MatMul fusion failed at scalar input

[yz6790]

Describe the bug
When fusing MatMul and Mul for an input of rank=0, there is an error.

Perhaps it is because Mul outputs a result of rank=1, and the input's rank was not raised to 1 in fusion? Since MatMul does not support inputs of rank=0, this error happens?
Urgency
None.

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): macOS Mojave
• ONNX Runtime installed from (source or binary): binary
• ONNX Runtime version: 1.10.0
• Python version: 3.9.7
• Visual Studio version (if applicable): n/a
• GCC/Compiler version (if compiling from source): n/a
• CUDA/cuDNN version: n/a
• GPU model and memory: n/a
• PyTorch version: 1.10.2

To Reproduce

• Describe steps/code to reproduce the behavior.

import torch

class Model(torch.nn.Module):
    @torch.no_grad()
    def forward(self, i1):
        y = torch.mul(i1, torch.ones(1,dtype=torch.float32))
        o2 = torch.matmul(y, torch.ones((1,64),dtype=torch.float32))

        return o2


model = Model()
inputs = (torch.randn((), dtype=torch.float32))
output_names = ["o1"]
torch.onnx.export(model, inputs, "1.onnx", verbose=False,
                  input_names=["i1"], output_names=output_names, opset_version=14)

• Attach the ONNX model to the issue (where applicable) to expedite investigation.
  MatMul.onnx .zip

Expected behavior
MatMul will take the input as rank=1 after Mul.
Screenshots

[ytaous]

thanks for reporting the issue, let me take a look

[ytaous]

x1 = np.array((1.0)).astype(np.float32)
session = ort.InferenceSession("./1.onnx", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
res = session.run(None,{"i1":x1})