I’m not familiar with LZO, is it a parallelizable algorithm? I know LZ77 is essentially a sequential algorithm and it’s not trivial to parallelize it (though it is possible and has been done) so if LZO is anything like LZ77, it probably won’t be easy. Then again, I’m no specialist on compression algorithms (and I only replied to this thread because no one else did :) )
To achieve those great speed-ups CUDA advertises, the algorithm must not only be computationally intensive but also (and mainly) massively parallelizable.
Where the discussion ended in the other thread was, “Do you really have a need?” If, for you, the answer is yes, then there’s always ways and means (like cutting up the data into pieces and doing it multi-serial).