Function moe_w4a16_fused_gate_up_n128 

pub fn moe_w4a16_fused_gate_up_n128(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    a: DevicePtr,
    gate_packed_ptrs: DevicePtr,
    gate_scale_ptrs: DevicePtr,
    gate_scale2_vals: DevicePtr,
    up_packed_ptrs: DevicePtr,
    up_scale_ptrs: DevicePtr,
    up_scale2_vals: DevicePtr,
    c_gate: DevicePtr,
    c_up: DevicePtr,
    expert_offsets: DevicePtr,
    sorted_token_ids: DevicePtr,
    num_experts: u32,
    n_out: u32,
    k: u32,
    max_m_tiles: u32,
    stream: u64,
) -> Result<()>

Fused gate+up grouped GEMM — single launch for both projections.

Grid: (ceil(2*n_out/128), max_m_tiles, num_experts) Block: (128, 1, 1) First N cols → gate weights/output, last N cols → up weights/output.