#include <taskflow/taskflow.hpp>
std::mutex cout_mutex;
void format_str(std::string const& str) {
std::lock_guard<std::mutex> lock(cout_mutex);
std::cout << str << std::endl;
}
std::string createTaskName(int i, int j) {
return "Task" + std::to_string(i) + "_" + std::to_string(j);
}
int main() {
int m1 = 1000, n = 1000, m2 = 1000; // m1行n列 * n行m2列
std::vector<std::vector<int>> nums1(m1, std::vector<int>(n, 0));
std::vector<std::vector<int>> nums2(n, std::vector<int>(m2, 0));
std::vector<std::vector<int>> ans1(m1, std::vector<int>(m2, 0)); // 结果矩阵
std::vector<std::vector<int>> ans2(m1, std::vector<int>(m2, 0)); // 结果矩阵
// 随机往两个数据中填入数据
for(int i = 0; i < m1; i++) {
for(int j = 0; j < n; j++) {
nums1[i][j] = 1+i*j;
}
}
for(int i = 0; i < n; i++) {
for(int j = 0; j < m2; j++) {
nums2[i][j] = 2*i*j;
}
}
// // 统计矩阵乘法的耗时
auto start = std::chrono::high_resolution_clock::now();
// // 单线程的矩阵乘法
for(int i = 0; i < m1; i++) {
for(int j = 0; j < m2; j++) {
for(int k = 0; k < n; k++) {
ans1[i][j] += nums1[i][k] * nums2[k][j];
}
}
}
auto end = std::chrono::high_resolution_clock::now();
std::chrono::duration<double, std::milli> duration = end - start;
format_str("Single-thread matrix multiplication takes " + std::to_string(duration.count()) + "ms");
// 使用Taskflow 对上述矩阵乘法进行并行加速
tf::Executor executor;
// 统计矩阵乘法的耗时
start = std::chrono::high_resolution_clock::now();
for(int i = 0; i < m1; i++) {
executor.silent_async([&,i](){
for(int j = 0; j < m2; j++) {
for(int k = 0; k < n; k++) {
ans2[i][j] += nums1[i][k] * nums2[k][j];
}
}
});
}
executor.wait_for_all(); // 等待所有任务完成
end = std::chrono::high_resolution_clock::now();
duration = end - start;
format_str("Multi-thread matrix multiplication takes " + std::to_string(duration.count()) + "ms");
// 检查结果是否正确
for(int i = 0; i < m1; i++) {
for(int j = 0; j < m2; j++) {
assert(ans1[i][j] == ans2[i][j]);
}
}
}
Taskflow应用:矩阵乘法
于 2024-04-01 10:48:43 首次发布