/usr/lib/python2.7/dist-packages/pyopencl/algorithm.py is in python-pyopencl 2016.1+git20161130-1.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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 | """Scan primitive."""
from __future__ import division
from __future__ import absolute_import
from six.moves import range
from six.moves import zip
__copyright__ = """Copyright 2011-2012 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 numpy as np
import pyopencl as cl
import pyopencl.array # noqa
from pyopencl.scan import ScanTemplate
from pyopencl.tools import dtype_to_ctype
from pytools import memoize, memoize_method, Record
from mako.template import Template
# {{{ copy_if
_copy_if_template = ScanTemplate(
arguments="item_t *ary, item_t *out, scan_t *count",
input_expr="(%(predicate)s) ? 1 : 0",
scan_expr="a+b", neutral="0",
output_statement="""
if (prev_item != item) out[item-1] = ary[i];
if (i+1 == N) *count = item;
""",
template_processor="printf")
def extract_extra_args_types_values(extra_args):
from pyopencl.tools import VectorArg, ScalarArg
extra_args_types = []
extra_args_values = []
for name, val in extra_args:
if isinstance(val, cl.array.Array):
extra_args_types.append(VectorArg(val.dtype, name, with_offset=False))
extra_args_values.append(val)
elif isinstance(val, np.generic):
extra_args_types.append(ScalarArg(val.dtype, name))
extra_args_values.append(val)
else:
raise RuntimeError("argument '%d' not understood" % name)
return tuple(extra_args_types), extra_args_values
def copy_if(ary, predicate, extra_args=[], preamble="", queue=None, wait_for=None):
"""Copy the elements of *ary* satisfying *predicate* to an output array.
:arg predicate: a C expression evaluating to a `bool`, represented as a string.
The value to test is available as `ary[i]`, and if the expression evaluates
to `true`, then this value ends up in the output.
:arg extra_args: |scan_extra_args|
:arg preamble: |preamble|
:arg wait_for: |explain-waitfor|
:returns: a tuple *(out, count, event)* where *out* is the output array, *count*
is an on-device scalar (fetch to host with `count.get()`) indicating
how many elements satisfied *predicate*, and *event* is a
:class:`pyopencl.Event` for dependency management. *out* is allocated
to the same length as *ary*, but only the first *count* entries carry
meaning.
.. versionadded:: 2013.1
"""
if len(ary) > np.iinfo(np.int32).max:
scan_dtype = np.int64
else:
scan_dtype = np.int32
extra_args_types, extra_args_values = extract_extra_args_types_values(extra_args)
knl = _copy_if_template.build(ary.context,
type_aliases=(("scan_t", scan_dtype), ("item_t", ary.dtype)),
var_values=(("predicate", predicate),),
more_preamble=preamble, more_arguments=extra_args_types)
out = cl.array.empty_like(ary)
count = ary._new_with_changes(data=None, offset=0,
shape=(), strides=(), dtype=scan_dtype)
# **dict is a Py2.5 workaround
evt = knl(ary, out, count, *extra_args_values,
**dict(queue=queue, wait_for=wait_for))
return out, count, evt
# }}}
# {{{ remove_if
def remove_if(ary, predicate, extra_args=[], preamble="", queue=None, wait_for=None):
"""Copy the elements of *ary* not satisfying *predicate* to an output array.
:arg predicate: a C expression evaluating to a `bool`, represented as a string.
The value to test is available as `ary[i]`, and if the expression evaluates
to `false`, then this value ends up in the output.
:arg extra_args: |scan_extra_args|
:arg preamble: |preamble|
:arg wait_for: |explain-waitfor|
:returns: a tuple *(out, count, event)* where *out* is the output array, *count*
is an on-device scalar (fetch to host with `count.get()`) indicating
how many elements did not satisfy *predicate*, and *event* is a
:class:`pyopencl.Event` for dependency management.
.. versionadded:: 2013.1
"""
return copy_if(ary, "!(%s)" % predicate, extra_args=extra_args,
preamble=preamble, queue=queue, wait_for=wait_for)
# }}}
# {{{ partition
_partition_template = ScanTemplate(
arguments=(
"item_t *ary, item_t *out_true, item_t *out_false, "
"scan_t *count_true"),
input_expr="(%(predicate)s) ? 1 : 0",
scan_expr="a+b", neutral="0",
output_statement="""//CL//
if (prev_item != item)
out_true[item-1] = ary[i];
else
out_false[i-item] = ary[i];
if (i+1 == N) *count_true = item;
""",
template_processor="printf")
def partition(ary, predicate, extra_args=[], preamble="", queue=None, wait_for=None):
"""Copy the elements of *ary* into one of two arrays depending on whether
they satisfy *predicate*.
:arg predicate: a C expression evaluating to a `bool`, represented as a string.
The value to test is available as `ary[i]`.
:arg extra_args: |scan_extra_args|
:arg preamble: |preamble|
:arg wait_for: |explain-waitfor|
:returns: a tuple *(out_true, out_false, count, event)* where *count*
is an on-device scalar (fetch to host with `count.get()`) indicating
how many elements satisfied the predicate, and *event* is a
:class:`pyopencl.Event` for dependency management.
.. versionadded:: 2013.1
"""
if len(ary) > np.iinfo(np.uint32).max:
scan_dtype = np.uint64
else:
scan_dtype = np.uint32
extra_args_types, extra_args_values = extract_extra_args_types_values(extra_args)
knl = _partition_template.build(
ary.context,
type_aliases=(("item_t", ary.dtype), ("scan_t", scan_dtype)),
var_values=(("predicate", predicate),),
more_preamble=preamble, more_arguments=extra_args_types)
out_true = cl.array.empty_like(ary)
out_false = cl.array.empty_like(ary)
count = ary._new_with_changes(data=None, offset=0,
shape=(), strides=(), dtype=scan_dtype)
# **dict is a Py2.5 workaround
evt = knl(ary, out_true, out_false, count, *extra_args_values,
**dict(queue=queue, wait_for=wait_for))
return out_true, out_false, count, evt
# }}}
# {{{ unique
_unique_template = ScanTemplate(
arguments="item_t *ary, item_t *out, scan_t *count_unique",
input_fetch_exprs=[
("ary_im1", "ary", -1),
("ary_i", "ary", 0),
],
input_expr="(i == 0) || (IS_EQUAL_EXPR(ary_im1, ary_i) ? 0 : 1)",
scan_expr="a+b", neutral="0",
output_statement="""
if (prev_item != item) out[item-1] = ary[i];
if (i+1 == N) *count_unique = item;
""",
preamble="#define IS_EQUAL_EXPR(a, b) %(macro_is_equal_expr)s\n",
template_processor="printf")
def unique(ary, is_equal_expr="a == b", extra_args=[], preamble="",
queue=None, wait_for=None):
"""Copy the elements of *ary* into the output if *is_equal_expr*, applied to the
array element and its predecessor, yields false.
Works like the UNIX command :program:`uniq`, with a potentially custom
comparison. This operation is often used on sorted sequences.
:arg is_equal_expr: a C expression evaluating to a `bool`,
represented as a string. The elements being compared are
available as `a` and `b`. If this expression yields `false`, the
two are considered distinct.
:arg extra_args: |scan_extra_args|
:arg preamble: |preamble|
:arg wait_for: |explain-waitfor|
:returns: a tuple *(out, count, event)* where *out* is the output array, *count*
is an on-device scalar (fetch to host with `count.get()`) indicating
how many elements satisfied the predicate, and *event* is a
:class:`pyopencl.Event` for dependency management.
.. versionadded:: 2013.1
"""
if len(ary) > np.iinfo(np.uint32).max:
scan_dtype = np.uint64
else:
scan_dtype = np.uint32
extra_args_types, extra_args_values = extract_extra_args_types_values(extra_args)
knl = _unique_template.build(
ary.context,
type_aliases=(("item_t", ary.dtype), ("scan_t", scan_dtype)),
var_values=(("macro_is_equal_expr", is_equal_expr),),
more_preamble=preamble, more_arguments=extra_args_types)
out = cl.array.empty_like(ary)
count = ary._new_with_changes(data=None, offset=0,
shape=(), strides=(), dtype=scan_dtype)
# **dict is a Py2.5 workaround
evt = knl(ary, out, count, *extra_args_values,
**dict(queue=queue, wait_for=wait_for))
return out, count, evt
# }}}
# {{{ radix_sort
def to_bin(n):
# Py 2.5 has no built-in bin()
digs = []
while n:
digs.append(str(n % 2))
n >>= 1
return ''.join(digs[::-1])
def _padded_bin(i, l):
s = to_bin(i)
while len(s) < l:
s = '0' + s
return s
@memoize
def _make_sort_scan_type(device, bits, index_dtype):
name = "pyopencl_sort_scan_%s_%dbits_t" % (
index_dtype.type.__name__, bits)
fields = []
for mnr in range(2**bits):
fields.append(('c%s' % _padded_bin(mnr, bits), index_dtype))
dtype = np.dtype(fields)
from pyopencl.tools import get_or_register_dtype, match_dtype_to_c_struct
dtype, c_decl = match_dtype_to_c_struct(device, name, dtype)
dtype = get_or_register_dtype(name, dtype)
return name, dtype, c_decl
# {{{ types, helpers preamble
RADIX_SORT_PREAMBLE_TPL = Template(r"""//CL//
typedef ${scan_ctype} scan_t;
typedef ${key_ctype} key_t;
typedef ${index_ctype} index_t;
// #define DEBUG
#ifdef DEBUG
#define dbg_printf(ARGS) printf ARGS
#else
#define dbg_printf(ARGS) /* */
#endif
index_t get_count(scan_t s, int mnr)
{
return ${get_count_branch("")};
}
#define BIN_NR(key_arg) ((key_arg >> base_bit) & ${2**bits - 1})
""", strict_undefined=True)
# }}}
# {{{ scan helpers
RADIX_SORT_SCAN_PREAMBLE_TPL = Template(r"""//CL//
scan_t scan_t_neutral()
{
scan_t result;
%for mnr in range(2**bits):
result.c${padded_bin(mnr, bits)} = 0;
%endfor
return result;
}
// considers bits (base_bit+bits-1, ..., base_bit)
scan_t scan_t_from_value(
key_t key,
int base_bit,
int i
)
{
// extract relevant bit range
key_t bin_nr = BIN_NR(key);
dbg_printf(("i: %d key:%d bin_nr:%d\n", i, key, bin_nr));
scan_t result;
%for mnr in range(2**bits):
result.c${padded_bin(mnr, bits)} = (bin_nr == ${mnr});
%endfor
return result;
}
scan_t scan_t_add(scan_t a, scan_t b, bool across_seg_boundary)
{
%for mnr in range(2**bits):
<% field = "c"+padded_bin(mnr, bits) %>
b.${field} = a.${field} + b.${field};
%endfor
return b;
}
""", strict_undefined=True)
RADIX_SORT_OUTPUT_STMT_TPL = Template(r"""//CL//
{
key_t key = ${key_expr};
key_t my_bin_nr = BIN_NR(key);
index_t previous_bins_size = 0;
%for mnr in range(2**bits):
previous_bins_size +=
(my_bin_nr > ${mnr})
? last_item.c${padded_bin(mnr, bits)}
: 0;
%endfor
index_t tgt_idx =
previous_bins_size
+ get_count(item, my_bin_nr) - 1;
%for arg_name in sort_arg_names:
sorted_${arg_name}[tgt_idx] = ${arg_name}[i];
%endfor
}
""", strict_undefined=True)
# }}}
# {{{ driver
# import hoisted here to be used as a default argument in the constructor
from pyopencl.scan import GenericScanKernel
class RadixSort(object):
"""Provides a general `radix sort <https://en.wikipedia.org/wiki/Radix_sort>`_
on the compute device.
.. seealso:: :class:`pyopencl.algorithm.BitonicSort`
.. versionadded:: 2013.1
"""
def __init__(self, context, arguments, key_expr, sort_arg_names,
bits_at_a_time=2, index_dtype=np.int32, key_dtype=np.uint32,
scan_kernel=GenericScanKernel, options=[]):
"""
:arg arguments: A string of comma-separated C argument declarations.
If *arguments* is specified, then *input_expr* must also be
specified. All types used here must be known to PyOpenCL.
(see :func:`pyopencl.tools.get_or_register_dtype`).
:arg key_expr: An integer-valued C expression returning the
key based on which the sort is performed. The array index
for which the key is to be computed is available as `i`.
The expression may refer to any of the *arguments*.
:arg sort_arg_names: A list of argument names whose corresponding
array arguments will be sorted according to *key_expr*.
"""
# {{{ arg processing
from pyopencl.tools import parse_arg_list
self.arguments = parse_arg_list(arguments)
del arguments
self.sort_arg_names = sort_arg_names
self.bits = int(bits_at_a_time)
self.index_dtype = np.dtype(index_dtype)
self.key_dtype = np.dtype(key_dtype)
self.options = options
# }}}
# {{{ kernel creation
scan_ctype, scan_dtype, scan_t_cdecl = \
_make_sort_scan_type(context.devices[0], self.bits, self.index_dtype)
from pyopencl.tools import VectorArg, ScalarArg
scan_arguments = (
list(self.arguments)
+ [VectorArg(arg.dtype, "sorted_"+arg.name) for arg in self.arguments
if arg.name in sort_arg_names]
+ [ScalarArg(np.int32, "base_bit")])
def get_count_branch(known_bits):
if len(known_bits) == self.bits:
return "s.c%s" % known_bits
boundary_mnr = known_bits + "1" + (self.bits-len(known_bits)-1)*"0"
return ("((mnr < %s) ? %s : %s)" % (
int(boundary_mnr, 2),
get_count_branch(known_bits+"0"),
get_count_branch(known_bits+"1")))
codegen_args = dict(
bits=self.bits,
key_ctype=dtype_to_ctype(self.key_dtype),
key_expr=key_expr,
index_ctype=dtype_to_ctype(self.index_dtype),
index_type_max=np.iinfo(self.index_dtype).max,
padded_bin=_padded_bin,
scan_ctype=scan_ctype,
sort_arg_names=sort_arg_names,
get_count_branch=get_count_branch,
)
preamble = scan_t_cdecl+RADIX_SORT_PREAMBLE_TPL.render(**codegen_args)
scan_preamble = preamble \
+ RADIX_SORT_SCAN_PREAMBLE_TPL.render(**codegen_args)
self.scan_kernel = scan_kernel(
context, scan_dtype,
arguments=scan_arguments,
input_expr="scan_t_from_value(%s, base_bit, i)" % key_expr,
scan_expr="scan_t_add(a, b, across_seg_boundary)",
neutral="scan_t_neutral()",
output_statement=RADIX_SORT_OUTPUT_STMT_TPL.render(**codegen_args),
preamble=scan_preamble, options=self.options)
for i, arg in enumerate(self.arguments):
if isinstance(arg, VectorArg):
self.first_array_arg_idx = i
# }}}
def __call__(self, *args, **kwargs):
"""Run the radix sort. In addition to *args* which must match the
*arguments* specification on the constructor, the following
keyword arguments are supported:
:arg key_bits: specify how many bits (starting from least-significant)
there are in the key.
:arg allocator: See the *allocator* argument of :func:`pyopencl.array.empty`.
:arg queue: A :class:`pyopencl.CommandQueue`, defaulting to the
one from the first argument array.
:arg wait_for: |explain-waitfor|
:returns: A tuple ``(sorted, event)``. *sorted* consists of sorted
copies of the arrays named in *sorted_args*, in the order of that
list. *event* is a :class:`pyopencl.Event` for dependency management.
"""
wait_for = kwargs.pop("wait_for", None)
# {{{ run control
key_bits = kwargs.pop("key_bits", None)
if key_bits is None:
key_bits = int(np.iinfo(self.key_dtype).bits)
n = len(args[self.first_array_arg_idx])
allocator = kwargs.pop("allocator", None)
if allocator is None:
allocator = args[self.first_array_arg_idx].allocator
queue = kwargs.pop("allocator", None)
if queue is None:
queue = args[self.first_array_arg_idx].queue
args = list(args)
base_bit = 0
while base_bit < key_bits:
sorted_args = [
cl.array.empty(queue, n, arg_descr.dtype, allocator=allocator)
for arg_descr in self.arguments
if arg_descr.name in self.sort_arg_names]
scan_args = args + sorted_args + [base_bit]
last_evt = self.scan_kernel(*scan_args,
**dict(queue=queue, wait_for=wait_for))
wait_for = [last_evt]
# substitute sorted
for i, arg_descr in enumerate(self.arguments):
if arg_descr.name in self.sort_arg_names:
args[i] = sorted_args[self.sort_arg_names.index(arg_descr.name)]
base_bit += self.bits
return [arg_val
for arg_descr, arg_val in zip(self.arguments, args)
if arg_descr.name in self.sort_arg_names], last_evt
# }}}
# }}}
# }}}
# {{{ generic parallel list builder
# {{{ kernel template
_LIST_BUILDER_TEMPLATE = Template("""//CL//
% if double_support:
#if __OPENCL_C_VERSION__ < 120
#pragma OPENCL EXTENSION cl_khr_fp64: enable
#endif
#define PYOPENCL_DEFINE_CDOUBLE
% endif
#include <pyopencl-complex.h>
${preamble}
// {{{ declare helper macros for user interface
typedef ${index_type} index_type;
%if is_count_stage:
#define PLB_COUNT_STAGE
%for name, dtype in list_names_and_dtypes:
%if name in count_sharing:
#define APPEND_${name}(value) { /* nothing */ }
%else:
#define APPEND_${name}(value) { ++(*plb_loc_${name}_count); }
%endif
%endfor
%else:
#define PLB_WRITE_STAGE
%for name, dtype in list_names_and_dtypes:
%if name in count_sharing:
#define APPEND_${name}(value) \
{ plb_${name}_list[(*plb_${count_sharing[name]}_index) - 1] \
= value; }
%else:
#define APPEND_${name}(value) \
{ plb_${name}_list[(*plb_${name}_index)++] = value; }
%endif
%endfor
%endif
#define LIST_ARG_DECL ${user_list_arg_decl}
#define LIST_ARGS ${user_list_args}
#define USER_ARG_DECL ${user_arg_decl}
#define USER_ARGS ${user_args}
// }}}
${generate_template}
// {{{ kernel entry point
__kernel
%if do_not_vectorize:
__attribute__((reqd_work_group_size(1, 1, 1)))
%endif
void ${kernel_name}(${kernel_list_arg_decl} USER_ARG_DECL index_type n)
{
%if not do_not_vectorize:
int lid = get_local_id(0);
index_type gsize = get_global_size(0);
index_type work_group_start = get_local_size(0)*get_group_id(0);
for (index_type i = work_group_start + lid; i < n; i += gsize)
%else:
const int chunk_size = 128;
index_type chunk_base = get_global_id(0)*chunk_size;
index_type gsize = get_global_size(0);
for (; chunk_base < n; chunk_base += gsize*chunk_size)
for (index_type i = chunk_base; i < min(n, chunk_base+chunk_size); ++i)
%endif
{
%if is_count_stage:
%for name, dtype in list_names_and_dtypes:
%if name not in count_sharing:
index_type plb_loc_${name}_count = 0;
%endif
%endfor
%else:
%for name, dtype in list_names_and_dtypes:
%if name not in count_sharing:
index_type plb_${name}_index;
if (plb_${name}_start_index)
plb_${name}_index = plb_${name}_start_index[i];
else
plb_${name}_index = 0;
%endif
%endfor
%endif
generate(${kernel_list_arg_values} USER_ARGS i);
%if is_count_stage:
%for name, dtype in list_names_and_dtypes:
%if name not in count_sharing:
if (plb_${name}_count)
plb_${name}_count[i] = plb_loc_${name}_count;
%endif
%endfor
%endif
}
}
// }}}
""", strict_undefined=True)
# }}}
def _get_arg_decl(arg_list):
result = ""
for arg in arg_list:
result += arg.declarator() + ", "
return result
def _get_arg_list(arg_list, prefix=""):
result = ""
for arg in arg_list:
result += prefix + arg.name + ", "
return result
class BuiltList(Record):
pass
class ListOfListsBuilder:
"""Generates and executes code to produce a large number of variable-size
lists, simply.
.. note:: This functionality is provided as a preview. Its interface
is subject to change until this notice is removed.
.. versionadded:: 2013.1
Here's a usage example::
from pyopencl.algorithm import ListOfListsBuilder
builder = ListOfListsBuilder(context, [("mylist", np.int32)], \"\"\"
void generate(LIST_ARG_DECL USER_ARG_DECL index_type i)
{
int count = i % 4;
for (int j = 0; j < count; ++j)
{
APPEND_mylist(count);
}
}
\"\"\", arg_decls=[])
result, event = builder(queue, 2000)
inf = result["mylist"]
assert inf.count == 3000
assert (inf.list.get()[-6:] == [1, 2, 2, 3, 3, 3]).all()
The function `generate` above is called once for each "input object".
Each input object can then generate zero or more list entries.
The number of these input objects is given to :meth:`__call__` as *n_objects*.
List entries are generated by calls to `APPEND_<list name>(value)`.
Multiple lists may be generated at once.
.. automethod:: __init__
.. automethod:: __call__
"""
def __init__(self, context, list_names_and_dtypes, generate_template,
arg_decls, count_sharing=None, devices=None,
name_prefix="plb_build_list", options=[], preamble="",
debug=False, complex_kernel=False):
"""
:arg context: A :class:`pyopencl.Context`.
:arg list_names_and_dtypes: a list of `(name, dtype)` tuples
indicating the lists to be built.
:arg generate_template: a snippet of C as described below
:arg arg_decls: A string of comma-separated C argument declarations.
:arg count_sharing: A mapping consisting of `(child, mother)`
indicating that `mother` and `child` will always have the
same number of indices, and the `APPEND` to `mother`
will always happen *before* the `APPEND` to the child.
:arg name_prefix: the name prefix to use for the compiled kernels
:arg options: OpenCL compilation options for kernels using
*generate_template*.
:arg complex_kernel: If `True`, prevents vectorization on CPUs.
*generate_template* may use the following C macros/identifiers:
* `index_type`: expands to C identifier for the index type used
for the calculation
* `USER_ARG_DECL`: expands to the C declarator for `arg_decls`
* `USER_ARGS`: a list of C argument values corresponding to
`user_arg_decl`
* `LIST_ARG_DECL`: expands to a C argument list representing the
data for the output lists. These are escaped prefixed with
`"plg_"` so as to not interfere with user-provided names.
* `LIST_ARGS`: a list of C argument values corresponding to
`LIST_ARG_DECL`
* `APPEND_name(entry)`: inserts `entry` into the list `name`.
*entry* must be a valid C expression of the correct type.
All argument-list related macros have a trailing comma included
if they are non-empty.
*generate_template* must supply a function:
.. code-block:: c
void generate(USER_ARG_DECL LIST_ARG_DECL index_type i)
{
APPEND_mylist(5);
}
Internally, the `kernel_template` is expanded (at least) twice. Once,
for a 'counting' stage where the size of all the lists is determined,
and a second time, for a 'generation' stage where the lists are
actually filled. A `generate` function that has side effects beyond
calling `append` is therefore ill-formed.
"""
if devices is None:
devices = context.devices
if count_sharing is None:
count_sharing = {}
self.context = context
self.devices = devices
self.list_names_and_dtypes = list_names_and_dtypes
self.generate_template = generate_template
from pyopencl.tools import parse_arg_list
self.arg_decls = parse_arg_list(arg_decls)
self.count_sharing = count_sharing
self.name_prefix = name_prefix
self.preamble = preamble
self.options = options
self.debug = debug
self.complex_kernel = complex_kernel
# {{{ kernel generators
@memoize_method
def get_scan_kernel(self, index_dtype):
from pyopencl.scan import GenericScanKernel
return GenericScanKernel(
self.context, index_dtype,
arguments="__global %s *ary" % dtype_to_ctype(index_dtype),
input_expr="ary[i]",
scan_expr="a+b", neutral="0",
output_statement="ary[i+1] = item;",
devices=self.devices)
def do_not_vectorize(self):
from pytools import any
return (self.complex_kernel
and any(dev.type & cl.device_type.CPU
for dev in self.context.devices))
@memoize_method
def get_count_kernel(self, index_dtype):
index_ctype = dtype_to_ctype(index_dtype)
from pyopencl.tools import VectorArg, OtherArg
kernel_list_args = [
VectorArg(index_dtype, "plb_%s_count" % name)
for name, dtype in self.list_names_and_dtypes
if name not in self.count_sharing]
user_list_args = []
for name, dtype in self.list_names_and_dtypes:
if name in self.count_sharing:
continue
name = "plb_loc_%s_count" % name
user_list_args.append(OtherArg("%s *%s" % (
index_ctype, name), name))
kernel_name = self.name_prefix+"_count"
from pyopencl.characterize import has_double_support
src = _LIST_BUILDER_TEMPLATE.render(
is_count_stage=True,
kernel_name=kernel_name,
double_support=all(has_double_support(dev) for dev in
self.context.devices),
debug=self.debug,
do_not_vectorize=self.do_not_vectorize(),
kernel_list_arg_decl=_get_arg_decl(kernel_list_args),
kernel_list_arg_values=_get_arg_list(user_list_args, prefix="&"),
user_list_arg_decl=_get_arg_decl(user_list_args),
user_list_args=_get_arg_list(user_list_args),
user_arg_decl=_get_arg_decl(self.arg_decls),
user_args=_get_arg_list(self.arg_decls),
list_names_and_dtypes=self.list_names_and_dtypes,
count_sharing=self.count_sharing,
name_prefix=self.name_prefix,
generate_template=self.generate_template,
preamble=self.preamble,
index_type=index_ctype,
)
src = str(src)
prg = cl.Program(self.context, src).build(self.options)
knl = getattr(prg, kernel_name)
from pyopencl.tools import get_arg_list_scalar_arg_dtypes
knl.set_scalar_arg_dtypes(get_arg_list_scalar_arg_dtypes(
kernel_list_args+self.arg_decls) + [index_dtype])
return knl
@memoize_method
def get_write_kernel(self, index_dtype):
index_ctype = dtype_to_ctype(index_dtype)
from pyopencl.tools import VectorArg, OtherArg
kernel_list_args = []
kernel_list_arg_values = ""
user_list_args = []
for name, dtype in self.list_names_and_dtypes:
list_name = "plb_%s_list" % name
list_arg = VectorArg(dtype, list_name)
kernel_list_args.append(list_arg)
user_list_args.append(list_arg)
if name in self.count_sharing:
kernel_list_arg_values += "%s, " % list_name
continue
kernel_list_args.append(
VectorArg(index_dtype, "plb_%s_start_index" % name))
index_name = "plb_%s_index" % name
user_list_args.append(OtherArg("%s *%s" % (
index_ctype, index_name), index_name))
kernel_list_arg_values += "%s, &%s, " % (list_name, index_name)
kernel_name = self.name_prefix+"_write"
from pyopencl.characterize import has_double_support
src = _LIST_BUILDER_TEMPLATE.render(
is_count_stage=False,
kernel_name=kernel_name,
double_support=all(has_double_support(dev) for dev in
self.context.devices),
debug=self.debug,
do_not_vectorize=self.do_not_vectorize(),
kernel_list_arg_decl=_get_arg_decl(kernel_list_args),
kernel_list_arg_values=kernel_list_arg_values,
user_list_arg_decl=_get_arg_decl(user_list_args),
user_list_args=_get_arg_list(user_list_args),
user_arg_decl=_get_arg_decl(self.arg_decls),
user_args=_get_arg_list(self.arg_decls),
list_names_and_dtypes=self.list_names_and_dtypes,
count_sharing=self.count_sharing,
name_prefix=self.name_prefix,
generate_template=self.generate_template,
preamble=self.preamble,
index_type=index_ctype,
)
src = str(src)
prg = cl.Program(self.context, src).build(self.options)
knl = getattr(prg, kernel_name)
from pyopencl.tools import get_arg_list_scalar_arg_dtypes
knl.set_scalar_arg_dtypes(get_arg_list_scalar_arg_dtypes(
kernel_list_args+self.arg_decls) + [index_dtype])
return knl
# }}}
# {{{ driver
def __call__(self, queue, n_objects, *args, **kwargs):
"""
:arg args: arguments corresponding to arg_decls in the constructor.
:class:`pyopencl.array.Array` are not allowed directly and should
be passed as their :attr:`pyopencl.array.Array.data` attribute instead.
:arg allocator: optionally, the allocator to use to allocate new
arrays.
:arg omit_lists: An iterable of list names that should *not* be built
with this invocation. The kernel code may *not* call ``APPEND_name``
for these omitted lists. If it does, undefined behavior will result.
The returned *lists* dictionary will not contain an entry for names
in *omit_lists*.
:arg wait_for: |explain-waitfor|
:returns: a tuple ``(lists, event)``, where
*lists* a mapping from (built) list names to objects which
have attributes
* ``count`` for the total number of entries in all lists combined
* ``lists`` for the array containing all lists.
* ``starts`` for the array of starting indices in `lists`.
`starts` is built so that it has n+1 entries, so that
the *i*'th entry is the start of the *i*'th list, and the
*i*'th entry is the index one past the *i*'th list's end,
even for the last list.
This implies that all lists are contiguous.
*event* is a :class:`pyopencl.Event` for dependency management.
.. versionchanged:: 2016.2
Added omit_lists.
"""
if n_objects >= int(np.iinfo(np.int32).max):
index_dtype = np.int64
else:
index_dtype = np.int32
index_dtype = np.dtype(index_dtype)
allocator = kwargs.pop("allocator", None)
omit_lists = kwargs.pop("omit_lists", [])
wait_for = kwargs.pop("wait_for", None)
if kwargs:
raise TypeError("invalid keyword arguments: '%s'" % ", ".join(kwargs))
for l in omit_lists:
if not any(l == name for name, _ in self.list_names_and_dtypes):
raise ValueError("invalid list name '%s' in omit_lists")
result = {}
count_list_args = []
if wait_for is None:
wait_for = []
count_kernel = self.get_count_kernel(index_dtype)
write_kernel = self.get_write_kernel(index_dtype)
scan_kernel = self.get_scan_kernel(index_dtype)
# {{{ allocate memory for counts
for name, dtype in self.list_names_and_dtypes:
if name in self.count_sharing:
continue
if name in omit_lists:
count_list_args.append(None)
continue
counts = cl.array.empty(queue,
(n_objects + 1), index_dtype, allocator=allocator)
counts[-1] = 0
wait_for = wait_for + counts.events
# The scan will turn the "counts" array into the "starts" array
# in-place.
result[name] = BuiltList(starts=counts)
count_list_args.append(counts.data)
# }}}
if self.debug:
gsize = (1,)
lsize = (1,)
elif self.complex_kernel and queue.device.type == cl.device_type.CPU:
gsize = (4*queue.device.max_compute_units,)
lsize = (1,)
else:
from pyopencl.array import splay
gsize, lsize = splay(queue, n_objects)
count_event = count_kernel(queue, gsize, lsize,
*(tuple(count_list_args) + args + (n_objects,)),
**dict(wait_for=wait_for))
# {{{ run scans
scan_events = []
for name, dtype in self.list_names_and_dtypes:
if name in self.count_sharing:
continue
if name in omit_lists:
continue
info_record = result[name]
starts_ary = info_record.starts
evt = scan_kernel(starts_ary, wait_for=[count_event],
size=n_objects)
starts_ary.setitem(0, 0, queue=queue, wait_for=[evt])
scan_events.extend(starts_ary.events)
# retrieve count
info_record.count = int(starts_ary[-1].get())
# }}}
# {{{ deal with count-sharing lists, allocate memory for lists
write_list_args = []
for name, dtype in self.list_names_and_dtypes:
if name in omit_lists:
write_list_args.append(None)
if name not in self.count_sharing:
write_list_args.append(None)
continue
if name in self.count_sharing:
sharing_from = self.count_sharing[name]
info_record = result[name] = BuiltList(
count=result[sharing_from].count,
starts=result[sharing_from].starts,
)
else:
info_record = result[name]
info_record.lists = cl.array.empty(queue,
info_record.count, dtype, allocator=allocator)
write_list_args.append(info_record.lists.data)
if name not in self.count_sharing:
write_list_args.append(info_record.starts.data)
# }}}
evt = write_kernel(queue, gsize, lsize,
*(tuple(write_list_args) + args + (n_objects,)),
**dict(wait_for=scan_events))
return result, evt
# }}}
# }}}
# {{{ key-value sorting
class _KernelInfo(Record):
pass
def _make_cl_int_literal(value, dtype):
iinfo = np.iinfo(dtype)
result = str(int(value))
if dtype.itemsize == 8:
result += "l"
if int(iinfo.min) < 0:
result += "u"
return result
class KeyValueSorter(object):
"""Given arrays *values* and *keys* of equal length
and a number *nkeys* of keys, returns a tuple `(starts,
lists)`, as follows: *values* and *keys* are sorted
by *keys*, and the sorted *values* is returned as
*lists*. Then for each index *i* in `range(nkeys)`,
*starts[i]* is written to indicating where the
group of *values* belonging to the key with index
*i* begins. It implicitly ends at *starts[i+1]*.
`starts` is built so that it has `nkeys+1` entries, so that
the *i*'th entry is the start of the *i*'th list, and the
*i*'th entry is the index one past the *i*'th list's end,
even for the last list.
This implies that all lists are contiguous.
.. note:: This functionality is provided as a preview. Its
interface is subject to change until this notice is removed.
.. versionadded:: 2013.1
"""
def __init__(self, context):
self.context = context
@memoize_method
def get_kernels(self, key_dtype, value_dtype, starts_dtype):
from pyopencl.algorithm import RadixSort
from pyopencl.tools import VectorArg, ScalarArg
by_target_sorter = RadixSort(
self.context, [
VectorArg(value_dtype, "values"),
VectorArg(key_dtype, "keys"),
],
key_expr="keys[i]",
sort_arg_names=["values", "keys"])
from pyopencl.elementwise import ElementwiseTemplate
start_finder = ElementwiseTemplate(
arguments="""//CL//
starts_t *key_group_starts,
key_t *keys_sorted_by_key,
""",
operation=r"""//CL//
key_t my_key = keys_sorted_by_key[i];
if (i == 0 || my_key != keys_sorted_by_key[i-1])
key_group_starts[my_key] = i;
""",
name="find_starts").build(self.context,
type_aliases=(
("key_t", starts_dtype),
("starts_t", starts_dtype),
),
var_values=())
from pyopencl.scan import GenericScanKernel
bound_propagation_scan = GenericScanKernel(
self.context, starts_dtype,
arguments=[
VectorArg(starts_dtype, "starts"),
# starts has length n+1
ScalarArg(key_dtype, "nkeys"),
],
input_expr="starts[nkeys-i]",
scan_expr="min(a, b)",
neutral=_make_cl_int_literal(
np.iinfo(starts_dtype).max, starts_dtype),
output_statement="starts[nkeys-i] = item;")
return _KernelInfo(
by_target_sorter=by_target_sorter,
start_finder=start_finder,
bound_propagation_scan=bound_propagation_scan)
def __call__(self, queue, keys, values, nkeys,
starts_dtype, allocator=None, wait_for=None):
if allocator is None:
allocator = values.allocator
knl_info = self.get_kernels(keys.dtype, values.dtype,
starts_dtype)
(values_sorted_by_key, keys_sorted_by_key), evt = knl_info.by_target_sorter(
values, keys, queue=queue, wait_for=wait_for)
starts = (cl.array.empty(queue, (nkeys+1), starts_dtype, allocator=allocator)
.fill(len(values_sorted_by_key), wait_for=[evt]))
evt, = starts.events
evt = knl_info.start_finder(starts, keys_sorted_by_key,
range=slice(len(keys_sorted_by_key)),
wait_for=[evt])
evt = knl_info.bound_propagation_scan(starts, nkeys,
queue=queue, wait_for=[evt])
return starts, values_sorted_by_key, evt
# }}}
# vim: filetype=pyopencl:fdm=marker
|