updated hetero_traversal benchmark

twhuang-utah · twhuang-utah · commit d4b45363b03c · 2020-03-30T11:01:55.000-06:00
diff --git a/benchmarks/hetero_traversal/main.cu b/benchmarks/hetero_traversal/main.cu
@@ -46,14 +46,14 @@ int main(int argc, char* argv[]) {
 
   for(int i=10; i<=10000; i += 500) {
 
-    double runtime {0.0};
-
     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};
+    
     for(unsigned j=0; j<num_rounds; ++j) {
       if(model == "tf") {
         runtime += measure_time_taskflow(graph, num_threads, num_gpus).count();
diff --git a/benchmarks/hetero_traversal/omp.cu b/benchmarks/hetero_traversal/omp.cu
@@ -89,9 +89,9 @@ void omp(const Graph& g, unsigned num_cpus, unsigned num_gpus) {
   int* gx = nullptr;
   int* gy = nullptr;
   int* gz = nullptr;
-  TF_CHECK_CUDA(cudaMalloc(&gx, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gy, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gz, N*sizeof(int)), "failed at cudaMalloc");
+  TF_CHECK_CUDA(cudaMallocManaged(&gx, N*sizeof(int)), "failed at cudaMallocManaged");
+  TF_CHECK_CUDA(cudaMallocManaged(&gy, N*sizeof(int)), "failed at cudaMallocManaged");
+  TF_CHECK_CUDA(cudaMallocManaged(&gz, N*sizeof(int)), "failed at cudaMallocManaged");
   
   for(size_t l=0; l<levellist.size(); ++l) {
     #pragma omp parallel for num_threads(num_cpus + num_gpus)
diff --git a/benchmarks/hetero_traversal/taskflow.cu b/benchmarks/hetero_traversal/taskflow.cu
@@ -12,9 +12,9 @@ void taskflow(const Graph& g, unsigned num_cpus, unsigned num_gpus) {
   int* gx = nullptr;
   int* gy = nullptr;
   int* gz = nullptr;
-  TF_CHECK_CUDA(cudaMalloc(&gx, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gy, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gz, N*sizeof(int)), "failed at cudaMalloc");
+  TF_CHECK_CUDA(cudaMallocManaged(&gx, N*sizeof(int)), "failed at cudaMalloc");
+  TF_CHECK_CUDA(cudaMallocManaged(&gy, N*sizeof(int)), "failed at cudaMalloc");
+  TF_CHECK_CUDA(cudaMallocManaged(&gz, N*sizeof(int)), "failed at cudaMalloc");
   
   tf::Taskflow taskflow;
   tf::Executor executor(num_cpus, num_gpus);
diff --git a/benchmarks/hetero_traversal/tbb.cu b/benchmarks/hetero_traversal/tbb.cu
@@ -54,9 +54,9 @@ void TBB(const Graph& g, unsigned num_cpus, unsigned num_gpus) {
   int* gx = nullptr;
   int* gy = nullptr;
   int* gz = nullptr;
-  TF_CHECK_CUDA(cudaMalloc(&gx, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gy, N*sizeof(int)), "failed at cudaMalloc");
-  TF_CHECK_CUDA(cudaMalloc(&gz, N*sizeof(int)), "failed at cudaMalloc");
+  TF_CHECK_CUDA(cudaMallocManaged(&gx, N*sizeof(int)), "failed at cudaMallocManaged");
+  TF_CHECK_CUDA(cudaMallocManaged(&gy, N*sizeof(int)), "failed at cudaMallocManaged");
+  TF_CHECK_CUDA(cudaMallocManaged(&gz, N*sizeof(int)), "failed at cudaMallocManaged");
 
   std::vector<std::unique_ptr<continue_node<continue_msg>>> tasks(g.num_nodes);
   std::vector<size_t> indegree(g.num_nodes, 0);
