Graph Construction by Using DataFlow APIs

Construct FlowGraph.

Use FlowNode and FunctionPp to construct a graph. The following is an example:

// Construct FlowGraph of DataFlow.
dflow::FlowGraph flow_graph("flow_graph");

// Construct the input node.
auto data0 = dflow::FlowData("Data0", 0);
auto data1 = dflow::FlowData("Data1", 1);

auto node0 = dflow::FlowNode("node0", 2, 1).SetInput(0, data0).SetInput(1, data1);
// function_pp
auto pp0 = dflow::FunctionPp("func_pp0")
               .SetCompileConfig("./add_func.json")
               .SetInitParam("out_type", ge::DT_UINT32);
node0.AddPp(&pp0);

Run the graph.

You can use either of the following modes:

Session run mode. The following is an example:

// Create a session.
ge::Session* session1 = new Session(options);
assert(session1 != NULL);
ret = session1->AddGraph(graphId, graph);
if (!ret) { return; }
// Run the graph.
ret = session1->RunGraph(graphId, input, output);
if (!ret) { return; }
GEFinalize();
return;

DataFlow mode. The following is an example:

// Execute the graph using the asynchronous DataFlow API.
ge::DataFlowInfo dataFlowInfo;
dataFlowInfo.SetStartTime(0);
dataFlowInfo.SetEndTime(5);
std::vector<ge::Tensor> inputsData = {inputTensor, inputTensor};
geRet = session->FeedDataFlowGraph(0, inputsData, dataFlowInfo, 3000);
if (geRet != ge::SUCCESS)
{
	std::cout << "TEST====Feed input data failed============." << std::endl;
	ge::GEFinalize();
	return -1;
}
// Obtain outputs.
std::vector<ge::Tensor> outputsData;
geRet = session->FetchDataFlowGraph(0, outputsData, dataFlowInfo, 3000);
if (geRet != ge::SUCCESS)
{
	std::cout << "TEST====Fetch output data failed." << std::endl;
	ge::GEFinalize();
	return -1;
}

Parent topic: UDF Development