forked from sureshgl/taskflow
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathfor_each.hpp
More file actions
205 lines (162 loc) · 5.45 KB
/
for_each.hpp
File metadata and controls
205 lines (162 loc) · 5.45 KB
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
// 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
#include "../core/executor.hpp"
namespace tf {
// ----------------------------------------------------------------------------
// default parallel for
// ----------------------------------------------------------------------------
// Function: for_each
template <typename B, typename E, typename C>
Task FlowBuilder::for_each(B beg, E end, C c) {
using B_t = std::decay_t<unwrap_ref_decay_t<B>>;
using E_t = std::decay_t<unwrap_ref_decay_t<E>>;
using namespace std::string_literals;
Task task = emplace([b=beg, e=end, c] (Subflow& sf) mutable {
// fetch the stateful values
B_t beg = b;
E_t end = e;
if(beg == end) {
return;
}
size_t chunk_size = 1;
size_t W = sf._executor.num_workers();
size_t N = std::distance(beg, end);
// only myself - no need to spawn another graph
if(W <= 1 || N <= chunk_size) {
std::for_each(beg, end, c);
return;
}
if(N < W) {
W = N;
}
std::atomic<size_t> next(0);
for(size_t w=0; w<W; w++) {
//sf.emplace([&next, beg, N, chunk_size, W, c] () mutable {
sf._named_silent_async(
sf._worker, "part-"s + std::to_string(w), [=, &next] () mutable {
size_t z = 0;
size_t p1 = 2 * W * (chunk_size + 1);
double p2 = 0.5 / static_cast<double>(W);
size_t s0 = next.load(std::memory_order_relaxed);
while(s0 < N) {
size_t r = N - s0;
// fine-grained
if(r < p1) {
while(1) {
s0 = next.fetch_add(chunk_size, std::memory_order_relaxed);
if(s0 >= N) {
return;
}
size_t e0 = (chunk_size <= (N - s0)) ? s0 + chunk_size : N;
std::advance(beg, s0-z);
for(size_t x=s0; x<e0; x++) {
c(*beg++);
}
z = e0;
}
break;
}
// coarse-grained
else {
size_t q = static_cast<size_t>(p2 * r);
if(q < chunk_size) {
q = chunk_size;
}
size_t e0 = (q <= r) ? s0 + q : N;
if(next.compare_exchange_strong(s0, e0, std::memory_order_relaxed,
std::memory_order_relaxed)) {
std::advance(beg, s0-z);
for(size_t x = s0; x< e0; x++) {
c(*beg++);
}
z = e0;
s0 = next.load(std::memory_order_relaxed);
}
}
}
});
}
sf.join();
});
return task;
}
// Function: for_each_index
template <typename B, typename E, typename S, typename C>
Task FlowBuilder::for_each_index(B beg, E end, S inc, C c){
using namespace std::string_literals;
using B_t = std::decay_t<unwrap_ref_decay_t<B>>;
using E_t = std::decay_t<unwrap_ref_decay_t<E>>;
using S_t = std::decay_t<unwrap_ref_decay_t<S>>;
Task task = emplace([b=beg, e=end, a=inc, c] (Subflow& sf) mutable {
// fetch the iterator values
B_t beg = b;
E_t end = e;
S_t inc = a;
if(is_range_invalid(beg, end, inc)) {
TF_THROW("invalid range [", beg, ", ", end, ") with step size ", inc);
}
size_t chunk_size = 1;
size_t W = sf._executor.num_workers();
size_t N = distance(beg, end, inc);
// only myself - no need to spawn another graph
if(W <= 1 || N <= chunk_size) {
for(size_t x=0; x<N; x++, beg+=inc) {
c(beg);
}
return;
}
if(N < W) {
W = N;
}
std::atomic<size_t> next(0);
for(size_t w=0; w<W; w++) {
//sf.emplace([&next, beg, inc, N, chunk_size, W, c] () mutable {
sf._named_silent_async(
sf._worker, "part-"s + std::to_string(w), [=, &next] () mutable {
size_t p1 = 2 * W * (chunk_size + 1);
double p2 = 0.5 / static_cast<double>(W);
size_t s0 = next.load(std::memory_order_relaxed);
while(s0 < N) {
size_t r = N - s0;
// fine-grained
if(r < p1) {
while(1) {
s0 = next.fetch_add(chunk_size, std::memory_order_relaxed);
if(s0 >= N) {
return;
}
size_t e0 = (chunk_size <= (N - s0)) ? s0 + chunk_size : N;
auto s = static_cast<B_t>(s0) * inc + beg;
for(size_t x=s0; x<e0; x++, s+=inc) {
c(s);
}
}
break;
}
// coarse-grained
else {
size_t q = static_cast<size_t>(p2 * r);
if(q < chunk_size) {
q = chunk_size;
}
size_t e0 = (q <= r) ? s0 + q : N;
if(next.compare_exchange_strong(s0, e0, std::memory_order_relaxed,
std::memory_order_relaxed)) {
auto s = static_cast<B_t>(s0) * inc + beg;
for(size_t x=s0; x<e0; x++, s+= inc) {
c(s);
}
s0 = next.load(std::memory_order_relaxed);
}
}
}
//}).name("pfg_"s + std::to_string(w));
});
}
sf.join();
});
return task;
}
} // end of namespace tf -----------------------------------------------------