Add more ops (including all Reduce ops) using the relay frontend (apa…

…che#11)

* [WIP] add more ops. Some fail at the moment

* skip some tests

* Remove duplicate tests for squeeze

python/tvm/relax/frontend/onnx_frontend.py

@@ -587,6 +587,26 @@ def _get_convert_map(opset):
  "Constant": Constant,
  "Sub": Sub,
  "LayerNormalization": relay.frontend.onnx.LayerNormalization,
+ "SkipLayerNormalization": relay.frontend.onnx.SkipLayerNormalization,
+ "EmbedLayerNormalization": relay.frontend.onnx.EmbedLayerNormalization,
+ # defs/reduction
+ "ReduceMax": relay.frontend.onnx.ReduceMax,
+ "ReduceMin": relay.frontend.onnx.ReduceMin,
+ "ReduceSum": relay.frontend.onnx.ReduceSum,
+ "ReduceMean": relay.frontend.onnx.ReduceMean,
+ "ReduceProd": relay.frontend.onnx.ReduceProd,
+ "ReduceLogSumExp": relay.frontend.onnx.ReduceLogSumExp,
+ "ReduceLogSum": relay.frontend.onnx.ReduceLogSum,
+ "ReduceSumSquare": relay.frontend.onnx.ReduceSumSquare,
+ "ReduceL1": relay.frontend.onnx.ReduceL1,
+ "ReduceL2": relay.frontend.onnx.ReduceL2,
+ "Expand": relay.frontend.onnx.Expand,
+ "ConstantOfShape": relay.frontend.onnx.ConstantOfShape,
+ "Slice": relay.frontend.onnx.Slice,
+ "Attention": relay.frontend.onnx.Attention,
+ "Pad": relay.frontend.onnx.Pad,
+ "Split": relay.frontend.onnx.Split,
+ "Tile": relay.frontend.onnx.Tile,
  }
 
 
@@ -780,7 +800,7 @@ def _parse_attr(self, attr_proto):
 
  def _relay_input_adapter(self, inputs):
  """Creates equivalent input Relay vars from the input Relax vars"""
- relay_vars = []
+ relay_vars = onnx_input()
  for relax_var in inputs:
  shape_values = []
  for shape_value in relax_var.struct_info.shape.values:
@@ -896,7 +916,9 @@ def _convert_operator(self, op_name, inputs, attrs, opset):
  if issubclass(convert_class, RelayOnnxOpConverter):
  relay_inputs = self._relay_input_adapter(inputs)
  # The op_function might change the inputs to the relay op. Use a copy of the inputs.
- relay_inputs_copy = [relay_input for relay_input in relay_inputs]
+ relay_inputs_copy = onnx_input()
+ for relay_input in relay_inputs:
+ relay_inputs_copy.append(relay_input)
  # TODO handle params passing
  relay_output = op_function(relay_inputs_copy, attrs, params=[])
  sym = self._relay_output_adapter(inputs, relay_inputs, relay_output)