pub fn nvfp4_silu_mul_scaled(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
gate: DevicePtr,
up: DevicePtr,
out: DevicePtr,
gate_scale: f32,
up_scale: f32,
total: u32,
stream: u64,
) -> Result<()>Expand description
SiLU(gate×gs)×(up×us) with the per-projection scale2 fold (swiglu ±10 clamp, mirrors moe_silu_mul). In-place safe (out may alias gate).