add time count

starrygl/sample/count_static.py

@@ -32,7 +32,7 @@ class time_count:
     @staticmethod
     def start_gpu():
         # Uncomment for better breakdown timings
-        torch.cuda.synchronize()
+        #torch.cuda.synchronize()
         start_event = torch.cuda.Event(enable_timing=True)
         end_event = torch.cuda.Event(enable_timing=True)
         start_event.record()

starrygl/sample/data_loader.py

@@ -293,6 +293,7 @@ class DistributedDataLoader:
         t3 = time.time()
         self.result_queue.append((batch_data,dist_nid,dist_eid,edge_feat,node_feat))
         tt.time_memory_fetch += tt.elapsed_event(t1)
+        #print(tt.time_memory_fetch)
         t1 = tt.start_gpu()
         self.submit()
         tt.time_sample_and_build+=tt.elapsed_event(t1)
@@ -326,6 +327,7 @@ class DistributedDataLoader:
                 node_feat,mem = get_node_feature_by_dist(self.graph,self.mailbox,dist_nid, is_local,out_device=self.device)
                 prepare_input(node_feat,edge_feat,mem,mfgs,dist_nid,dist_eid)
                 tt.time_memory_fetch += tt.elapsed_event(t1)
+                #print(tt.time_memory_fetch)
                 if(self.mailbox is not None and self.mailbox.historical_cache is not None):
                         id = batch_data[1][0][0].srcdata['ID']
                         mask = DistIndex(id).is_shared
@@ -363,6 +365,8 @@ class DistributedDataLoader:
                     batch_data[1][0][0].srcdata['his_mem']  = batch_data[1][0][0].srcdata['mem'].clone()
                     batch_data[1][0][0].srcdata['his_ts']  = batch_data[1][0][0].srcdata['mail_ts'].clone()
                 tt.time_memory_fetch+=tt.elapsed_event(t1)
+                
+                #print(tt.time_memory_fetch)
                 #if(self.mailbox is not None and self.mailbox.historical_cache is not None):
                 #    id = batch_data[1][0][0].srcdata['ID']
                 #    mask = DistIndex(id).is_shared