Function ba_gates_warp 

pub fn ba_gates_warp(local_out: u32, lane: u32) -> u32

The block-wide warp index that owns lane lane of output local_out.

The parent writes its warp partial to smem[local_out * 2 + (lane / 32)]; the twin writes red[g * BA_GATES_WARPS + threadIdx.x / 32]. Those are the same slot only because threadIdx.x / 32 == local_out * 2 + lane / 32, which is what this function and its test pin.