NXP backend: Remove ONNX related and unused code.

Author: MartinPavella

backends/nxp/backend/ir/conversion_config.py

@@ -1,4 +1,4 @@
-# Copyright 2024 NXP
+# Copyright 2024-2025 NXP
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
@@ -14,7 +14,6 @@ def __init__(self, args: dict | None = None):
         :param args: Optional dictionary with conversion arguments. Unknown arguments are ignored.
         """
         self.keep_io_format: bool = False
-        self.skip_shape_inference: bool = False
         self.allow_inputs_stripping: bool = True
         self.qdq_aware_conversion: bool = True
         self.symbolic_dimensions_mapping: dict[str, int] | None = None
@@ -46,15 +45,6 @@ def __repr__(self):
         return "ConversionConfig[" + ", ".join(attrs) + "]"
 
 
-class SkipShapeInferenceConfig(ConversionConfig):
-
-    def __init__(self):
-        """
-        Conversion config shortcut with disabled shape inference.
-        """
-        super().__init__({"skip_shape_inference": True})
-
-
 class QDQAwareConfig(ConversionConfig):
 
     def __init__(self):

backends/nxp/backend/ir/converter/builder/model_builder.py

@@ -1,6 +1,6 @@
 #
 # Copyright 2023 Martin Pavella
-# Copyright 2023-2024 NXP
+# Copyright 2023-2025 NXP
 #
 # License: MIT
 # See the LICENSE_MIT for more details.
@@ -795,29 +795,8 @@ def _remove_tensor_with_name(self, name):
 
     def append_new_tensor(self, t_tensor: tflite_model.Tensor, overwrite: bool = False):
         """Append the TFLite tensor 't_tensor' to the 'SubGraph.tensors' and register it."""
-
-        if t_tensor.name in self._tensor_name_map.keys():
-            """Tensor has already been added. Sometimes however, ONNX models
-            will have tensors in their 'inputs' or 'outputs', which don't
-            belong there and are in fact static. I this case we need to
-            overwrite the existing tensors."""
-
-            if overwrite:
-                self._remove_tensor_with_name(t_tensor.name)
-
-                # If the tenor previously appeared in ONNX 'inputs' or 'outputs',
-                # the old version MUST be removed from there.
-                self._remove_input_with_name(t_tensor.name)
-                self._remove_output_with_name(t_tensor.name)
-
-                self.get_tensors().append(t_tensor)
-                self._tensor_name_map[t_tensor.name] = t_tensor
-            else:
-                logger.w(f"Tensor '{t_tensor.name}' is already in the tensors!")
-
-        else:
-            self._tensor_name_map[t_tensor.name] = t_tensor
-            self.get_tensors().append(t_tensor)
+        self._tensor_name_map[t_tensor.name] = t_tensor
+        self.get_tensors().append(t_tensor)
 
     def append_new_buffer(self, buffer: tflite_model.Buffer):
         """Append the 'buffer' to the 'model.buffers'."""

backends/nxp/backend/ir/converter/conversion/common.py

@@ -1,6 +1,6 @@
 #
 # Copyright 2023 Martin Pavella
-# Copyright 2023-2024 NXP
+# Copyright 2023-2025 NXP
 #
 # License: MIT
 # See the LICENSE_MIT for more details.
@@ -12,7 +12,7 @@
 'conversion/builtin/' directory.
 """
 
-from typing import Any, List, MutableSequence, Optional
+from typing import List, MutableSequence, Optional
 
 import executorch.backends.nxp.backend.ir.logger as logger
 from executorch.backends.nxp.backend.ir.tflite_generator import tflite_model
@@ -22,28 +22,8 @@
     max_pool_2d_options,
     transpose_conv_options,
 )
-from torch.fx import Node
-
-
-def exactly_one_is_none(obj1: Optional, obj2: Optional) -> bool:
-    """Determine if exactly 1 of the arguments is None, or not."""
-    return (obj1 is None and obj2 is not None) or (obj1 is not None and obj2 is None)
-
-
-def contains_duplicates(list_to_check: List[Any]) -> bool:
-    """Determine if given list has duplicate elements or not."""
-    return len(list_to_check) != len(set(list_to_check))
-
-
-def clamp(val: int, start: int, end: int) -> int:
-    """Clamp an int value between start and end (inclusive) and return it."""
-    if val < start:
-        return start
-
-    elif val > end:
-        return end
 
-    return val
+from torch.fx import Node
 
 
 def try_get_input(t_op: tflite_model.Operator, idx: int) -> tflite_model.Tensor | None:
@@ -62,11 +42,6 @@ def try_get_input(t_op: tflite_model.Operator, idx: int) -> tflite_model.Tensor
 
     tensor = t_op.tmp_inputs[idx]
 
-    if tensor.name == "":
-        # ONNX allows the name "" for optional tensors. It indicates that the tensor should be ignored, and a default
-        #  value should be used. Just like if the tensor was omitted altogether.
-        return None
-
     return tensor
 
 
@@ -188,32 +163,6 @@ def node_uses_shape_broadcasting(node: Node) -> bool:
     )
 
 
-def uses_multiple_input_types(t_op: tflite_model.Operator) -> bool:
-    """Determine if the input tensors of given TFLite operator use different data types or not.
-
-    :param t_op: TFLite operator with 'tmp_inputs' initialized.
-    :return: True, if any two input tensors have a different data type.
-             False, if all input tensors use the same data type.
-    """
-
-    if t_op.tmp_inputs is None:
-        logger.e(
-            logger.Code.INTERNAL_ERROR,
-            "common.uses_multiple_input_types(): 'tmp_inputs' are None!",
-        )
-
-    if len(t_op.tmp_inputs) == 0:
-        logger.e(
-            logger.Code.INTERNAL_ERROR,
-            "common.uses_multiple_input_types(): Operator has no inputs!",
-        )
-
-    first_input_type = t_op.tmp_inputs[0].type
-    return any(
-        input_tensor.type != first_input_type for input_tensor in t_op.tmp_inputs[1:]
-    )
-
-
 class OpsList:
     """
     Holder of TFLite operator (middle_op) that can be prefixed (pre_ops) of suffixed (post_ops)