Thank you. But how could the program predict the device's losing? Or just copy the data every frame? It seems not so efficient.
Once you lose the device, you lose all of your render targets. You would have to copy the data before you lose the device.
Because those render targets are used for saving states of a GPU-based particle system. I think it'll be weird when all particles are disappeared or reseted at their initial position or other states after a device' losing.
Why not just re-render what was originally rendered into your render targets?