def test_cmp_type():
    for op, ref in ((relay.greater, np.greater),
               (relay.greater_equal, np.greater_equal),
               (relay.less, np.less),
               (relay.less_equal, np.less_equal),
               (relay.equal, np.equal),
               (relay.not_equal, np.not_equal)):
        x = relay.var("x", relay.TensorType((10, 4), "float32"))
        y = relay.var("y", relay.TensorType((5, 10, 1), "float32"))
        z = op(x, y)
        z.astext()
        zz = relay.ir_pass.infer_type(z)
        assert zz.checked_type == relay.TensorType((5, 10, 4), "bool")

        if ref is not None:
            x_shape = (10, 4)
            y_shape = (5, 10, 1)
            t1 = relay.TensorType(x_shape)
            t2 = relay.TensorType(y_shape)
            x = relay.var("x", t1)
            y = relay.var("y", t2)
            z = op(x, y)
            x_data = np.random.rand(*x_shape).astype(t1.dtype)
            y_data = np.random.rand(*y_shape).astype(t2.dtype)
            ref_res = ref(x_data, y_data)
            func = relay.Function([x, y], z)

            for target, ctx in ctx_list():
                intrp = relay.create_executor("graph", ctx=ctx, target=target)
                op_res = intrp.evaluate(func)(x_data, y_data)
                tvm.testing.assert_allclose(op_res.asnumpy(), ref_res)

def test_mul_param():
    x = relay.var('x', shape=(10, 10))
    y = relay.var('y', shape=(1, 10))
    func = relay.Function([x, y], relay.multiply(x, y))
    x_data = np.random.rand(10, 10).astype('float32')
    y_data = np.random.rand(1, 10).astype('float32')
    check_eval(func, [x_data, y_data], x_data * y_data)

def dense_add_bias(data, weight=None, bias=None, units=None, **kwargs):
    """Wrapper of dense which automatically creates weights if not given.

    Parameters
    ----------
    data : relay.Expr
        The input expression.

    weight : relay.Expr
        The weight to conv2d.

    bias : relay.Expr
        The bias.

    kwargs : dict
        Additional arguments.

    Returns
    -------
    result : relay.Expr
        The result.
    """
    name = kwargs.get("name")
    kwargs.pop("name")
    if not weight:
        weight = relay.var(name + "_weight")
    if not bias:
        bias = relay.var(name + "_bias")
    data = relay.nn.dense(data, weight, units, **kwargs)
    data = relay.nn.bias_add(data, bias, axis=-1)
    return data

def test_annotate_none():
    ctx1 = tvm.context(1)
    ctx2 = tvm.context(2)
    x = relay.var("x", shape=(3,))
    y = relay.var("y", shape=(3,))
    z = relay.var("z", shape=(3,))

    def annotated():
        add = relay.add(x, y)
        sub = relay.subtract(add, z)
        func = relay.Function([x, y, z], sub)
        func = relay.ir_pass.infer_type(func)
        func = relay.ir_pass.rewrite_annotated_ops(func,
                                                   ctx1.device_type)
        return func

    def expected():
        add = relay.add(x, y)
        sub = relay.subtract(add, z)
        func = relay.Function([x, y, z], sub)
        return func

    annotated_func = relay.ir_pass.infer_type(annotated())
    expected_func = relay.ir_pass.infer_type(expected())
    assert relay.ir_pass.alpha_equal(annotated_func, expected_func)

def test_conv():
    batch_size = 1
    input_channel = 3
    h = 224
    w = 224
    output_channel = 64
    kh = 7
    kw = 7
    h_padding = 1
    w_padding = 1
    oh = h + h_padding * 2 - kh + 1
    ow = w + w_padding * 2 - kw + 1
    dshape = (batch_size, input_channel, h, w)
    weight = relay.var("weight", shape=(output_channel, input_channel, kh, kw))
    data = relay.var("data", shape=dshape)
    conv2d = relay.nn.conv2d(
        data,
        weight,
        channels=output_channel,
        kernel_size=(kh, kw),
        padding=(1, 1))
    func = relay.Function([data, weight],
                            relay.Tuple(tvm.convert([conv2d])))
    func = relay.ir_pass.infer_type(func)
    compute_count = relay.ir_pass.get_total_mac_number(func)
    expect_count = batch_size * input_channel * oh * ow * output_channel * kh * kw
    assert compute_count == expect_count

def test_tuple():
    shape = (10, 10)
    dtype = 'float32'
    t = relay.TensorType(shape, dtype)
    x = relay.var("x", t)
    y = relay.var("y", t)
    z = relay.var("z", t)
    tup = relay.Var("tup")
    func = relay.Function([x, y, z], relay.Let(tup, relay.Tuple([x, y, z]),
                                               relay.TupleGetItem(tup, 0) +
                                               relay.TupleGetItem(tup, 1) -
                                               relay.TupleGetItem(tup, 2)))
    back_func = relay.ir_pass.infer_type(gradient(func))
    assert back_func.checked_type == relay.FuncType([t, t, t], relay.TupleType([t, relay.TupleType([t, t, t])]))
    x_nd = rand(dtype, *shape)
    y_nd = rand(dtype, *shape)
    z_nd = rand(dtype, *shape)
    x_np = x_nd.asnumpy()
    y_np = y_nd.asnumpy()
    z_np = z_nd.asnumpy()
    expected_forward = x_np + y_np - z_np
    ex = create_executor()
    forward, (grad_x, grad_y, grad_z) = ex.evaluate(back_func)(x_nd, y_nd, z_nd)
    tvm.testing.assert_allclose(forward.asnumpy(), expected_forward)
    tvm.testing.assert_allclose(grad_x.asnumpy(), np.ones_like(grad_x.asnumpy()))
    tvm.testing.assert_allclose(grad_y.asnumpy(), np.ones_like(grad_y.asnumpy()))
    tvm.testing.assert_allclose(grad_z.asnumpy(), -1 * np.ones_like(grad_z.asnumpy()))

def test_recursion():
    """
    Program:
       def @f(%n: int32, %data: float32) -> float32 {
          if (%n == 0) {
              %data
          } else {
              @f(%n - 1, log(%data))
          }
       }
    """
    sb = relay.ScopeBuilder()
    f = relay.GlobalVar("f")
    ti32 = relay.scalar_type("int32")
    tf32 = relay.scalar_type("float32")
    n = relay.var("n", ti32)
    data = relay.var("data", tf32)

    with sb.if_scope(relay.equal(n, relay.const(0, ti32))):
        sb.ret(data)
    with sb.else_scope():
        sb.ret(f(relay.subtract(n, relay.const(1, ti32)), relay.log(data)))
    mod = relay.Module()
    mod[f] = relay.Function([n, data], sb.get())
    assert "@f(%1, %2) /* ty=float32 */" in mod.astext()
    assert mod[f].checked_type == relay.FuncType([ti32, tf32], tf32)

def _test_upsampling(layout, method):
    n, c, h, w = tvm.var("n"), 16, 32, 32
    scale = 2
    dtype = "float32"
    def get_shape():
        if layout == "NCHW":
            return (c, h, w), (c, h*scale, w*scale)
        else:
            return (h, w, c), (h*scale, w*scale, c)
    ishape, oshape = get_shape()
    x = relay.var("x", relay.TensorType((n,) + ishape, dtype))
    y = relay.nn.upsampling(x, scale=scale, layout=layout, method=method)
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((n,) + oshape, dtype)
    dshape = (1,) + ishape
    x = relay.var("x", shape=dshape)
    y = relay.nn.upsampling(x, scale=scale, layout=layout, method=method)
    func = relay.Function([x], y)
    data = np.random.uniform(size=dshape).astype(dtype)
    if method == "NEAREST_NEIGHBOR":
        ref = topi.testing.upsampling_python(data, scale, layout)
    else:
        ref = topi.testing.bilinear_resize_python(data, (h*scale, w*scale), layout)
    for target, ctx in ctx_list():
        executor = relay.create_executor("graph", ctx=ctx, target=target)
        out = executor.evaluate(func)(data)
        tvm.testing.assert_allclose(out.asnumpy(), ref, rtol=1e-5, atol=1e-5)

    def run_test_conv2d(dtype, out_dtype, scale, dshape, kshape,
                        padding=(1, 1),
                        fref=None,
                        groups=1,
                        dilation=(1, 1),
                        except_targets=None,
                        **attrs):
        if except_targets is None:
          except_targets = []
          
        x = relay.var("x", shape=dshape, dtype=dtype)
        w = relay.var("w", dtype=dtype)
        y = relay.nn.conv2d(x, w,
                            padding=padding,
                            dilation=dilation,
                            groups=groups,
                            **attrs)
        func = relay.Function([x, w], y)
        data = np.random.uniform(-scale, scale, size=dshape).astype(dtype)
        kernel = np.random.uniform(-scale, scale, size=kshape).astype(dtype)
        dkernel = topi.testing.dilate_python(kernel, (1, 1) + dilation)
        if fref is None:
            ref_res = topi.testing.conv2d_nchw_python(
                data.astype(out_dtype), dkernel.astype(out_dtype), 1, padding,
                groups=groups)
        else:
            ref_res = fref(data.astype(out_dtype), dkernel.astype(out_dtype))


        for target, ctx in ctx_list():
            if target in except_targets:
                continue
            intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
            op_res1 = intrp1.evaluate(func)(data, kernel)
            tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5, atol=1e-5)

def test_lrn():
    n, c , h, w = tvm.var("n"), tvm.var("c"), tvm.var("h"), tvm.var("w")
    x = relay.var("x", shape=(n, c , h, w))
    y = relay.nn.lrn(x, size=10, axis=2, bias=0.5, alpha=.00001, beta=0.75)
    "alpha=" in y.astext()
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((n, c , h, w))

    shape = (1, 5, 10, 10)
    dtype = "float32"
    x = relay.var("x", relay.TensorType(shape, dtype))
    size=5
    axis=1
    bias=0.5
    alpha=.00001
    beta=0.75
    z = relay.nn.lrn(x, size=size, axis=axis, bias=bias, alpha=alpha, beta=beta)
    yy = relay.ir_pass.infer_type(z)
    assert yy.checked_type == relay.TensorType(shape, dtype)
    func = relay.Function([x], z)
    x_data = np.random.uniform(low=-1, high=1, size=shape).astype(dtype)
    ref_res = topi.testing.lrn_python(x_data, size, axis, bias, alpha, beta)

    for target, ctx in ctx_list():
        intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
        intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
        op_res1 = intrp1.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5)
        op_res2 = intrp2.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-5)

def test_l2_normalize():
    n, c , h, w = tvm.var("n"), tvm.var("c"), tvm.var("h"), tvm.var("w")
    x = relay.var("x", shape=(n, c , h, w))
    y = relay.nn.l2_normalize(x, eps=0.001, axis=[1])
    "axis=" in y.astext()
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((n, c , h, w))

    shape = (1, 5, 10, 10)
    dtype = "float32"
    x = relay.var("x", relay.TensorType(shape, dtype))
    eps=0.001
    axis=1
    z = relay.nn.l2_normalize(x, eps=0.001, axis=[axis])
    yy = relay.ir_pass.infer_type(z)
    assert yy.checked_type == relay.TensorType(shape, dtype)
    func = relay.Function([x], z)
    x_data = np.random.uniform(low=-1, high=1, size=shape).astype(dtype)
    ref_res = topi.testing.l2_normalize_python(x_data, eps, axis)

    for target, ctx in ctx_list():
        intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
        intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
        op_res1 = intrp1.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5)
        op_res2 = intrp2.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-5)

def test_flatten_infer_type():
    d1, d2, d3, d4 = tvm.var("d1"), tvm.var("d2"), tvm.var("d3"), tvm.var("d4")
    x = relay.var("x", relay.TensorType((d1, d2, d3, d4), "float32"))
    y = relay.nn.batch_flatten(x)
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((d1, ((d2*d3)*d4)), "float32")

    x = relay.var("x", relay.TensorType((3, 2, 4, 3), "float32"))
    y = relay.nn.batch_flatten(x)
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((3, 24), "float32")

    x = relay.var("x", relay.TensorType((d1, 2, d3, 3), "float32"))
    y = relay.nn.batch_flatten(x)
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((d1, ((2*d3)*3)), "float32")

    shape = (1, 5, 10, 10)
    o_shape = (1, 500)
    dtype = "float32"
    x = relay.var("x", relay.TensorType(shape, dtype))
    z = relay.nn.batch_flatten(x)
    yy = relay.ir_pass.infer_type(z)
    assert yy.checked_type == relay.TensorType(o_shape, dtype)
    func = relay.Function([x], z)
    x_data = np.random.uniform(low=-1, high=1, size=shape).astype(dtype)
    ref_res = x_data.flatten().reshape(o_shape)

    for target, ctx in ctx_list():
        intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
        intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
        op_res1 = intrp1.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5)
        op_res2 = intrp2.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-5)

def test_conv2d_transpose_infer_type():
    # symbolic in batch dimension
    n, c, h, w = tvm.var("n"), 10, 10, 12
    x = relay.var("x", relay.TensorType((n, c, h, w), "float32"))
    w = relay.var("w", relay.IncompleteType())
    y = relay.nn.conv2d_transpose(x, w,
                                  kernel_size=(3, 3),
                                  padding=(1, 1),
                                  channels=15)
    assert "channels=15" in y.astext()
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType(
        (n, 15, 10, 12), "float32")
    assert yy.args[1].checked_type == relay.TensorType(
        (10, 15, 3, 3), "float32")

    # infer by shape of w, mixed precision
    n, c, h, w = tvm.var("n"), 10, 10, 12
    x = relay.var("x", relay.TensorType((n, c, h, w), "float32"))
    w = relay.var("w", relay.TensorType((12, 11, 5, 5), "float32"))
    y = relay.nn.conv2d_transpose(x, w,
                                  output_padding=(1, 1),
                                  channels=11,
                                  data_layout="NHWC")
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType(
        (n, 15, 15, 11), "float32")

def test_binary_int_broadcast():
    for op, ref in [(relay.right_shift, np.right_shift),
               (relay.left_shift, np.left_shift),
                (relay.mod, np.mod),
               (relay.maximum, np.maximum),
               (relay.minimum, np.minimum)]:
        x = relay.var("x", relay.TensorType((10, 4), "int32"))
        y = relay.var("y", relay.TensorType((5, 10, 1), "int32"))
        z = op(x, y)
        zz = relay.ir_pass.infer_type(z)
        assert zz.checked_type == relay.TensorType((5, 10, 4), "int32")

    if ref is not None:
        x_shape = (10, 4)
        y_shape = (5, 10, 1)
        t1 = relay.TensorType(x_shape, 'int32')
        t2 = relay.TensorType(y_shape, 'int32')
        x_data = np.random.rand(*x_shape).astype(t1.dtype)
        y_data = np.random.rand(*y_shape).astype(t2.dtype)
        func = relay.Function([x, y], z)
        ref_res = ref(x_data, y_data)

        for target, ctx in ctx_list():
            intrp = relay.create_executor("graph", ctx=ctx, target=target)
            op_res = intrp.evaluate(func)(x_data, y_data)
            tvm.testing.assert_allclose(op_res.asnumpy(), ref_res)

def test_infer_type_leaky_relu():
    n, c , h, w = tvm.var("n"), tvm.var("c"), tvm.var("h"), tvm.var("w")
    x = relay.var("x", relay.TensorType((n, c, h, w), "float32"))
    y = relay.nn.leaky_relu(x, alpha=0.1)
    "alpha=0.1" in y.astext()
    yy = relay.ir_pass.infer_type(y)
    assert yy.checked_type == relay.TensorType((n, c, h, w), "float32")

    shape = (1, 5, 10, 10)
    dtype = "float32"
    x = relay.var("x", relay.TensorType(shape, dtype))
    z = relay.nn.leaky_relu(x, alpha=0.1)
    assert "alpha=0.1" in z.astext()
    yy = relay.ir_pass.infer_type(z)
    assert yy.checked_type == relay.TensorType(shape, dtype)
    func = relay.Function([x], z)
    x_data = np.random.uniform(low=-1, high=1, size=shape).astype(dtype)
    ref_res = np.where(x_data > 0, x_data, x_data * 0.1)

    for target, ctx in ctx_list():
        intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
        intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
        op_res1 = intrp1.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5)
        op_res2 = intrp2.evaluate(func)(x_data)
        tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-5)

    def test_threshold():
        num_anchors = 5
        num_classes = 5
        n = tvm.var("n")
        cls_prob = relay.var(
            "cls_prob",
            relay.ty.TensorType((n, num_anchors, num_classes), "float32"))
        loc_pred = relay.var(
            "loc_pred", relay.ty.TensorType((n, num_anchors * 4), "float32"))
        anchors = relay.var(
            "anchors", relay.ty.TensorType((1, num_anchors, 4), "float32"))
        threshold = 0.02
        variances = (0.2, 0.2, 0.3, 0.3)

        ret = relay.vision.multibox_transform_loc(
            cls_prob=cls_prob,
            loc_pred=loc_pred,
            anchor=anchors,
            threshold=threshold,
            variances=variances)
        ret = relay.ir_pass.infer_type(ret.astuple())
        ref_type = relay.ty.TupleType(
            tvm.convert([
                relay.ty.TensorType((n, num_anchors, 6), "float32"),
                relay.ty.TensorType((n, ), "int")
            ]))
        assert ret.checked_type == ref_type

def verify_infer_type_prelu(data, alpha, axis, output, dtype="float32"):
    x = relay.var("data", relay.TensorType(data, dtype))
    if alpha:
        y = relay.var("alpha", relay.TensorType(alpha, dtype))
    else:
        y = relay.var("alpha", relay.IncompleteType())
    z = relay.nn.prelu(x, y, axis=axis)
    zz = relay.ir_pass.infer_type(z)
    if axis != 1:
        assert "axis" in z.astext()
    assert zz.checked_type == relay.ty.TensorType(output, dtype)
    if not alpha:
        axis = axis if axis else 1
        alpha_shape = (data[axis],)
        assert zz.args[1].checked_type == relay.TensorType(alpha_shape, "float32")

    if all(isinstance(v, tvm.expr.Var) == 1 for v in data) or not alpha:
        return

    func = relay.Function([x, y], z)
    x_data = np.random.uniform(low=-1, high=1, size=data).astype(dtype)
    a_data = np.random.uniform(low=-1, high=1, size=alpha).astype(dtype)

    if axis == 1:
        ref_res = (x_data < 0) * (x_data * a_data.reshape(3, 1, 1)) + (x_data>=0) * x_data
    else:
        ref_res = (x_data < 0) * (x_data * a_data.reshape(1, 1, 3)) + (x_data>=0) * x_data

    for target, ctx in ctx_list():
        intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
        intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
        op_res1 = intrp1.evaluate(func)(x_data, a_data)
        tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5)
        op_res2 = intrp2.evaluate(func)(x_data, a_data)
        tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-5)

    def verify_roi_pool(data_shape, rois_shape, pooled_size, spatial_scale):
        data = relay.var("data", relay.ty.TensorType(data_shape, "float32"))
        rois = relay.var("rois", relay.ty.TensorType(rois_shape, "float32"))
        z = relay.vision.roi_pool(data, rois, pooled_size=(pooled_size, pooled_size),
                                   spatial_scale=spatial_scale, layout="NCHW")
        zz = relay.ir_pass.infer_type(z)

        batch, channel, in_size, _ = data_shape
        num_roi = rois_shape[0]
        assert zz.checked_type == relay.ty.TensorType(
                (num_roi, channel, pooled_size, pooled_size), "float32")

        func = relay.Function([data, rois], z)
        func = relay.ir_pass.infer_type(func)
        np_data = np.random.uniform(size=data_shape).astype("float32")
        np_rois = np.random.uniform(size=rois_shape).astype('float32') * in_size
        np_rois[:, 0] = np.random.randint(low = 0, high = batch, size = num_roi).astype('float32')
        ref_res = topi.testing.roi_pool_nchw_python(np_data, np_rois, pooled_size=pooled_size,
                                                     spatial_scale=spatial_scale)
        for target, ctx in ctx_list():
            intrp1 = relay.create_executor("graph", ctx=ctx, target=target)
            op_res1 = intrp1.evaluate(func)(np_data, np_rois)
            tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-4)
            intrp2 = relay.create_executor("debug", ctx=ctx, target=target)
            op_res2 = intrp2.evaluate(func)(np_data, np_rois)
            tvm.testing.assert_allclose(op_res2.asnumpy(), ref_res, rtol=1e-4)

def test_broadcast_add():
    shape1 = (3, 4, 1)
    shape2 = (1, 5)
    dtype = 'float32'
    x_nd = rand(dtype, *shape1)
    y_nd = rand(dtype, *shape2)
    x_np = x_nd.asnumpy()
    y_np = y_nd.asnumpy()
    expected_forward = x_np + y_np
    t1 = relay.TensorType(shape1, dtype)
    t2 = relay.TensorType(shape2, dtype)
    x = relay.var("x", t1)
    y = relay.var("y", t2)
    func = relay.Function([x, y], x + y)
    full_func = relay.ir_pass.infer_type(gradient(func))
    assert full_func.checked_type == relay.FuncType([t1, t2],
                                                    relay.TupleType([relay.TensorType(expected_forward.shape, dtype),
                                                                     relay.TupleType([t1, t2])]))
    ex = create_executor()
    forward, (grad_x, grad_y) = ex.evaluate(full_func)(x_nd, y_nd)
    tvm.testing.assert_allclose(forward.asnumpy(), expected_forward)
    tvm.testing.assert_allclose(grad_x.asnumpy(),
                                np.ones_like(expected_forward).sum(axis=2, keepdims=True))
    tvm.testing.assert_allclose(grad_y.asnumpy(),
                                np.ones_like(expected_forward).sum(axis=(0, 1), keepdims=True).squeeze(axis=0))

    def test_run(batch, in_channel, size, out_channel, deformable_groups, groups):
        kernel_size = (3, 3)
        data_shape = (batch, in_channel, size, size)
        offset_shape = (batch, 2 * kernel_size[0] * kernel_size[1] * deformable_groups, size, size)
        kernel_shape = (out_channel, in_channel // groups, kernel_size[0], kernel_size[1])
        dtype = 'float32'
        data = relay.var("data", shape=data_shape, dtype=dtype)
        offset = relay.var("offset")
        kernel = relay.var("kernel")
        y = relay.nn.deformable_conv2d(data, offset, kernel,
            strides=(1, 1),
            padding=(1, 1),
            dilation=(1, 1),
            kernel_size=kernel_size,
            deformable_groups=deformable_groups,
            groups=groups,
            channels=out_channel)
        func = relay.Function([data, offset, kernel], y)
        data = np.random.uniform(size=data_shape).astype(dtype)
        offset = np.random.uniform(size=offset_shape).astype(dtype)
        kernel = np.random.uniform(size=kernel_shape).astype(dtype)
        ref_res = topi.testing.deformable_conv2d_nchw_python(data, offset, kernel, stride=(1, 1), padding=(1, 1), dilation=(1, 1), deformable_groups=deformable_groups, groups=groups)

        for target, ctx in ctx_list():
            for kind in ["graph", "debug"]:
                intrp1 = relay.create_executor(kind, ctx=ctx, target=target)
                op_res1 = intrp1.evaluate(func)(data, offset, kernel)
                tvm.testing.assert_allclose(op_res1.asnumpy(), ref_res, rtol=1e-5, atol=1e-5)