I’ve never interacted with one of these offload engines without many abstraction layers between me and it but I’m curious what ‘stateless’ means here. Formats like AV1 are intrinsically stateful. Most frames use inter-frame compression and so provide motion vectors and so on to apply to the previous one. I think AV1 (optionally) does bidirectional compression so may also refer to things in the next keyframe. This seems like it would require a non-trivial amount of state in the hardware. Does anyone know how they avoid this?
I’ve never interacted with one of these offload engines without many abstraction layers between me and it but I’m curious what ‘stateless’ means here. Formats like AV1 are intrinsically stateful. Most frames use inter-frame compression and so provide motion vectors and so on to apply to the previous one. I think AV1 (optionally) does bidirectional compression so may also refer to things in the next keyframe. This seems like it would require a non-trivial amount of state in the hardware. Does anyone know how they avoid this?