Function w8a16_gemv_batch16_ncol2 

pub fn w8a16_gemv_batch16_ncol2(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    weight: DevicePtr,
    block_scale: DevicePtr,
    output: DevicePtr,
    m: u32,
    n: u32,
    k: u32,
    stream: u64,
) -> Result<()>

N_COLS=2, contiguous A [M, K] and C [M, N].

Kernel: w8a16_gemv_batch16_ncol2 (module w8a16_gemv_ncol). Grid: (ceil(N/8), 1, 1) Block: (256, 1, 1)