#!/usr/bin/env python3
"""Smallest official Gluon boundary: Python launcher to one GPU kernel."""

import torch
from triton.experimental import gluon
from triton.experimental.gluon import language as gl


@gluon.jit
def copy_scalar_kernel(input_pointer, output_pointer):
    value = gl.load(input_pointer)
    gl.store(output_pointer, value)


def run() -> torch.Tensor:
    source = torch.tensor([1.0], device="cuda")
    destination = torch.empty_like(source)
    copy_scalar_kernel[(1,)](source, destination)
    return destination


if __name__ == "__main__":
    print(run())