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
|
// The example shows how to use syclFlow to multiply two 2D matrices.
#include <taskflow/taskflow.hpp>
#include <taskflow/sycl/syclflow.hpp>
// Matrix multiplication using GPU
auto gpu(int M, int N, int K) {
std::vector<int> ha, hb, hc;
int *da, *db, *dc;
tf::Executor executor;
tf::Taskflow taskflow("MatrixMultiplication");
sycl::queue queue;
// allocate the host and device storage for a
auto allocate_a = taskflow.emplace([&](){
ha.resize(M*N, M+N);
da = sycl::malloc_device<int>(M*N, queue);
}).name("allocate_a");
// allocate the host and device storage for b
auto allocate_b = taskflow.emplace([&](){
hb.resize(N*K, N+K);
db = sycl::malloc_device<int>(N*K, queue);
}).name("allocate_b");
// allocate the host and device storage for c
auto allocate_c = taskflow.emplace([&](){
hc.resize(M*K);
dc = sycl::malloc_device<int>(M*K, queue);
}).name("allocate_c");
// create a syclFlow to run the matrix multiplication
auto syclFlow = taskflow.emplace_on([&](tf::syclFlow& sf){
// copy data to da, db, and dc
auto copy_da = sf.copy(da, ha.data(), M*N).name("H2D_a");
auto copy_db = sf.copy(db, hb.data(), N*K).name("H2D_b");
auto copy_hc = sf.copy(hc.data(), dc, M*K).name("D2H_c");
auto _M = (M % 16 == 0) ? M : (M + 16 - M % 16);
auto _K = (K % 16 == 0) ? K : (K + 16 - K % 16);
auto kmatmul = sf.parallel_for(
sycl::nd_range<2>{sycl::range<2>(_M, _K ), sycl::range<2>(16, 16)},
[=](sycl::nd_item<2> item) {
int row = item.get_global_id(0);
int col = item.get_global_id(1);
if(row < M && col < K) {
int sum = 0;
for(int n = 0; n < N; n++) {
sum += da[row * N + n] * db[n * K + col];
}
dc[row * K + col] = sum;
}
}
).name("matmul");
// It is also possible to just use range and let the runtime decide the
// partition of groups, but the result is less efficient.
//
//auto kmatmul = sf.parallel_for(
// sycl::range<2>(M, K),
// [=](sycl::id<2> id) {
// int row = id[0];
// int col = id[1];
// int sum = 0;
// for(int n = 0; n < N; n++) {
// sum += da[row * N + n] * db[n * K + col];
// }
// dc[row * K + col] = sum;
// }
//).name("matmul");
kmatmul.succeed(copy_da, copy_db)
.precede(copy_hc);
}, queue).name("syclFlow");
auto free = taskflow.emplace([&](){
sycl::free(da, queue);
sycl::free(db, queue);
sycl::free(dc, queue);
}).name("free");
syclFlow.succeed(allocate_a, allocate_b, allocate_c)
.precede(free);
executor.run(taskflow).wait();
// You may uncomment the line below to dump the task graph
//taskflow.dump(std::cout);
return hc;
}
// Matrix multiplication using CPU
auto cpu(int M, int N, int K) {
std::vector<int> a, b, c;
tf::Executor executor;
tf::Taskflow taskflow;
auto ha = taskflow.emplace([&](){
a.resize(M*N, M+N);
}).name("allocate_a");
auto hb = taskflow.emplace([&](){
b.resize(N*K, N+K);
}).name("allocate_b");
auto hc = taskflow.emplace([&](){
c.resize(M*K, 0);
}).name("allocate_c");
auto pf = taskflow.for_each_index(0, M, 1, [&] (int m) {
for(int k=0; k<K; k++) {
for(int n=0; n<N; n++) {
c[m*K+k] += (a[m*N+n]*b[n*K+k]);
}
}
});
pf.succeed(ha, hb, hc);
//taskflow.dump(std::cout);
executor.run(taskflow).wait();
return c;
}
// Function: main
int main(int argc, char *argv[]) {
if(argc != 4) {
std::cerr << "usage: matrix-multiplication M N K\n";
std::exit(EXIT_FAILURE);
}
int M = std::atoi(argv[1]);
int N = std::atoi(argv[2]);
int K = std::atoi(argv[3]);
std::cout << "matrix A: " << M << 'x' << N << '\n'
<< "matrix B: " << N << 'x' << K << '\n'
<< "matrix C: " << M << 'x' << K << '\n';
// matrix multiplication using gpu
std::cout << "running gpu matrix multiplication ... ";
auto gbeg = std::chrono::steady_clock::now();
auto gres = gpu(M, N, K);
auto gend = std::chrono::steady_clock::now();
std::cout << "completed with "
<< std::chrono::duration_cast<std::chrono::milliseconds>(gend-gbeg).count()
<< " ms\n";
// matrix multiplication using cpu
std::cout << "running cpu matrix multiplication ... ";
auto cbeg = std::chrono::steady_clock::now();
auto cres = cpu(M, N, K);
auto cend = std::chrono::steady_clock::now();
std::cout << "completed with "
<< std::chrono::duration_cast<std::chrono::milliseconds>(cend-cbeg).count()
<< " ms\n";
// verify the result
int64_t error = 0;
std::cout << "verifying results ... ";
for(int i=0; i<M*K; ++i) {
error += abs(gres[i] - cres[i]);
}
std::cout << "abs-error=" << error << '\n';
return 0;
}
|
