Some switch chips have a feature (also supported in SwOS - forwarding table) where for N ports there is matrix of NxN bits to allow forwarding between any combination of ingress/egress ports. Why isn't bridge split-horizon hw-offload implemented simply by grouping ports by the same nonzero horizon value and setting these bits in the switch chip appropriately for all such groups of ports? Seems simple to do, or am I missing something? Pseudo code:

Some switch chips have a feature to deal with VLAN tags and VLAN-aware switching. Bridge vlan-filtering is not HW-offloaded simply by using this functionality ...

I use bridge VLAN filtering, and hw-offload works fine here on CRS328-24P-4S+ with ROS 6.42.7 and now 6.43.2 - as long as I don't try to isolate ports by setting horizon.

I use bridge VLAN filtering, and hw-offload works fine here on CRS328-24P-4S+ with ROS 6.42.7 and now 6.43.2 - as long as I don't try to isolate ports by setting horizon.

CRS series 300 is AFAIK currently only HW that does HW offload non-trivial tasks. HW offload on all other device types is currently severely limited.

Good to know, I thought VLAN was something basic that should be hw-offloaded on all products (except mixing tagged and untagged on the same port - "hybrid ports" are not supported by some cheap "smart" switches, it could be that some switch chips really can't do it even though they can do VLANs).

...I thought VLAN was something basic that should be hw-offloaded on all products ...

So did I