Some questions about torch operator in iree.

[FlintWangacc]

Hi guys,
I am curious about the how torch operator is handled in iree.
For example, I have a simple matmul model

class MatMulModel(nn.Module):
    def __init__(self, input_size: int, hidden_size: int, output_size: int):
        super().__init__()
        self.weight1 = nn.Parameter(torch.randn(input_size, hidden_size,
                                    dtype=torch.float32))
        self.weight2 = nn.Parameter(torch.randn(hidden_size, output_size,
                                    dtype=torch.float32))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # Two matrix multiplications
        x = torch.matmul(x, self.weight1)
        x = torch.relu(x)  # Add non-linearity
        x = torch.matmul(x, self.weight2)
        return x

I didn't tell iree how to handle torch.matmul. After some investigation, I think the torch.matmul will become a hal command in iree bytecode sequence. But I didn't tell iree how to generate code for torch.matmul. Which pass add the mlir source code for this MatMulModel ?