Circumvents/fixes #672: In the simpleCudaGraphs example, keeping alive information the CUDA graph template instantiation relies on until after the instantiation has occurred (which we shouldn't have to do, except for the CUDA driver's unwillingness to copy relevant data)

Also, removed a redundant line break printing command.

Author: eyalroz

examples/modified_cuda_samples/simpleCudaGraphs/simpleCudaGraphs.cu

@@ -191,7 +191,6 @@ void use(const cuda::device_t &device, const cuda::graph::template_t &graph, con
 		cloned_graph_instance.launch(stream_for_graph);
 	}
 	stream_for_graph.synchronize();
-	std::cout << std::endl;
 }
 
 void cudaGraphsManual(
@@ -205,6 +204,12 @@ void cudaGraphsManual(
 	report_mode(graph_construction_mode);
 	report_attempt("construction", graph_construction_mode);
 	double result_h = 0.0;
+	// This ugly hack is due to a problem with the CUDA graph API, and specifically
+	// with kernel launch node insertion: When you insert a node into a template,
+	// the kernel argument values are _not_ copied by the CUDA driver, nor apparently
+	// is the array of pointers to them. So, we need to extend the lifetime of this
+	// information until after the graph template is instantiated. 
+	std::vector<std::vector<void*>> argument_ptr_sequences{};
 
 	using node_kind_t = cuda::graph::node::kind_t;
 	auto graph = cuda::graph::create();
@@ -238,9 +243,9 @@ void cudaGraphsManual(
 			.grid_size(outputVec_d.size())
 			.block_size(THREADS_PER_BLOCK)
 			.build();
-		auto kernel_arg_pointers = cuda::graph::make_kernel_argument_pointers(
-			inputVec_d.data(), outputVec_d.data(), inputVec_d.size(), outputVec_d.size());
-		auto kernel_node_args = cuda::graph::make_launch_primed_kernel(reduce_kernel, launch_config, kernel_arg_pointers);
+		argument_ptr_sequences.emplace_back(cuda::graph::make_kernel_argument_pointers(
+			inputVec_d.data(), outputVec_d.data(), inputVec_d.size(), outputVec_d.size()));
+		auto kernel_node_args = cuda::graph::make_launch_primed_kernel(reduce_kernel, launch_config, argument_ptr_sequences.back());
 		return graph.insert.node<node_kind_t::kernel_launch>(kernel_node_args);
 	}();
 
@@ -261,8 +266,8 @@ void cudaGraphsManual(
 			.grid_size(1)
 			.block_size(THREADS_PER_BLOCK)
 			.build();
-		auto arg_ptrs = cuda::graph::make_kernel_argument_pointers(outputVec_d.data(), result_d.data(), outputVec_d.size());
-		return graph.insert.node<node_kind_t::kernel_launch>(kernel, launch_config, arg_ptrs);
+		argument_ptr_sequences.emplace_back(cuda::graph::make_kernel_argument_pointers(outputVec_d.data(), result_d.data(), outputVec_d.size()));
+		return graph.insert.node<node_kind_t::kernel_launch>(kernel, launch_config, argument_ptr_sequences.back());
 	}();
 
 	graph.insert.edge(reduce_node, reduce_final_node);