it is somewhat complicated. The stepgens do not get their position commands directly from the g-code. There are a few layers in between.
One of the motion planners (there is more than one) sends the commands for movement to the kinematics model, that then sends the joint position commands either directly to or via a PID loop to the stepgen. (I understand that the motion planner used for jogging isn't the same as the one used for program execution.)
The motion planner reads the G-code, calculates what the current commanded position should be and adds required offsets, before sending the axis positio commands to the Kinematics program.
After the Kinematics program translates the axis positions to joint positions, then any additional joint offsets will be added (such as home offsets.
