FazBrowse GitHub Viewer
|
Trending
|
URL:
|
Home
Tools:
[Download Repo ZIP]
[View Raw Code]
[Original HTTPS Page]
taskflow/taskflow/algorithm/partitioner.hpp at master · PentaBug/taskflow · GitHub
PentaBug
taskflow
Repository navigation
Code
Pull requests
Actions
Projects
Security and quality
Insights
Expand file tree
Breadcrumbs
taskflow
/
taskflow
/
algorithm
/
partitioner.hpp
Copy path
More file actions
More file actions
Latest commit
History
History
History
809 lines (655 loc) · 21.5 KB
Breadcrumbs
taskflow
/
taskflow
/
algorithm
/
partitioner.hpp
Copy path
File metadata and controls
809 lines (655 loc) · 21.5 KB
Raw
Copy raw file
Download raw file
Open symbols panel
Edit and raw actions
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
//
reference:
//
- gomp: https://github.com/gcc-mirror/gcc/blob/master/libgomp/iter.c
//
- komp: https://github.com/llvm-mirror/openmp/blob/master/runtime/src/kmp_dispatch.cpp
#
pragma
once
/*
*
@file partitioner.hpp
@brief partitioner include file
*/
namespace
tf
{
/*
*
@enum PartitionerType
@brief enumeration of all partitioner types
*/
enum
class
PartitionerType
:
int
{
/*
* @brief static partitioner type
*/
STATIC
,
/*
* @brief dynamic partitioner type
*/
DYNAMIC
};
//
template <typename C>
//
class PartitionInvoker : public PartitionerBase {
//
//
protected
//
//
C _closure;
//
//
template <typename... ArgsT>
//
auto operator()(ArgsT&&... args) {
//
return std::invoke(closure, std::forward<ArgsT>(args)...);
//
}
//
//
template <typename... ArgsT>
//
auto operator()(ArgsT&&... args) const {
//
return std::invoke(closure, std::forward<ArgsT>(args)...);
//
}
//
//
};
/*
*
@struct DefaultClosureWrapper
@brief default closure wrapper that simplies runs the given closure as is
*/
struct
DefaultClosureWrapper
{
};
/*
*
@private
*/
struct
IsPartitioner
{
};
//
----------------------------------------------------------------------------
//
Partitioner Base
//
----------------------------------------------------------------------------
/*
*
@class PartitionerBase
@brief class to derive a partitioner for scheduling parallel algorithms
@tparam C closure wrapper type
The class provides base methods to derive a partitioner that can be used
to schedule parallel iterations (e.g., tf::Taskflow::for_each).
An partitioner defines the scheduling method for running parallel algorithms,
such tf::Taskflow::for_each, tf::Taskflow::reduce, and so on.
By default, we provide the following partitioners:
+ tf::GuidedPartitioner to enable guided scheduling algorithm of adaptive chunk size
+ tf::DynamicPartitioner to enable dynamic scheduling algorithm of equal chunk size
+ tf::StaticPartitioner to enable static scheduling algorithm of static chunk size
+ tf::RandomPartitioner to enable random scheduling algorithm of random chunk size
Depending on applications, partitioning algorithms can impact the performance
a lot.
For example, if a parallel-iteration workload contains a regular work unit per
iteration, tf::StaticPartitioner can deliver the best performance.
On the other hand, if the work unit per iteration is irregular and unbalanced,
tf::GuidedPartitioner or tf::DynamicPartitioner can outperform tf::StaticPartitioner.
In most situations, tf::GuidedPartitioner can deliver decent performance and
is thus used as our default partitioner.
@note
Giving the partition size of 0 lets the %Taskflow runtime automatically determines
the partition size for the given partitioner.
In addition to partition size, the application can specify a closure wrapper
for a partitioner.
A closure wrapper allows the application to wrapper a partitioned task
(i.e., closure) with a custom function object that performs additional tasks.
For example:
@code{.cpp}
std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1,
[](){
printf("%d\n", i);
},
tf::StaticPartitioner(0, [](auto&& closure){
// do something before invoking the partitioned task
// ...
// invoke the partitioned task
closure();
// do something else after invoking the partitioned task
// ...
}
);
executor.run(taskflow).wait();
@endcode
@note
The default closure wrapper (tf::DefaultClosureWrapper) does nothing but invoke
the partitioned task (closure).
*/
template
<
typename
C = DefaultClosureWrapper>
class
PartitionerBase
:
public
IsPartitioner
{
public:
/*
*
@brief the closure type
*/
using
closure_wrapper_type = C;
/*
*
@brief default constructor
*/
PartitionerBase
() =
default
;
/*
*
@brief construct a partitioner with the given chunk size
*/
explicit
PartitionerBase
(
size_t
chunk_size) : _chunk_size {chunk_size} {}
/*
*
@brief construct a partitioner with the given chunk size and closure wrapper
*/
PartitionerBase
(
size_t
chunk_size, C&& closure_wrapper) :
_chunk_size {chunk_size},
_closure_wrapper {std::forward<C>(closure_wrapper)} {
}
/*
*
@brief query the chunk size of this partitioner
*/
size_t
chunk_size
()
const
{
return
_chunk_size; }
/*
*
@brief update the chunk size of this partitioner
*/
void
chunk_size
(
size_t
cz) { _chunk_size = cz; }
/*
*
@brief acquire an immutable access to the closure wrapper object
*/
const
C&
closure_wrapper
()
const
{
return
_closure_wrapper; }
/*
*
@brief modify the closure wrapper object
*/
template
<
typename
F>
void
closure_wrapper
(F&& fn) { _closure_wrapper = std::forward<F>(fn); }
protected:
/*
*
@brief chunk size
*/
size_t
_chunk_size{
0
};
/*
*
@brief closure wrapper
*/
C _closure_wrapper;
};
//
----------------------------------------------------------------------------
//
Guided Partitioner
//
----------------------------------------------------------------------------
/*
*
@class GuidedPartitioner
@tparam C closure wrapper type (default tf::DefaultClosureWrapper)
@brief class to construct a guided partitioner for scheduling parallel algorithms
The size of a partition is proportional to the number of unassigned iterations
divided by the number of workers,
and the size will gradually decrease to the given chunk size.
The last partition may be smaller than the chunk size.
In addition to partition size, the application can specify a closure wrapper
for a guided partitioner.
A closure wrapper allows the application to wrapper a partitioned task
(i.e., closure) with a custom function object that performs additional tasks.
For example:
@code{.cpp}
std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1,
[](){
printf("%d\n", i);
},
tf::GuidedPartitioner(0, [](auto&& closure){
// do something before invoking the partitioned task
// ...
// invoke the partitioned task
closure();
// do something else after invoking the partitioned task
// ...
}
);
executor.run(taskflow).wait();
@endcode
*/
template
<
typename
C = DefaultClosureWrapper>
class
GuidedPartitioner
:
public
PartitionerBase
<C> {
public:
/*
*
@brief queries the partition type (dynamic)
*/
static
constexpr
PartitionerType
type
() {
return
PartitionerType::
DYNAMIC
; }
/*
*
@brief default constructor
*/
GuidedPartitioner
() =
default
;
/*
*
@brief construct a guided partitioner with the given chunk size
*/
explicit
GuidedPartitioner
(
size_t
sz) : PartitionerBase<C> (sz) {}
/*
*
@brief construct a guided partitioner with the given chunk size and the closure
*/
explicit
GuidedPartitioner
(
size_t
sz, C&& closure) :
PartitionerBase<C>(sz, std::forward<C>(closure)) {
}
//
--------------------------------------------------------------------------
//
scheduling methods
//
--------------------------------------------------------------------------
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
void
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop
(
size_t
N,
size_t
W, std::atomic<
size_t
>& next, F&& func
)
const
{
size_t
chunk_size = (
this
->
_chunk_size
==
0
) ?
size_t
{
1
} :
this
->
_chunk_size
;
size_t
p1 =
2
* W * (chunk_size +
1
);
float
p2 =
0
.
5f
/
static_cast
<
float
>(W);
size_t
curr_b = next.
load
(std::memory_order_relaxed);
while
(curr_b < N) {
size_t
r = N - curr_b;
//
fine-grained
if
(r < p1) {
while
(
1
) {
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
if
(curr_b >= N) {
return
;
}
func
(curr_b,
std::min
(curr_b + chunk_size, N));
}
break
;
}
//
coarse-grained
else
{
size_t
q =
static_cast
<
size_t
>(p2 * r);
if
(q < chunk_size) {
q = chunk_size;
}
//
size_t curr_e = (q <= r) ? curr_b + q : N;
size_t
curr_e =
std::min
(curr_b + q, N);
if
(next.
compare_exchange_strong
(curr_b, curr_e, std::memory_order_relaxed,
std::memory_order_relaxed)) {
func
(curr_b, curr_e);
curr_b = next.
load
(std::memory_order_relaxed);
}
}
}
}
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
bool
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop_until
(
size_t
N,
size_t
W, std::atomic<
size_t
>& next, F&& func
)
const
{
size_t
chunk_size = (
this
->
_chunk_size
==
0
) ?
size_t
{
1
} :
this
->
_chunk_size
;
size_t
p1 =
2
* W * (chunk_size +
1
);
float
p2 =
0
.
5f
/
static_cast
<
float
>(W);
size_t
curr_b = next.
load
(std::memory_order_relaxed);
while
(curr_b < N) {
size_t
r = N - curr_b;
//
fine-grained
if
(r < p1) {
while
(
1
) {
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
if
(curr_b >= N) {
return
;
}
if
(
func
(curr_b,
std::min
(curr_b + chunk_size, N))) {
return
;
}
}
break
;
}
//
coarse-grained
else
{
size_t
q =
static_cast
<
size_t
>(p2 * r);
if
(q < chunk_size) {
q = chunk_size;
}
//
size_t curr_e = (q <= r) ? curr_b + q : N;
size_t
curr_e =
std::min
(curr_b + q, N);
if
(next.
compare_exchange_strong
(curr_b, curr_e, std::memory_order_relaxed,
std::memory_order_relaxed)) {
if
(
func
(curr_b, curr_e)) {
return
;
}
curr_b = next.
load
(std::memory_order_relaxed);
}
}
}
}
};
//
----------------------------------------------------------------------------
//
Dynamic Partitioner
//
----------------------------------------------------------------------------
/*
*
@class DynamicPartitioner
@brief class to construct a dynamic partitioner for scheduling parallel algorithms
@tparam C closure wrapper type (default tf::DefaultClosureWrapper)
The partitioner splits iterations into many partitions each of size equal to
the given chunk size.
Different partitions are distributed dynamically to workers
without any specific order.
In addition to partition size, the application can specify a closure wrapper
for a dynamic partitioner.
A closure wrapper allows the application to wrapper a partitioned task
(i.e., closure) with a custom function object that performs additional tasks.
For example:
@code{.cpp}
std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1,
[](){
printf("%d\n", i);
},
tf::DynamicPartitioner(0, [](auto&& closure){
// do something before invoking the partitioned task
// ...
// invoke the partitioned task
closure();
// do something else after invoking the partitioned task
// ...
}
);
executor.run(taskflow).wait();
@endcode
*/
template
<
typename
C = DefaultClosureWrapper>
class
DynamicPartitioner
:
public
PartitionerBase
<C> {
public:
/*
*
@brief queries the partition type (dynamic)
*/
static
constexpr
PartitionerType
type
() {
return
PartitionerType::
DYNAMIC
; }
/*
*
@brief default constructor
*/
DynamicPartitioner
() =
default
;
/*
*
@brief construct a dynamic partitioner with the given chunk size
*/
explicit
DynamicPartitioner
(
size_t
sz) : PartitionerBase<C>(sz) {}
/*
*
@brief construct a dynamic partitioner with the given chunk size and the closure
*/
explicit
DynamicPartitioner
(
size_t
sz, C&& closure) :
PartitionerBase<C>(sz, std::forward<C>(closure)) {
}
//
--------------------------------------------------------------------------
//
scheduling methods
//
--------------------------------------------------------------------------
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
void
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop
(
size_t
N,
size_t
, std::atomic<
size_t
>& next, F&& func
)
const
{
size_t
chunk_size = (
this
->
_chunk_size
==
0
) ?
size_t
{
1
} :
this
->
_chunk_size
;
size_t
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
while
(curr_b < N) {
func
(curr_b,
std::min
(curr_b + chunk_size, N));
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
}
}
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
bool
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop_until
(
size_t
N,
size_t
, std::atomic<
size_t
>& next, F&& func
)
const
{
size_t
chunk_size = (
this
->
_chunk_size
==
0
) ?
size_t
{
1
} :
this
->
_chunk_size
;
size_t
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
while
(curr_b < N) {
if
(
func
(curr_b,
std::min
(curr_b + chunk_size, N))) {
return
;
}
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
}
}
};
//
----------------------------------------------------------------------------
//
Static Partitioner
//
----------------------------------------------------------------------------
/*
*
@class StaticPartitioner
@brief class to construct a static partitioner for scheduling parallel algorithms
@tparam C closure wrapper type (default tf::DefaultClosureWrapper)
The partitioner divides iterations into chunks and distributes chunks
to workers in order.
If the chunk size is not specified (default @c 0), the partitioner resorts to a chunk size
that equally distributes iterations into workers.
@code{.cpp}
std::vector<int> data = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
taskflow.for_each(
data.begin(), data.end(), [](int i){}, StaticPartitioner(0)
);
executor.run(taskflow).run();
@endcode
In addition to partition size, the application can specify a closure wrapper
for a static partitioner.
A closure wrapper allows the application to wrapper a partitioned task
(i.e., closure) with a custom function object that performs additional tasks.
For example:
@code{.cpp}
std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1,
[](){
printf("%d\n", i);
},
tf::StaticPartitioner(0, [](auto&& closure){
// do something before invoking the partitioned task
// ...
// invoke the partitioned task
closure();
// do something else after invoking the partitioned task
// ...
}
);
executor.run(taskflow).wait();
@endcode
*/
template
<
typename
C = DefaultClosureWrapper>
class
StaticPartitioner
:
public
PartitionerBase
<C> {
public:
/*
*
@brief queries the partition type (static)
*/
static
constexpr
PartitionerType
type
() {
return
PartitionerType::
STATIC
; }
/*
*
@brief default constructor
*/
StaticPartitioner
() =
default
;
/*
*
@brief construct a static partitioner with the given chunk size
*/
explicit
StaticPartitioner
(
size_t
sz) : PartitionerBase<C>(sz) {}
/*
*
@brief construct a static partitioner with the given chunk size and the closure
*/
explicit
StaticPartitioner
(
size_t
sz, C&& closure) :
PartitionerBase<C>(sz, std::forward<C>(closure)) {
}
/*
*
@brief queries the adjusted chunk size
Returns the given chunk size if it is not zero, or returns
<tt>N/W + (w < N%W)</tt>, where @c N is the number of iterations,
@c W is the number of workers, and @c w is the worker ID.
*/
size_t
adjusted_chunk_size
(
size_t
N,
size_t
W,
size_t
w)
const
{
return
this
->
_chunk_size
?
this
->
_chunk_size
: N/W + (w < N%W);
}
//
--------------------------------------------------------------------------
//
scheduling methods
//
--------------------------------------------------------------------------
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
void
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop
(
size_t
N,
size_t
W,
size_t
curr_b,
size_t
chunk_size, F&& func
) {
size_t
stride = W * chunk_size;
while
(curr_b < N) {
size_t
curr_e =
std::min
(curr_b + chunk_size, N);
func
(curr_b, curr_e);
curr_b += stride;
}
}
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
bool
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop_until
(
size_t
N,
size_t
W,
size_t
curr_b,
size_t
chunk_size, F&& func
) {
size_t
stride = W * chunk_size;
while
(curr_b < N) {
size_t
curr_e =
std::min
(curr_b + chunk_size, N);
if
(
func
(curr_b, curr_e)) {
return
;
}
curr_b += stride;
}
}
};
//
----------------------------------------------------------------------------
//
RandomPartitioner
//
----------------------------------------------------------------------------
/*
*
@class RandomPartitioner
@brief class to construct a random partitioner for scheduling parallel algorithms
@tparam C closure wrapper type (default tf::DefaultClosureWrapper)
Similar to tf::DynamicPartitioner,
the partitioner splits iterations into many partitions but each with a random
chunk size in the range, <tt>c = [alpha * N * W, beta * N * W]</tt>.
By default, @c alpha is <tt>0.01</tt> and @c beta is <tt>0.5</tt>, respectively.
In addition to partition size, the application can specify a closure wrapper
for a random partitioner.
A closure wrapper allows the application to wrapper a partitioned task
(i.e., closure) with a custom function object that performs additional tasks.
For example:
@code{.cpp}
std::atomic<int> count = 0;
tf::Taskflow taskflow;
taskflow.for_each_index(0, 100, 1,
[](){
printf("%d\n", i);
},
tf::RandomPartitioner(0, [](auto&& closure){
// do something before invoking the partitioned task
// ...
// invoke the partitioned task
closure();
// do something else after invoking the partitioned task
// ...
}
);
executor.run(taskflow).wait();
@endcode
*/
template
<
typename
C = DefaultClosureWrapper>
class
RandomPartitioner
:
public
PartitionerBase
<C> {
public:
/*
*
@brief queries the partition type (dynamic)
*/
static
constexpr
PartitionerType
type
() {
return
PartitionerType::
DYNAMIC
; }
/*
*
@brief default constructor
*/
RandomPartitioner
() =
default
;
/*
*
@brief construct a dynamic partitioner with the given chunk size
*/
explicit
RandomPartitioner
(
size_t
sz) : PartitionerBase<C>(sz) {}
/*
*
@brief construct a random partitioner with the given chunk size and the closure
*/
explicit
RandomPartitioner
(
size_t
sz, C&& closure) :
PartitionerBase<C>(sz, std::forward<C>(closure)) {
}
/*
*
@brief constructs a random partitioner with the given parameters
*/
RandomPartitioner
(
float
alpha,
float
beta) : _alpha{alpha}, _beta{beta} {}
/*
*
@brief constructs a random partitioner with the given parameters and the closure
*/
RandomPartitioner
(
float
alpha,
float
beta, C&& closure) :
_alpha {alpha}, _beta {beta},
PartitionerBase<C>(
0
, std::forward<C>(closure)) {
}
/*
*
@brief queries the @c alpha value
*/
float
alpha
()
const
{
return
_alpha; }
/*
*
@brief queries the @c beta value
*/
float
beta
()
const
{
return
_beta; }
/*
*
@brief queries the range of chunk size
@param N number of iterations
@param W number of workers
*/
std::pair<
size_t
,
size_t
>
chunk_size_range
(
size_t
N,
size_t
W)
const
{
size_t
b1 =
static_cast
<
size_t
>(_alpha * N * W);
size_t
b2 =
static_cast
<
size_t
>(_beta * N * W);
if
(b1 > b2) {
std::swap
(b1, b2);
}
b1 =
std::max
(b1,
size_t
{
1
});
b2 =
std::max
(b2, b1 +
1
);
return
{b1, b2};
}
//
--------------------------------------------------------------------------
//
scheduling methods
//
--------------------------------------------------------------------------
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
void
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop
(
size_t
N,
size_t
W, std::atomic<
size_t
>& next, F&& func
)
const
{
auto
[b1, b2] =
chunk_size_range
(N, W);
std::default_random_engine engine {std::random_device{}()};
std::uniform_int_distribution<
size_t
>
dist
(b1, b2);
size_t
chunk_size =
dist
(engine);
size_t
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
while
(curr_b < N) {
func
(curr_b,
std::min
(curr_b + chunk_size, N));
chunk_size =
dist
(engine);
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
}
}
/*
*
@private
*/
template
<
typename
F,
std::
enable_if_t
<std::is_invocable_r_v<
bool
, F,
size_t
,
size_t
>,
void
>* =
nullptr
>
void
loop_until
(
size_t
N,
size_t
W, std::atomic<
size_t
>& next, F&& func
)
const
{
auto
[b1, b2] =
chunk_size_range
(N, W);
std::default_random_engine engine {std::random_device{}()};
std::uniform_int_distribution<
size_t
>
dist
(b1, b2);
size_t
chunk_size =
dist
(engine);
size_t
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
while
(curr_b < N) {
if
(
func
(curr_b,
std::min
(curr_b + chunk_size, N))){
return
;
}
chunk_size =
dist
(engine);
curr_b = next.
fetch_add
(chunk_size, std::memory_order_relaxed);
}
}
private:
float
_alpha {
0
.
01f
};
float
_beta {
0
.
5f
};
};
/*
*
@brief default partitioner set to tf::GuidedPartitioner
Guided partitioner can achieve decent performance for most parallel algorithms,
especially for those with irregular and unbalanced workload per iteration.
*/
using
DefaultPartitioner = GuidedPartitioner<>;
/*
*
@brief determines if a type is a partitioner
A partitioner is a derived type from tf::PartitionerBase.
*/
template
<
typename
P>
inline
constexpr
bool
is_partitioner_v = std::is_base_of<IsPartitioner, P>::value;
}
//
end of namespace tf -----------------------------------------------------
Back
|
FazBrowse Home
|
New Git URL