refactor: add concrete.onnx mod with previous conv2d

the module is meant to contain machine learning operations as defined by
ONNX

concrete/numpy/__init__.py

@@ -10,6 +10,6 @@ from .compilation import (
     EncryptionStatus,
     compiler,
 )
-from .extensions import LookupTable, conv2d
+from .extensions import LookupTable
 from .mlir.utils import MAXIMUM_BIT_WIDTH
 from .representation import Graph

concrete/numpy/extensions/__init__.py

@@ -2,5 +2,4 @@
 Provide additional features that are not present in numpy.
 """
 
-from .convolution import conv2d
 from .table import LookupTable

concrete/onnx/__init__.py

@@ -0,0 +1,5 @@
+"""
+Implement machine learning operations as specified by ONNX.
+"""
+
+from .convolution import conv2d

concrete/onnx/convolution.py

@@ -8,9 +8,9 @@ from typing import List, Optional, Tuple, Union
 import numpy as np
 import torch
 
-from ..representation import Node
-from ..tracing import Tracer
-from ..values import EncryptedTensor
+from ..numpy.representation import Node
+from ..numpy.tracing import Tracer
+from ..numpy.values import EncryptedTensor
 
 SUPPORTED_AUTO_PAD = {
     "NOTSET",

tests/execution/test_convolution.py

@@ -6,6 +6,7 @@ import numpy as np
 import pytest
 
 import concrete.numpy as cnp
+import concrete.onnx as connx
 
 
 @pytest.mark.parametrize(
@@ -56,7 +57,7 @@ def test_conv2d(input_shape, weight_shape, strides, dilations, has_bias, helpers
 
     @cnp.compiler({"x": "encrypted"}, configuration=configuration)
     def function(x):
-        return cnp.conv2d(x, weight, bias, strides=strides, dilations=dilations)
+        return connx.conv2d(x, weight, bias, strides=strides, dilations=dilations)
 
     inputset = [np.random.randint(0, 4, size=input_shape) for i in range(100)]
     circuit = function.compile(inputset)
@@ -177,7 +178,7 @@ def test_bad_conv2d_tracing(
 
     @cnp.compiler({"x": "encrypted"}, configuration=configuration)
     def function(x):
-        return cnp.conv2d(x, weight, bias, pads, strides, dilations, auto_pad)
+        return connx.conv2d(x, weight, bias, pads, strides, dilations, auto_pad)
 
     inputset = [np.random.randint(0, 4, size=input_shape) for i in range(100)]
     with pytest.raises(expected_error) as excinfo:
@@ -194,13 +195,13 @@ def test_bad_conv2d_evaluation():
     x = np.random.randint(0, 4, size=(1, 1, 4, 4))
 
     with pytest.raises(ValueError) as excinfo:
-        cnp.conv2d(x, "abc")
+        connx.conv2d(x, "abc")
 
     assert str(excinfo.value) == "Weight should be of type np.ndarray for evaluation"
 
     weight = np.random.randint(0, 4, size=(1, 1, 2, 2))
 
     with pytest.raises(ValueError) as excinfo:
-        cnp.conv2d(x, weight, "abc")
+        connx.conv2d(x, weight, "abc")
 
     assert str(excinfo.value) == "Bias should be of type np.ndarray for evaluation"

tests/mlir/test_graph_converter.py

@@ -6,6 +6,7 @@ import numpy as np
 import pytest
 
 import concrete.numpy as cnp
+import concrete.onnx as connx
 
 # pylint: disable=line-too-long
 
@@ -91,7 +92,7 @@ return %2
             """,  # noqa: E501
         ),
         pytest.param(
-            lambda x, w: cnp.conv2d(x, w),
+            lambda x, w: connx.conv2d(x, w),
             {"x": "encrypted", "w": "encrypted"},
             [
                 (