Function conv1d_update_prefill 

pub fn conv1d_update_prefill(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    conv1d_prefill_tp_k: KernelHandle,
    conv_state: DevicePtr,
    input: DevicePtr,
    weight: &DenseWeight,
    bias: DevicePtr,
    output: DevicePtr,
    d_inner: u32,
    d_conv: u32,
    seq_len: u32,
    input_stride: u32,
    output_stride: u32,
    stream: u64,
) -> Result<()>

Multi-token conv1d sliding window update + SiLU for prefill.

Processes seq_len tokens sequentially per channel in registers. Input/output may be non-contiguous (different strides between tokens).

Kernel: causal_conv1d_update_prefill(conv_state, input, weight, bias, output, dim, d_conv, seq_len, input_stride, output_stride) Grid: (ceil(dim/256), 1, 1) Block: (256, 1, 1)