BNNSGraphContextExecute(_:_:_:_:_:_:)

Declaration

func BNNSGraphContextExecute(_ context: bnns_graph_context_t, _ function: UnsafePointer<CChar>?, _ argument_count: Int, _ arguments: UnsafeMutablePointer<bnns_graph_argument_t>, _ workspace_size: Int, _ workspace: UnsafeMutablePointer<CChar>?) -> Int32

Parameters

• context:

  The graph context. You may only use this context on a single thread at a time.

• function:

  The function. Specify as nil if the graph only contains one function.

• argument_count:

  The number of elements in the arguments array.

• arguments:

  An array of bnns_graph_argument_t objects that supply the input data and the outputs. Use the Bnnsgraphgetargumentposition(_:_:_:) to derive the argument positions for this parameter. If you supply unallocated output arguments, this function allocates the required memory and reuses that memory for future executions.

• workspace_size:

  The size of the workspace. Call Bnnsgraphcontextgetworkspacesize(_:_:) to get the required size or pass 0 and a null workspace parameter to specify that this function allocates its own workspace.

• workspace:

  The temporary workspace that the function uses during execution. Call Bnnsgraphcontextgetworkspacesize(_:_:) to get the required size or pass nil and a zero workspace_size parameter to specify that this function allocates its own workspace.

Return Value

0 on success, nonzero on failure.

Discussion

If the underlying model contains dynamic shaped inputs or outputs, you must set these prior to calling this routine with either BNNSGraphContextSetDynamicShapes(_:_:_:_:) or BNNSGraphContextSetBatchSize(_:_:_:). Don’t modify these shapes during execution.

This function doesn’t perform memory allocation if your code meets the following conditions:

• The workspace isn’t nil.

• The graph doesn’t contain tensors that BNNS couldn’t bound prior to calling this function.

• The arguments array doesn’t contain any new shape information.

• The code provides all output storage.

The following code executes the graph with the BNNSTensor structures inputTensor and outputTensor as the input and output arguments, respectively:

withUnsafeMutablePointer(to: &inputTensor) { input in
    withUnsafeMutablePointer(to: &outputTensor) { output in
        
        var inputArgument = bnns_graph_argument_t()
        inputArgument.tensor = input
        
        var outputArgument = bnns_graph_argument_t()
        outputArgument.tensor = output
        
        var arguments = [outputArgument, inputArgument]
        
        BNNSGraphContextSetArgumentType(context, BNNSGraphArgumentTypeTensor)
        
        BNNSGraphContextExecute(context, nil, 2, &arguments, 0, nil)
    }
}

See Also

Executing a graph

• BNNSGraphArgumentType