Function dense_gemm_m16_bf16 

pub fn dense_gemm_m16_bf16(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    weight: &DenseWeight,
    output: DevicePtr,
    m: u32,
    n: u32,
    k: u32,
    a_row_stride: u32,
    c_row_stride: u32,
    stream: u64,
) -> Result<()>

The default 32-wide CTA. input is [m, a_row_stride] BF16 with k used, weight.weight is the raw [n, k] BF16 checkpoint tensor (NOT copied, NOT quantized — the same pointer dense_gemv_batchm reads), and output is [m, c_row_stride] BF16 with n used.