Additional note:
- Same fence and command buffers are used.
- Dispatches are with number of groups to cover data i.e (2,1,1) -> 256*2 for 400 data entires. with indices higher than 400 are ignored.
- I had same problems earlier, but, likely, it was because i was dispatching 256 groups!
- If compute shaders process same data, over and over, they dont crash.