Fast tile search

[00shiv]

Does anyone know if the techniques on dynamic shared memory from this old thread:

Thinking through distance matrix calculation

Note that GPU arrays are tied to the Julia GC, and are freed after they go out of scope on the host, so that might be your problem. That means you need to keep data that is alive on the GPU, alive on the CPU as well. Lastly, because there’s no way for us to indicate GPU memory pressure to the Julia GC, you might need to manually finalize(arr) at some points. I played some more with the example, now also mapping the outer loop into the GPU’s execution model (makes the code cleaner) and introduc…

are still valid in 1.8.2?
Thanks.
–shiv–