get_or_register_dtype('id_val', dtype)

    As this example shows, it is important to call
    :func:`get_or_register_dtype` on the modified `dtype` returned by this
    function, not the original one.
    """

    fields = sorted(six.iteritems(dtype.fields),
            key=lambda name_dtype_offset: name_dtype_offset[1][1])

    c_fields = []
    for field_name, dtype_and_offset in fields:
        field_dtype, offset = dtype_and_offset[:2]
        c_fields.append("  %s %s;" % (dtype_to_ctype(field_dtype), field_name))

    c_decl = "typedef struct {\n%s\n} %s;\n\n" % (
            "\n".join(c_fields),
            name)

    cdl = _CDeclList(device)
    for field_name, dtype_and_offset in fields:
        field_dtype, offset = dtype_and_offset[:2]
        cdl.add_dtype(field_dtype)

    pre_decls = cdl.get_declarations()

    offset_code = "\n".join(
            "result[%d] = pycl_offsetof(%s, %s);" % (i+1, name, field_name)
            for i, (field_name, _) in enumerate(fields))

    src = r"""
        #define pycl_offsetof(st, m) \
                 ((uint) ((__local char *) &(dummy.m) \
                 - (__local char *)&dummy ))

        %(pre_decls)s

        %(my_decl)s

        __kernel void get_size_and_offsets(__global uint *result)
        {
            result[0] = sizeof(%(my_type)s);
            __local %(my_type)s dummy;
            %(offset_code)s
        }
    """ % dict(
            pre_decls=pre_decls,
            my_decl=c_decl,
            my_type=name,
            offset_code=offset_code)

    if context is None:
        context = cl.Context([device])

    queue = cl.CommandQueue(context)

    prg = cl.Program(context, src)
    knl = prg.build(devices=[device]).get_size_and_offsets

    import pyopencl.array  # noqa
    result_buf = cl.array.empty(queue, 1+len(fields), np.uint32)
    knl(queue, (1,), (1,), result_buf.data)
    queue.finish()
    size_and_offsets = result_buf.get()

    size = int(size_and_offsets[0])

    from pytools import any
    offsets = size_and_offsets[1:]
    if any(ofs >= size for ofs in offsets):
        # offsets not plausible

        if dtype.itemsize == size:
            # If sizes match, use numpy's idea of the offsets.
            offsets = [dtype_and_offset[1]
                    for field_name, dtype_and_offset in fields]
        else:
            raise RuntimeError(
                    "OpenCL compiler reported offsetof() past sizeof() "
                    "for struct layout on '%s'. "
                    "This makes no sense, and it's usually indicates a "
                    "compiler bug. "
                    "Refusing to discover struct layout." % device)

    result_buf.data.release()
    del knl
    del prg
    del queue
    del context

    try:
        dtype_arg_dict = {
            'names': [field_name
                      for field_name, (field_dtype, offset) in fields],
            'formats': [field_dtype
                        for field_name, (field_dtype, offset) in fields],
            'offsets': [int(x) for x in offsets],
            'itemsize': int(size_and_offsets[0]),
            }
        dtype = np.dtype(dtype_arg_dict)
        if dtype.itemsize != size_and_offsets[0]:
            # "Old" versions of numpy (1.6.x?) silently ignore "itemsize". Boo.
            dtype_arg_dict["names"].append("_pycl_size_fixer")
            dtype_arg_dict["formats"].append(np.uint8)
            dtype_arg_dict["offsets"].append(int(size_and_offsets[0])-1)
            dtype = np.dtype(dtype_arg_dict)
    except NotImplementedError:
        def calc_field_type():
            total_size = 0
            padding_count = 0
            for offset, (field_name, (field_dtype, _)) in zip(offsets, fields):
                if offset > total_size:
                    padding_count += 1
                    yield ('__pycl_padding%d' % padding_count,
                           'V%d' % offset - total_size)
                yield field_name, field_dtype
                total_size = field_dtype.itemsize + offset
        dtype = np.dtype(list(calc_field_type()))

    assert dtype.itemsize == size_and_offsets[0]

    return dtype, c_decl


@memoize
def dtype_to_c_struct(device, dtype):
    if dtype.fields is None:
        return ""

    import pyopencl.cltypes
    if dtype in pyopencl.cltypes.vec_type_to_scalar_and_count:
        # Vector types are built-in. Don't try to redeclare those.
        return ""

    matched_dtype, c_decl = match_dtype_to_c_struct(
            device, dtype_to_ctype(dtype), dtype)

    def dtypes_match():
        result = len(dtype.fields) == len(matched_dtype.fields)

        for name, val in six.iteritems(dtype.fields):
            result = result and matched_dtype.fields[name] == val

        return result

    assert dtypes_match()

    return c_decl


# {{{ code generation/templating helper

def _process_code_for_macro(code):
    code = code.replace("//CL//", "\n")

    if "//" in code:
        raise RuntimeError("end-of-line comments ('//') may not be used in "
                "code snippets")

    return code.replace("\n", " \\\n")


class _SimpleTextTemplate:
    def __init__(self, txt):
        self.txt = txt

    def render(self, context):
        return self.txt


class _PrintfTextTemplate:
    def __init__(self, txt):
        self.txt = txt

    def render(self, context):
        return self.txt % context


class _MakoTextTemplate:
    def __init__(self, txt):
        from mako.template import Template
        self.template = Template(txt, strict_undefined=True)

    def render(self, context):
        return self.template.render(**context)


class _ArgumentPlaceholder:
    """A placeholder for subclasses of :class:`DtypedArgument`. This is needed
    because the concrete dtype of the argument is not known at template
    creation time--it may be a type alias that will only be filled in
    at run time. These types take the place of these proto-arguments until
    all types are known.

    See also :class:`_TemplateRenderer.render_arg`.
    """

    def __init__(self, typename, name, **extra_kwargs):
        self.typename = typename
        self.name = name
        self.extra_kwargs = extra_kwargs


class _VectorArgPlaceholder(_ArgumentPlaceholder):
    target_class = VectorArg


class _ScalarArgPlaceholder(_ArgumentPlaceholder):
    target_class = ScalarArg


class _TemplateRenderer(object):
    def __init__(self, template, type_aliases, var_values, context=None,
            options=[]):
        self.template = template
        self.type_aliases = dict(type_aliases)
        self.var_dict = dict(var_values)

        for name in self.var_dict:
            if name.startswith("macro_"):
                self.var_dict[name] = _process_code_for_macro(
                        self.var_dict[name])

        self.context = context
        self.options = options

    def __call__(self, txt):
        if txt is None:
            return txt

        result = self.template.get_text_template(txt).render(self.var_dict)

        return str(result)

    def get_rendered_kernel(self, txt, kernel_name):
        prg = cl.Program(self.context, self(txt)).build(self.options)

        kernel_name_prefix = self.var_dict.get("kernel_name_prefix")
        if kernel_name_prefix is not None:
            kernel_name = kernel_name_prefix+kernel_name

        return getattr(prg, kernel_name)

    def parse_type(self, typename):
        if isinstance(typename, str):
            try:
                return self.type_aliases[typename]
            except KeyError:
                from pyopencl.compyte.dtypes import NAME_TO_DTYPE
                return NAME_TO_DTYPE[typename]
        else:
            return np.dtype(typename)

    def render_arg(self, arg_placeholder):
        return arg_placeholder.target_class(
                self.parse_type(arg_placeholder.typename),
                arg_placeholder.name,
                **arg_placeholder.extra_kwargs)

    _C_COMMENT_FINDER = re.compile(r"/\*.*?\*/")

    def render_argument_list(self, *arg_lists, **kwargs):
        with_offset = kwargs.pop("with_offset", False)
        if kwargs:
            raise TypeError("unrecognized kwargs: " + ", ".join(kwargs))

        all_args = []

        for arg_list in arg_lists:
            if isinstance(arg_list, str):
                arg_list = str(
                        self.template
                        .get_text_template(arg_list).render(self.var_dict))
                arg_list = self._C_COMMENT_FINDER.sub("", arg_list)
                arg_list = arg_list.replace("\n", " ")

                all_args.extend(arg_list.split(","))
            else:
                all_args.extend(arg_list)

        if with_offset:
            def vec_arg_factory(typename, name):
                    return _VectorArgPlaceholder(typename, name, with_offset=True)
        else:
            vec_arg_factory = _VectorArgPlaceholder

        from pyopencl.compyte.dtypes import parse_c_arg_backend
        parsed_args = []
        for arg in all_args:
            if isinstance(arg, str):
                arg = arg.strip()
                if not arg:
                    continue

                ph = parse_c_arg_backend(arg,
                        _ScalarArgPlaceholder, vec_arg_factory,
                        name_to_dtype=lambda x: x)
                parsed_arg = self.render_arg(ph)

            elif isinstance(arg, Argument):
                parsed_arg = arg
            elif isinstance(arg, tuple):
                parsed_arg = ScalarArg(self.parse_type(arg[0]), arg[1])

            parsed_args.append(parsed_arg)

        return parsed_args

    def get_type_decl_preamble(self, device, decl_type_names, arguments=None):
        cdl = _CDeclList(device)

        for typename in decl_type_names:
            cdl.add_dtype(self.parse_type(typename))

        if arguments is not None:
            cdl.visit_arguments(arguments)

        for _, tv in sorted(six.iteritems(self.type_aliases)):
            cdl.add_dtype(tv)

        type_alias_decls = [
                "typedef %s %s;" % (dtype_to_ctype(val), name)
                for name, val in sorted(six.iteritems(self.type_aliases))
                ]

        return cdl.get_declarations() + "\n" + "\n".join(type_alias_decls)


class KernelTemplateBase(object):
    def __init__(self, template_processor=None):
        self.template_processor = template_processor

        self.build_cache = {}
        _first_arg_dependent_caches.append(self.build_cache)

    def get_preamble(self):
        pass

    _TEMPLATE_PROCESSOR_PATTERN = re.compile(r"^//CL(?::([a-zA-Z0-9_]+))?//")

    @memoize_method
    def get_text_template(self, txt):
        proc_match = self._TEMPLATE_PROCESSOR_PATTERN.match(txt)
        tpl_processor = None

        if proc_match is not None:
            tpl_processor = proc_match.group(1)
            # chop off //CL// mark
            txt = txt[len(proc_match.group(0)):]
        if tpl_processor is None:
            tpl_processor = self.template_processor

        if tpl_processor is None or tpl_processor == "none":
            return _SimpleTextTemplate(txt)
        elif tpl_processor == "printf":
            return _PrintfTextTemplate(txt)
        elif tpl_processor == "mako":
            return _MakoTextTemplate(txt)
        else:
            raise RuntimeError(
                    "unknown template processor '%s'" % proc_match.group(1))

    def get_renderer(self, type_aliases, var_values, context=None, options=[]):
        return _TemplateRenderer(self, type_aliases, var_values)

    def build(self, context, *args, **kwargs):
        """Provide caching for an :meth:`build_inner`."""

        cache_key = (context, args, tuple(sorted(six.iteritems(kwargs))))
        try:
            return self.build_cache[cache_key]
        except KeyError:
            result = self.build_inner(context, *args, **kwargs)
            self.build_cache[cache_key] = result
            return result

# }}}


# {{{ array_module

class _CLFakeArrayModule:
    def __init__(self, queue):
        self.queue = queue

    @property
    def ndarray(self):
        from pyopencl.array import Array
        return Array

    def dot(self, x, y):
        from pyopencl.array import dot
        return dot(x, y, queue=self.queue).get()

    def vdot(self, x, y):
        from pyopencl.array import vdot
        return vdot(x, y, queue=self.queue).get()

    def empty(self, shape, dtype, order="C"):
        from pyopencl.array import empty
        return empty(self.queue, shape, dtype, order=order)

    def hstack(self, arrays):
        from pyopencl.array import hstack
        return hstack(arrays, self.queue)


def array_module(a):
    if isinstance(a, np.ndarray):
        return np
    else:
        from pyopencl.array import Array
        if isinstance(a, Array):
            return _CLFakeArrayModule(a.queue)
        else:
            raise TypeError("array type not understood: %s" % type(a))

# }}}


def is_spirv(s):
    spirv_magic = b"\x07\x23\x02\x03"
    return (
            isinstance(s, six.binary_type)
            and (
                s[:4] == spirv_magic
                or s[:4] == spirv_magic[::-1]))


# {{{ numpy key types builder

class _NumpyTypesKeyBuilder(KeyBuilderBase):
    def update_for_type(self, key_hash, key):
        if issubclass(key, np.generic):
            self.update_for_str(key_hash, key.__name__)
            return

        raise TypeError("unsupported type for persistent hash keying: %s"
                % type(key))

# }}}

# vim: foldmethod=marker