129 lines
3.3 KiB
Text
129 lines
3.3 KiB
Text
#include "graph.hpp"
|
|
#include <CLI11.hpp>
|
|
|
|
int main(int argc, char* argv[]) {
|
|
|
|
CLI::App app{"HeteroTraversal"};
|
|
|
|
unsigned num_threads {1};
|
|
app.add_option("-t,--num_threads", num_threads, "number of threads (default=1)");
|
|
|
|
unsigned num_gpus {1};
|
|
app.add_option("-g,--num_gpus", num_gpus, "number of gpus (default=1)");
|
|
|
|
unsigned num_rounds {1};
|
|
app.add_option("-r,--num_rounds", num_rounds, "number of rounds (default=1)");
|
|
|
|
app.add_option("-n,--num_elements", N, "number of elements in GPU vector (default=1024)");
|
|
|
|
unsigned cuda_ratio {2};
|
|
app.add_option(
|
|
"-c,--cuda_ratio",
|
|
cuda_ratio,
|
|
"cpu/cuda task ratio (the higher, the fewer cuda tasks (default=2)"
|
|
);
|
|
|
|
std::string model = "tf";
|
|
app.add_option("-m,--model", model, "model name tf|tbb|omp (default=tf)")
|
|
->check([] (const std::string& m) {
|
|
if(m != "tf" && m != "tbb" && m != "omp") {
|
|
return "model name should be \"tbb\", \"tf\", or \"omp\"";
|
|
}
|
|
return "";
|
|
});
|
|
|
|
std::string path;
|
|
app.add_option("-f,--file", path, "file path");
|
|
|
|
CLI11_PARSE(app, argc, argv);
|
|
|
|
|
|
cudaDeviceReset();
|
|
|
|
if(path.empty()) {
|
|
|
|
std::cout << "model=" << model << ' '
|
|
<< "num_threads=" << num_threads << ' '
|
|
<< "num_gpus=" << num_gpus << ' '
|
|
<< "num_rounds=" << num_rounds << ' '
|
|
<< std::endl;
|
|
|
|
std::cout << std::setw(12) << "|V|+|E|"
|
|
<< std::setw(12) << "Runtime"
|
|
<< '\n';
|
|
|
|
for(int i=10; i<=20010; i += 500) {
|
|
|
|
Graph graph(i, 4*i, cuda_ratio);
|
|
|
|
//std::ofstream ofs(std::string("graph") + std::to_string(graph.size()) + ".txt");
|
|
//graph.dump(ofs);
|
|
//continue;
|
|
|
|
double runtime {0.0};
|
|
double elapsed;
|
|
|
|
for(unsigned j=0; j<=num_rounds; ++j) {
|
|
|
|
if(model == "tf") {
|
|
elapsed = measure_time_taskflow(graph, num_threads, num_gpus).count();
|
|
}
|
|
else if(model == "tbb") {
|
|
elapsed = measure_time_tbb(graph, num_threads, num_gpus).count();
|
|
}
|
|
else if(model == "omp") {
|
|
elapsed = measure_time_omp(graph, num_threads, num_gpus).count();
|
|
}
|
|
else {
|
|
throw std::runtime_error("unknown model");
|
|
}
|
|
|
|
if(j) {
|
|
runtime += elapsed;
|
|
}
|
|
}
|
|
|
|
std::cout << std::setw(12) << graph.size()
|
|
<< std::setw(12) << runtime / num_rounds / 1e3
|
|
<< std::endl;
|
|
}
|
|
}
|
|
else {
|
|
|
|
Graph graph(path);
|
|
|
|
double runtime {0.0};
|
|
double elapsed;
|
|
|
|
for(unsigned j=0; j<=num_rounds; ++j) {
|
|
|
|
if(model == "tf") {
|
|
elapsed = measure_time_taskflow(graph, num_threads, num_gpus).count();
|
|
}
|
|
else if(model == "tbb") {
|
|
elapsed = measure_time_tbb(graph, num_threads, num_gpus).count();
|
|
}
|
|
else if(model == "omp") {
|
|
elapsed = measure_time_omp(graph, num_threads, num_gpus).count();
|
|
}
|
|
else {
|
|
throw std::runtime_error("unknown model");
|
|
}
|
|
|
|
if(j) {
|
|
runtime += elapsed;
|
|
//std::cout << elapsed/1e3 << ',';
|
|
}
|
|
}
|
|
|
|
//std::cout << std::setw(12) << graph.size()
|
|
// << std::setw(12) << runtime / num_rounds / 1e3
|
|
// << std::endl;
|
|
std::cout << runtime/num_rounds/1e3 << std::endl;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
|
|
|