Fusion fusion;
    FusionGuard fg(&fusion);

    int y = 6, z = 8;
    DataType dtype = DataType::Float;
    auto options = at::TensorOptions().dtype(at::kFloat).device(at::kCUDA, 0);

    auto tv0 = makeDummyTensor(1);
    auto tv1 = makeDummyTensor(2);
    fusion.addInput(tv0);
    fusion.addInput(tv1);

    auto tv2 = broadcast(tv0, {true, false});
    auto tv3 = add(tv2, new Float(1.0));
    auto tv4 = add(tv3, tv1);

    fusion.addOutput(tv4);

    tv4->merge(0);

    tv4->split(0, 128);
    tv4->split(0, 4);

    tv2->computeAt(tv4, 1);

    fusion.printKernel();

    torch::jit::fuser::cuda::FusionExecutor fe;

    at::Tensor t0 = at::randn({z}, options).fill_(1.0);
    at::Tensor t1 = at::randn({y, z}, options).fill_(3.0);

    fe.compileFusion(&fusion);
    auto outputs = fe.runFusion({t0, t1});

    auto t3 = t0.add(1.0);
    auto t4 = t3.add(t1);

    TORCH_CHECK(t4.allclose(outputs[0]));