[Bug] [scatterND] large shape will get a Segmentation fault in x86

[chengven027]

I do a test with scatterND:

def test_scatternd():
    def before():
        data = relay.const(np.zeros((1, 900, 300), dtype="float32"), dtype="float32")
        indices = relay.const(np.ones((3, 1, 900, 300), dtype="int64"), dtype="int64")
        update = relay.const(np.ones((1, 900, 300), dtype="float32"), dtype="float32")
        b = relay.op.scatter_nd(data, indices, update)
        return relay.Function(relay.analysis.free_vars(b), b)

    passes = tvm.transform.Sequential(
        [
            relay.transform.InferType(),
            relay.transform.FoldConstant(),
        ]
    )

    before_mod = tvm.IRModule.from_expr(before())
    with tvm.transform.PassContext(opt_level=3):
        after_mod = passes(before_mod)

I found if the shape is (1, 900, 200), the after_mod will success execute. and if I use the shape of (1, 900, 300). I will get a Segmentation fault.

[chengven027]

close. #12200 to follow this issue.