cudaErrorNotReady is no longer treated as an error

milakov · May 30, 2013 · 2f7fd14 · 2f7fd14
1 parent 16e2ebc
commit 2f7fd14
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 6 deletions.
diff --git a/nnforge/cuda/network_updater_cuda.cu b/nnforge/cuda/network_updater_cuda.cu
@@ -127,6 +127,8 @@ namespace nnforge
 		{
 			std::vector<testing_result_smart_ptr> res;
 
+			entry_count_updated_in_profile_mode = 0;
+
 			unsigned int min_dropout_layer_id = testing_layer_count + 1;
 			for(std::map<unsigned int, float>::const_iterator it = layer_to_dropout_rate_map.begin(); it != layer_to_dropout_rate_map.end(); ++it)
 				if (it->first < min_dropout_layer_id)
@@ -250,7 +252,7 @@ namespace nnforge
 			if (cuda_config->is_flush_required())
 			{
 				cuda_safe_call(cudaEventRecord(data_processed_event, *command_stream));
-				cudaEventQuery(data_processed_event);
+				cuda_safe_call(cudaEventQuery(data_processed_event));
 			}
 			std::tr1::variate_generator<random_generator, std::tr1::uniform_int<unsigned int> > gen_random_offset(
 				rnd::get_random_generator(),
@@ -416,20 +418,20 @@ namespace nnforge
 						if (((input_entry_id % 16) == 1) && cuda_config->is_flush_required())
 						{
 							cuda_safe_call(cudaEventRecord(data_processed_event, *command_stream));
-							cudaEventQuery(data_processed_event);
+							cuda_safe_call(cudaEventQuery(data_processed_event));
 						}
 					} // for(unsigned int input_entry_id
 
 					if (profile_mode)
-						entry_count_updated_in_profile_mode = entries_available_for_processing_count;
+						entry_count_updated_in_profile_mode += entries_available_for_processing_count;
 
 					for(std::vector<testing_result_smart_ptr>::iterator it = res.begin(); it != res.end(); ++it)
 						(*it)->entry_count += entries_available_for_processing_count;
 
 					if (cuda_config->is_flush_required())
 					{
 						cuda_safe_call(cudaEventRecord(data_processed_event, *command_stream));
-						cudaEventQuery(data_processed_event);
+						cuda_safe_call(cudaEventQuery(data_processed_event));
 					}
 				} // if (entries_available_for_processing_count > 0)
 

diff --git a/nnforge/cuda/neural_network_cuda_exception.h b/nnforge/cuda/neural_network_cuda_exception.h
@@ -32,4 +32,4 @@ namespace nnforge
 	}
 }
 
-#define cuda_safe_call(callstr) {cudaError_t error_code = callstr; if (error_code != cudaSuccess) throw nnforge::cuda::neural_network_cuda_exception(error_code);}
+#define cuda_safe_call(callstr) {cudaError_t error_code = callstr; if ((error_code != cudaSuccess) && (error_code != cudaErrorNotReady)) throw nnforge::cuda::neural_network_cuda_exception(error_code);}
diff --git a/nnforge/neural_network_toolset.cpp b/nnforge/neural_network_toolset.cpp
@@ -930,7 +930,7 @@ namespace nnforge
 			std::cout << (boost::format("%|1$.1f| GFLOPs, %|2$.2f| seconds") % gflops % time_to_complete_seconds) << std::endl;
 		}
 
-		std::cout << data[0]->get_stat() << std::endl;
+		std::cout << data[data.size()-1]->get_stat() << std::endl;
 	}
 
 	void neural_network_toolset::profile_hessian()
-Original file line number
+Diff line change
@@ Expand Up / @@ -32,4 +32,4 @@ namespace nnforge @@
     	}
     }
-    #define cuda_safe_call(callstr) {cudaError_t error_code = callstr; if (error_code != cudaSuccess) throw nnforge::cuda::neural_network_cuda_exception(error_code);}
+    #define cuda_safe_call(callstr) {cudaError_t error_code = callstr; if ((error_code != cudaSuccess) && (error_code != cudaErrorNotReady)) throw nnforge::cuda::neural_network_cuda_exception(error_code);}