Merge branch 'refactorone' of chef.heaplab.deib.polimi.it:nnicolosi/Raptor into refactorone

validation/operations/gen_tests.py

@@ -262,7 +262,7 @@ def conv_grouped_many_groups():
     X = helper.make_tensor_value_info("X", TensorProto.FLOAT, [1, 1024, 2, 2])
     Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [1, 1024, 2, 2])
     W = numpy_helper.from_array(
-        np.random.default_rng(77).uniform(-1, 1, (1024, 64, 1, 1)).astype(np.float32), name="W")
+        np.random.default_rng(77).uniform(-1, 1, (1024, 16, 1, 1)).astype(np.float32), name="W")
     node = helper.make_node("Conv", ["X", "W"], ["Y"],
                             kernel_shape=[1, 1], strides=[1, 1], pads=[0, 0, 0, 0], group=64)
     graph = helper.make_graph([node], "conv_grouped_many_groups", [X], [Y], initializer=[W])
@@ -779,28 +779,6 @@ def matmul_matrix_vector():
     save_model(model, "matmul/matrix_vector", "matmul_matrix_vector.onnx")
 
 
-def matmul_vector_vector_dot():
-    """Vector-vector MatMul producing a scalar output."""
-    A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [1024])
-    Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [])
-    B = numpy_helper.from_array(np.random.default_rng(97).uniform(-1, 1, (1024,)).astype(np.float32), name="B")
-    node = helper.make_node("MatMul", ["A", "B"], ["Y"])
-    graph = helper.make_graph([node], "matmul_vector_vector_dot", [A], [Y], initializer=[B])
-    model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 13)])
-    save_model(model, "matmul/vector_vector_dot", "matmul_vector_vector_dot.onnx")
-
-
-def matmul_batched_4d_broadcast():
-    """Batched 4D MatMul with broadcast across leading dimensions."""
-    A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [2, 1, 3, 4])
-    Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [2, 5, 3, 6])
-    B = numpy_helper.from_array(np.random.default_rng(98).uniform(-1, 1, (1, 5, 4, 6)).astype(np.float32), name="B")
-    node = helper.make_node("MatMul", ["A", "B"], ["Y"])
-    graph = helper.make_graph([node], "matmul_batched_4d_broadcast", [A], [Y], initializer=[B])
-    model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 13)])
-    save_model(model, "matmul/batched_4d_broadcast", "matmul_batched_4d_broadcast.onnx")
-
-
 # ---------------------------------------------------------------------------
 # Pooling tests
 # ---------------------------------------------------------------------------
@@ -1560,17 +1538,6 @@ def add_channel_broadcast_1024():
     save_model(model, "add/channel_broadcast_1024", "add_channel_broadcast_1024.onnx")
 
 
-def add_scalar_runtime():
-    """Elementwise Add with a runtime scalar RHS."""
-    A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [1, 1024, 1, 1])
-    B = helper.make_tensor_value_info("B", TensorProto.FLOAT, [1, 1, 1, 1])
-    Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [1, 1024, 1, 1])
-    node = helper.make_node("Add", ["A", "B"], ["Y"])
-    graph = helper.make_graph([node], "add_scalar_runtime", [A, B], [Y])
-    model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 13)])
-    save_model(model, "add/scalar_runtime", "add_scalar_runtime.onnx")
-
-
 def add_leading_dimension_broadcast():
     """Elementwise Add with trailing-dimension broadcasting."""
     A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [2, 3, 4])
@@ -1635,17 +1602,6 @@ def mul_channel_broadcast_1024():
     save_model(model, "mul/channel_broadcast_1024", "mul_channel_broadcast_1024.onnx")
 
 
-def mul_scalar_runtime():
-    """Elementwise Mul with a runtime scalar RHS."""
-    A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [1, 1024, 1, 1])
-    B = helper.make_tensor_value_info("B", TensorProto.FLOAT, [1, 1, 1, 1])
-    Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [1, 1024, 1, 1])
-    node = helper.make_node("Mul", ["A", "B"], ["Y"])
-    graph = helper.make_graph([node], "mul_scalar_runtime", [A, B], [Y])
-    model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 13)])
-    save_model(model, "mul/scalar_runtime", "mul_scalar_runtime.onnx")
-
-
 def mul_leading_dimension_broadcast():
     """Elementwise Mul with trailing-dimension broadcasting."""
     A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [2, 3, 4])
@@ -1721,17 +1677,6 @@ def div_runtime_scalar_rhs():
     save_model(model, "div/runtime_scalar_rhs", "div_runtime_scalar_rhs.onnx")
 
 
-def div_runtime_scalar_lhs():
-    """Elementwise Div with a scalar constant numerator."""
-    B = helper.make_tensor_value_info("B", TensorProto.FLOAT, [1, 1024, 1, 1])
-    Y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, [1, 1024, 1, 1])
-    A = numpy_helper.from_array(np.asarray([[[[2.0]]]], dtype=np.float32), name="A")
-    node = helper.make_node("Div", ["A", "B"], ["Y"])
-    graph = helper.make_graph([node], "div_runtime_scalar_lhs", [B], [Y], initializer=[A])
-    model = helper.make_model(graph, opset_imports=[helper.make_opsetid("", 13)])
-    save_model(model, "div/runtime_scalar_lhs", "div_runtime_scalar_lhs.onnx")
-
-
 def div_leading_dimension_broadcast():
     """Elementwise Div with trailing-dimension broadcasting."""
     A = helper.make_tensor_value_info("A", TensorProto.FLOAT, [2, 3, 4])
@@ -1812,8 +1757,6 @@ if __name__ == "__main__":
     matmul_huge_1024()
     matmul_vector_matrix()
     matmul_matrix_vector()
-    matmul_vector_vector_dot()
-    matmul_batched_4d_broadcast()
 
     print("\nGenerating Pooling tests:")
     maxpool_basic()
@@ -1899,7 +1842,6 @@ if __name__ == "__main__":
     add_broadcast_row()
     add_after_gemm()
     add_channel_broadcast_1024()
-    add_scalar_runtime()
     add_leading_dimension_broadcast()
 
     print("\nGenerating Mul tests:")
@@ -1907,7 +1849,6 @@ if __name__ == "__main__":
     mul_scalar_constant()
     mul_after_conv()
     mul_channel_broadcast_1024()
-    mul_scalar_runtime()
     mul_leading_dimension_broadcast()
 
     print("\nGenerating Div tests:")
@@ -1916,7 +1857,6 @@ if __name__ == "__main__":
     div_after_gemm()
     div_channel_broadcast_1024()
     div_runtime_scalar_rhs()
-    div_runtime_scalar_lhs()
     div_leading_dimension_broadcast()
 
     print("\nDone.")