import hipdnn
import torch


def build_transpose_graph(hipdnn_handle, torch_tensor_x, 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="transpose",
    )

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

    # Create transpose op
    # nhwc->nchw[0, 1, 2, 3] or nchw->nhwc[0, 2, 3, 1]
    hipdnn_tensor_y = graph.transpose(
        input=hipdnn_tensor_x,
        permutation=[0, 1, 2, 3],
        name="transpose",
    )
    hipdnn_tensor_y.set_output(True)
    graph.build(hipdnn_handle)

    return (graph, hipdnn_tensor_x, hipdnn_tensor_y)


if __name__ == "__main__":
    # Input dimensions
    batch, channels, height, width = 2, 3, 4, 5

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

    torch_tensor_x = torch.rand(
        batch, channels, height, width, dtype=torch_data_type, device="cuda"
    ).to(memory_format=torch.channels_last)
    hipdnn_handle = hipdnn.create_handle()

    graph, hipdnn_tensor_x, hipdnn_tensor_y = build_transpose_graph(
        hipdnn_handle, torch_tensor_x, 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("Transpose graph execution complete.")