import hipdnn
import torch


def build_softmax_graph(hipdnn_handle, torch_tensor_x, axis, hipdnn_data_type):
    # Create graph
    graph = hipdnn.pygraph(
        handle=hipdnn_handle,
        io_data_type=hipdnn_data_type,
        intermediate_data_type=hipdnn.data_type.FLOAT,
        compute_data_type=hipdnn.data_type.FLOAT,
        name="softmax",
    )

    # Create hipdnn tensors
    hipdnn_tensor_x = graph.tensor_like(torch_tensor_x)

    # Create softmax op
    hipdnn_tensor_y = graph.softmax(
        input=hipdnn_tensor_x,
        axis=axis,
        name="softmax",
    )
    hipdnn_tensor_y.set_output(True)
    graph.build(hipdnn_handle)

    return (graph, hipdnn_tensor_x, hipdnn_tensor_y)


if __name__ == "__main__":
    # Input dimensions
    n = 2  # Batch size
    c = 3  # Number of channels
    h = 4  # Height
    w = 5  # Width

    # Softmax parameters
    axis = 3  # Axis to apply softmax

    hipdnn_data_type = hipdnn.data_type.FLOAT
    torch_data_type = torch.float32

    torch_tensor_x = torch.rand(n, c, h, w, dtype=torch_data_type, device="cuda")

    hipdnn_handle = hipdnn.create_handle()
    graph, hipdnn_tensor_x, hipdnn_tensor_y = build_softmax_graph(
        hipdnn_handle,
        torch_tensor_x,
        axis,
        hipdnn_data_type,
    )

    torch_tensor_y = torch.empty(hipdnn_tensor_y.get_dim(), dtype=torch_data_type, device="cuda")
    variant_pack = {
        hipdnn_tensor_x: torch_tensor_x.data_ptr(),
        hipdnn_tensor_y: torch_tensor_y.data_ptr(),
    }

    workspace = torch.empty(graph.get_workspace_size(), dtype=torch.uint8, device="cuda")

    graph.exec(variant_pack=variant_pack, workspace=workspace.data_ptr())
    print("Softmax graph execution complete.")