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
|
#include "HCheckConfig.h"
#if 0
#include <omp.h>
#endif
#include <iostream>
#include "catch.hpp"
#include "matrix_multiplication.hpp"
#include "parallel/HighsParallel.h"
using namespace highs;
const int numThreads = (std::thread::hardware_concurrency() + 1) / 2;
const bool dev_run = false;
int64_t fib_sequential(const int64_t n) {
if (n <= 1) return 1;
return fib_sequential(n - 1) + fib_sequential(n - 2);
}
int64_t fib(const int64_t n) {
if (n <= 20) return fib_sequential(n);
int64_t n1;
parallel::spawn([&]() { n1 = fib(n - 1); });
int64_t n2 = fib(n - 2);
parallel::sync();
// printf("fib(%ld) = %ld + %ld = %ld\n", n, n1, n2, n1 + n2);
return n1 + n2;
}
#if 0
int64_t fib_omp(const int64_t n) {
if (n <= 20) return fib_sequential(n);
int64_t n1;
#pragma omp task shared(n1)
n1 = fib_omp(n - 1);
int64_t n2 = fib_omp(n - 2);
#pragma omp taskwait
// printf("fib(%ld) = %ld + %ld = %ld\n", n, n1, n2, n1 + n2);
return n1 + n2;
}
// matrix_multiplication_omp
// reference: https://computing.llnl.gov/tutorials/openMP/samples/C/omp_mm.c
void matrix_multiplication_omp(unsigned nthreads) {
omp_set_num_threads(nthreads);
int i, j, k;
#pragma omp parallel for private(i, j)
for (i = 0; i < N; ++i) {
for (j = 0; j < N; j++) {
a[i][j] = i + j;
}
}
#pragma omp parallel for private(i, j)
for (i = 0; i < N; ++i) {
for (j = 0; j < N; j++) {
b[i][j] = i * j;
}
}
#pragma omp parallel for private(i, j)
for (i = 0; i < N; ++i) {
for (j = 0; j < N; j++) {
c[i][j] = 0;
}
}
#pragma omp parallel for private(i, j, k)
for (i = 0; i < N; ++i) {
for (j = 0; j < N; j++) {
for (k = 0; k < N; k++) {
c[i][j] += a[i][k] * b[k][j];
}
}
}
// int edge;
//#pragma omp parallel shared(a, b, c, nthreads) private(i, j, k)
//{
// #pragma omp single private(i, j)
// for(i = 0; i<N; i++) {
// #pragma omp task private(j) firstprivate(i) depend(out: edge)
// for (j=0; j<N; j++)
// a[i][j]= i+j;
// }
// #pragma omp single private(i, j)
// for(i = 0; i<N; i++) {
// #pragma omp task private(j) firstprivate(i) depend(out: edge)
// for (j=0; j<N; j++)
// b[i][j]= i*j;
// }
// #pragma omp single private(i, j)
// for(i = 0; i<N; i++) {
// #pragma omp task private(j) firstprivate(i) depend(out: edge)
// for (j=0; j<N; j++)
// c[i][j]= 0;
// }
// #pragma omp single private(i, j)
// for(i = 0; i<N; i++) {
// #pragma omp task private(j, k) firstprivate(i) depend(in: edge)
// for(j=0; j<N; j++) {
// for (k=0; k<N; k++) {
// c[i][j] += a[i][k] * b[k][j];
// }
// }
// }
//}
// std::cout << reduce_sum() << std::endl;
}
#endif
void matrix_multiplication_highs(unsigned nthreads) {
// #pragma omp parallel for private(i, j)
parallel::for_each(0, N, [&](HighsInt start, HighsInt end) {
for (int i = start; i < end; ++i) {
for (int j = 0; j < N; j++) {
a[i][j] = i + j;
}
}
});
parallel::for_each(0, N, [&](HighsInt start, HighsInt end) {
for (int i = start; i < end; ++i) {
for (int j = 0; j < N; j++) {
b[i][j] = i * j;
}
}
});
parallel::for_each(0, N, [&](HighsInt start, HighsInt end) {
for (int i = start; i < end; ++i) {
for (int j = 0; j < N; j++) {
c[i][j] = 0;
}
}
});
parallel::for_each(0, N, [&](HighsInt start, HighsInt end) {
for (int i = start; i < end; ++i) {
for (int j = 0; j < N; j++) {
for (int k = 0; k < N; k++) {
c[i][j] += a[i][k] * b[k][j];
}
}
}
});
}
#if 0
std::chrono::microseconds measure_time_omp(unsigned num_threads) {
auto beg = std::chrono::high_resolution_clock::now();
matrix_multiplication_omp(num_threads);
auto end = std::chrono::high_resolution_clock::now();
return std::chrono::duration_cast<std::chrono::microseconds>(end - beg);
}
#endif
std::chrono::microseconds measure_time_highs(unsigned num_threads) {
auto beg = std::chrono::high_resolution_clock::now();
matrix_multiplication_highs(num_threads);
auto end = std::chrono::high_resolution_clock::now();
return std::chrono::duration_cast<std::chrono::microseconds>(end - beg);
}
int N = 0;
double **a = nullptr, **b = nullptr, **c = nullptr;
void matrix_multiplication(const std::string& model, const unsigned num_threads,
const unsigned num_rounds) {
if (dev_run)
std::cout << std::setw(12) << "size" << std::setw(12) << "runtime"
<< std::endl;
for (int i = 128; i <= 512; i += 128) {
N = i;
allocate_matrix();
double runtime{0.0};
for (unsigned j = 0; j < num_rounds; ++j) {
#if 0
if (model == "highs") {
runtime += measure_time_highs(num_threads).count();
} else if (model == "omp") {
runtime += measure_time_omp(num_threads).count();
} else
assert(false);
#else
runtime += measure_time_highs(num_threads).count();
#endif
}
if (dev_run)
std::cout << std::setw(12) << N << std::setw(12)
<< runtime / num_rounds / 1e3 << std::endl;
deallocate_matrix();
}
}
TEST_CASE("MatrixMultHighs", "[parallel]") {
HighsTaskExecutor::shutdown();
parallel::initialize_scheduler(numThreads);
if (dev_run) std::cout << "\nhighs workstealing for loop:" << std::endl;
matrix_multiplication("highs", parallel::num_threads(), 1);
}
TEST_CASE("FibonacciTasksHighs", "[parallel]") {
HighsTaskExecutor::shutdown();
auto beg = std::chrono::high_resolution_clock::now();
parallel::initialize_scheduler(numThreads);
int64_t result = fib(41);
auto end = std::chrono::high_resolution_clock::now();
if (dev_run)
std::cout << "time elapsed for fib(41) with HiGHS work stealing: "
<< (std::chrono::duration_cast<std::chrono::microseconds>(end -
beg)
.count() /
1e3)
<< "ms" << std::endl;
// REQUIRE(result == 4807526976);
// REQUIRE(result == 20365011074);
// fib 41
REQUIRE(result == 267914296);
}
#if 0
TEST_CASE("MatrixMultOmp", "[parallel]") {
if (dev_run)
std::cout << "\nomp for loop:" << std::endl;
matrix_multiplication("omp", numThreads, 1);
}
TEST_CASE("FibonacciTasksOmp", "[parallel]") {
int64_t result;
auto beg = std::chrono::high_resolution_clock::now();
#pragma omp parallel num_threads(numThreads)
{
#pragma omp single
{ result = fib_omp(41); }
}
auto end = std::chrono::high_resolution_clock::now();
if (dev_run)
std::cout << "time elapsed for fib(41) with omp tasks: "
<< (std::chrono::duration_cast<std::chrono::microseconds>(end - beg)
.count() / 1e3)
<< "ms" << std::endl;
// fib 41
REQUIRE(result == 267914296);
}
#endif
|
