Function nvfp4_silu_mul_quant 

pub fn nvfp4_silu_mul_quant(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    gate: DevicePtr,
    up: DevicePtr,
    out_y: DevicePtr,
    gate_scale: f32,
    up_scale: f32,
    m: u32,
    k: u32,
    stream: u64,
) -> Result<()>

Fused SiLU-mul + block_fp4_mmq quantize for the down-MMQ path: reads RAW gate/up MMQ outputs, applies the scale2 folds + swiglu clamp + SiLU-mul, and quantizes straight into the down GEMM’s y-format — the intermediate bf16 activation tensor is never written (this round-trip is why the unfused down arm measured neutral).