Function mamba2_ssm_prefill_persistent 

pub fn mamba2_ssm_prefill_persistent(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    h_state: DevicePtr,
    x: DevicePtr,
    b_proj: DevicePtr,
    c_proj: DevicePtr,
    dt_raw: DevicePtr,
    a_log: DevicePtr,
    d_param: DevicePtr,
    dt_bias: DevicePtr,
    output: DevicePtr,
    batch_size: u32,
    seq_len: u32,
    num_heads: u32,
    head_dim: u32,
    state_size: u32,
    n_groups: u32,
    dt_min: f32,
    dt_max: f32,
    x_stride: u32,
    bc_stride: u32,
    dt_stride: u32,
    y_stride: u32,
    stream: u64,
) -> Result<()>

2026-09-25: Mamba-2 SSM prefill with each block’s H slice in shared memory for the whole token loop (mamba2_ssm_prefill_persistent). Same arguments as mamba2_ssm_prefill, but launched with head_dim * SUB threads and dynamic shared memory.