Enabling profiler with empty Optional causes segmentation fault

[amancini-N]

Describe the issue

When enabling profiling over a model which includes empty Optional, a segfault is caused.

To reproduce

Following python script creates a model which reproduces the error:

"""Following module creates a relu model using Optional, OptionalHasElement, OptionalGetElement, If operators.
The model is saved as 'relu_with_optional.onnx' in the current directory.
It is then loaded and run using onnxruntime. The input is -1 and the expected output is 0.
Enabling profiling causes a segfault"""

import onnx
import onnxruntime as ort
import numpy as np

def make_optional_tensor_value_info(name, data_type, shape):
    value_info_proto = onnx.ValueInfoProto()
    value_info_proto.name = name

    tensor_type_proto = onnx.helper.make_tensor_type_proto(data_type, shape, None)
    optional_type_proto = onnx.helper.make_optional_type_proto(tensor_type_proto)
    value_info_proto.type.CopyFrom(optional_type_proto)

    return value_info_proto

def create_model():
    input_info_proto = onnx.helper.make_tensor_value_info("input", onnx.TensorProto.INT32, shape=[1])
    output_info_proto = onnx.helper.make_tensor_value_info("output", onnx.TensorProto.INT32, shape=[1])

    zero = onnx.helper.make_node("Constant", [], ["zero"], value=onnx.helper.make_tensor(name="zero", data_type=onnx.TensorProto.INT32, dims=[1], vals=[0]))
    is_input_positive = onnx.helper.make_node("Greater", ["input", "zero"], ["is_input_positive"])

    # Make a graph, return optional of input
    opt_wrap = onnx.helper.make_node("Optional", ["input"], ["opt_wrap"])
    then_graph = onnx.helper.make_graph(
        [opt_wrap],
        "then",
        [],
        [make_optional_tensor_value_info("opt_wrap", onnx.TensorProto.INT32, shape=[1])],
        [],
    )

    # Make a graph, return empty optional
    empty_opt_wrap = onnx.helper.make_node("Optional", [], ["empty_opt"], type=onnx.helper.make_tensor_type_proto( onnx.TensorProto.INT32, [1]))
    else_graph = onnx.helper.make_graph(
        [empty_opt_wrap],
        "else",
        [],
        [make_optional_tensor_value_info("empty_opt", onnx.TensorProto.INT32, shape=[1])],
        [],
    )

    # Make If node
    if_node = onnx.helper.make_node("If", ["is_input_positive"], ["intermediate"], then_branch=then_graph, else_branch=else_graph)

    # If condition: OptionalGet
    opt_has = onnx.helper.make_node("OptionalHasElement", ["intermediate"], ["has"])

    # If then branch: OptionalGet
    opt_get = onnx.helper.make_node("OptionalGetElement", ["intermediate"], ["get"])
    then_graph = onnx.helper.make_graph(
        [opt_get],
        "then",
        [],
        [onnx.helper.make_tensor_value_info("get", onnx.TensorProto.INT32, shape=[1])],
        [],
    )

    # If else branch: Constant
    const = onnx.helper.make_node("Constant", [], ["const"], value=onnx.helper.make_tensor(name="const", data_type=onnx.TensorProto.INT32, dims=[1], vals=[0]))
    else_graph = onnx.helper.make_graph(
        [const],
        "else",
        [],
        [onnx.helper.make_tensor_value_info("const", onnx.TensorProto.INT32, shape=[1])],
        [],
    )

    # Make If node
    if_node_2 = onnx.helper.make_node("If", ["has"], ["output"], then_branch=then_graph, else_branch=else_graph)

    # Make main graph
    graph = onnx.helper.make_graph(
        [zero, is_input_positive, if_node, opt_has, if_node_2],
        "main",
        [input_info_proto],
        [output_info_proto],
        [],
    )

    # Make model
    model = onnx.helper.make_model(
        graph,
        opset_imports=[onnx.helper.make_opsetid("", 17)],
        doc_string="Relu implementation using Optional, OptionalHasElement, OptionalGetElement, If operators",
    )

    # Save model
    onnx.save(model, "relu_with_optional.onnx")

def run_model():
    sess_options = ort.SessionOptions()
    sess_options.enable_profiling = True
    sess = ort.InferenceSession("relu_with_optional.onnx", sess_options)

    input = np.array([-1], dtype=np.int32)
    output = sess.run(["output"], {"input": input})

    print(output)

create_model()
run_model()

Urgency

No response

Platform

Linux

OS Version

Ubuntu 20.04

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

09c98433e7e6b352c7b878660267d2635c7bfcf6

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response