Function gdn_prefill_split4 

pub fn gdn_prefill_split4(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    h_state: DevicePtr,
    query: DevicePtr,
    key: DevicePtr,
    value: DevicePtr,
    gate: DevicePtr,
    beta: DevicePtr,
    output: DevicePtr,
    batch_size: u32,
    seq_len: u32,
    num_k_heads: u32,
    num_v_heads: u32,
    k_dim: u32,
    v_dim: u32,
    qk_stride: u32,
    v_stride: u32,
    gb_stride: u32,
    stream: u64,
) -> Result<()>

4-way split prefill: 4 CTAs per v-head, 32 threads each (128 total CTAs).

Kernel: gated_delta_rule_prefill_split4(h_state, query, key, value, gate, beta, output, batch_size, seq_len, num_k_heads, num_v_heads, k_dim, v_dim, qk_stride, v_stride, gb_stride) Grid: (num_v_heads * 4, batch, 1) Block: (32, 1, 1)