gather_bgmv_grids

metrale_model_layers::layers::ops::moe_lora_grouped

Function gather_bgmv_grids 

Source 
pub fn gather_bgmv_grids(
    max_rank: u32,
    n_out: u32,
    n_slots: u32,
) -> ([u32; 3], [u32; 3])
Expand description

2026-09-25: (shrink, expand) grids of the gather fold: [ceil(out/4), n_slots, 1], where out is max_rank for the shrink and n_out for the expand. Each 256-thread block computes 4 outputs, 64 threads per output (GBGMV_N_PER_BLOCK in moe_lora_gather_bgmv.cu).