Function w4a16_gemv_sw_raw 

pub fn w4a16_gemv_sw_raw(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    packed: DevicePtr,
    scale: DevicePtr,
    scale_2: f32,
    output: DevicePtr,
    n: u32,
    k: u32,
    stream: u64,
) -> Result<()>

Same launch as w4a16_gemv_sw for callers that hold the NVFP4 operand triple as loose pointers rather than a QuantizedWeight (GLM-5.3’s Nvfp4Proj).

Exists so the ceil(N/8) grid stays in this file — the one place that is SSOT with the kernel’s N_PER_BLOCK_SW.