def __call__(self, *inputs, **kwargs): pipeline = Pipeline.current() if pipeline is None: Pipeline._raise_no_current_pipeline("TorchPythonFunction") if self.stream is None: self.stream = torch.cuda.Stream(device=pipeline.device_id) return super(TorchPythonFunction, self).__call__(*inputs, **kwargs)
def __call__(self, *inputs, **kwargs): pipeline = Pipeline.current() if pipeline is None: Pipeline._raise_no_current_pipeline("NumbaFunction") inputs = ops._preprocess_inputs(inputs, self._impl_name, self._device, None) if pipeline is None: Pipeline._raise_pipeline_required("NumbaFunction operator") if (len(inputs) > self._schema.MaxNumInput() or len(inputs) < self._schema.MinNumInput()): raise ValueError(("Operator {} expects from {} to " + "{} inputs, but received {}.").format( type(self).__name__, self._schema.MinNumInput(), self._schema.MaxNumInput(), len(inputs))) for inp in inputs: if not isinstance(inp, _DataNode): raise TypeError(( "Expected inputs of type `DataNode`. Received input of type '{}'. " + "Python Operators do not support Multiple Input Sets." ).format(type(inp).__name__)) op_instance = ops._OperatorInstance(inputs, self, **kwargs) op_instance.spec.AddArg("run_fn", self.run_fn) op_instance.spec.AddArg("setup_fn", self.setup_fn) if self.setup_fn else None op_instance.spec.AddArg("out_types", self.out_types) op_instance.spec.AddArg("in_types", self.in_types) op_instance.spec.AddArg("outs_ndim", self.outs_ndim) op_instance.spec.AddArg("ins_ndim", self.ins_ndim) op_instance.spec.AddArg("device", self.device) op_instance.spec.AddArg("batch_processing", self.batch_processing) if self.device == 'gpu': op_instance.spec.AddArg("blocks", self.blocks) op_instance.spec.AddArg("threads_per_block", self.threads_per_block) if self.num_outputs == 0: t_name = self._impl_name + "_id_" + str(op_instance.id) + "_sink" t = _DataNode(t_name, self._device, op_instance) pipeline.add_sink(t) return outputs = [] for i in range(self.num_outputs): t_name = op_instance._name if self.num_outputs > 1: t_name += "[{}]".format(i) t = _DataNode(t_name, self._device, op_instance) op_instance.spec.AddOutput(t.name, t.device) op_instance.append_output(t) pipeline.add_sink(t) outputs.append(t) return outputs[0] if len(outputs) == 1 else outputs