Hi,

The acceleration that most of the 2D things provide isn't ever that
great, and shadowing is a lot more effective if done properly.

That is probably true for anything pci-ish, because those devices are
optimized for memory writes and reads are horribly slow.  So you surely
want avoid device memory reads and shadowing is a effective way to do
this.

On arm hardware the tradeoff may look quite different, the cpus are
relatively slow and I think most arm gpus don't have dedicated device
memory ...

cheers,
  Gerd