Function moe_build_tile_worklist 

pub fn moe_build_tile_worklist(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    expert_offsets: DevicePtr,
    weight_ptrs: DevicePtr,
    worklist: DevicePtr,
    total_tiles: DevicePtr,
    num_experts: u32,
    n_tiles: u32,
    m_tile: u32,
    stream: u64,
) -> Result<()>

Build the compacted (expert, m_tile, n_tile) work-list for the persistent grouped-GEMM grid. Single-block, thread-0 serial — mirrors the moe_sort_by_expert launch style (grid [1,1,1], block [256,1,1]).

n_tiles = div_ceil(N, 64) (PM4_N_TILE) and m_tile = 128 (PM4_M_TILE). Writes worklist[*total_tiles * 2] (word0=expert, word1=(m_tile<<6)|n_tile) and total_tiles[0].

SAME-STREAM INVARIANT: the caller MUST launch moe_fp8_grouped_gemm on the SAME stream so the kernel’s read of total_tiles/worklist happens-after this write (no cross-stream event is inserted).