Function moe_expert_silu_down_shared 

pub fn moe_expert_silu_down_shared(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    gate_out: DevicePtr,
    up_out: DevicePtr,
    packed_ptrs: DevicePtr,
    scale_ptrs: DevicePtr,
    scale2_vals: DevicePtr,
    output: DevicePtr,
    expert_indices: DevicePtr,
    sh_gate_in: DevicePtr,
    sh_up_in: DevicePtr,
    sh_down: &QuantizedWeight,
    sh_down_out: DevicePtr,
    n: u32,
    k: u32,
    top_k: u32,
    stream: u64,
) -> Result<()>

2026-09-25: NVFP4 SiLU+down GEMVs in one launch. Routed experts read gate_out/up_out; the shared expert reads sh_gate_in/sh_up_in. The kernel’s s_act is dynamic shared memory of K floats.