pub const BA_GATES_GROUPS: u32 = 8;Expand description
2026-09-25: BAH_GROUPS in the twin: output groups a thread accumulates per
read of the activation row. The row is read ceil(ceil(N/4)/8) * 4 times per
token, 12 at N = 96. The kernel source records the register and occupancy
table behind the choice of 8.