The critical path
We examine 10 bits from high end of r.
Those drive the logic described here.
Perhaps those choose 4 single irredundant numbers to be accumulated into the redundant r.
(By irredundant I mean one bit per bit as contrasted to the redundant two bits per bit for the low 50 bits of r. (CSA))
That summation is eight levels of logic as described here which is heavy duty in gate count but limited to 4 way fan-in and 4 way fan-out.
I judge not to be in the critical path.
To get the high bits for the next value of r we must sum 6 12 bit operands.
As we shift r left 8 bits each cycle we must make 8 more bits irredundant.
Then there are the four high order 10 (roughly) bits from the “choose 4” logic above.
That closes the loop.