No implementation for Where(9)

Describe the bug
[ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for Where(9) node with name 'Where_16' is raised when updating a boolean tensor.

Urgency
None

System information

• OS Platform and Distribution: Windows 10
• ONNX Runtime installed from: binary
• ONNX Runtime version: 1.9.0
• Python version: 3.8
• Visual Studio version (if applicable): 2019
• GCC/Compiler version (if compiling from source): N/A
• CUDA/cuDNN version: 11.1
• GPU model and memory: GTX 3080 Ti, 12 GB

To Reproduce

import torch
import torch.nn as nn
import onnxruntime as ort


class Model(nn.Module):
    def forward(self, x):
        candidate = torch.rand(x.size(0), device=x.device) < 0.5
        candidate[x == 0] = False
        return candidate


def main():
    dummy_input = torch.randint(256, (256,), dtype=torch.int32)
    model = Model()
    input_names = ['x']
    output_names = ['candidate']
    torch.onnx.export(model, dummy_input, 'm.onnx', verbose=True,
                      input_names=input_names, output_names=output_names,
                      dynamic_axes={
                          'x': {0: 'len'},
                          'candidate': {0: 'len'}
                      }, opset_version=13)
    ort_session = ort.InferenceSession('m.onnx')


if __name__ == '__main__':
    main()

Please also find the onnx model here: m.zip

Start with the provided Python reproduction and the attached m.onnx model, then inspect how the runtime handles the Where(9) node when updating a boolean tensor. Done means the InferenceSession loads the model and executes the boolean update without the NOT_IMPLEMENTED error.