Predicate store for GPU

[UncleHandsome]

Hello
I want to use predicate on GPU in Generator.
I tried following algorithm but didn't work at all.
Is there any suggestion to do the same thing??
Thank you

output(x, y) = select(x < output.width(), likely(1), undef<uint8_t>());

The schedule is simply something like this:

output
    .split(x, x, tx, 64, TailStrategy::RoundUp)
    .gpu_threads(tx)
    .gpu_blocks(x)