Struct GraphDecodeState 

pub struct GraphDecodeState {
    pub token_d: CudaSlice<u32>,
    pub pos_d: CudaSlice<i32>,
    pub graphs: HashMap<(bool, usize), CudaGraph>,
    pub bucket_max: HashMap<(bool, usize), usize>,
    pub captures: usize,
}

Persistent CUDA-graph decode state (CUDA-GRAPH-PLAN Phase 3). Holds the device-resident counters the captured graph reads/writes (token_d = current/next token id, pos_d = rope position) — both at FIXED addresses baked into every captured graph — plus the per-t_kv-bucket graph cache. The bucket key is the eager (fa_vec, n_splits) pair (see Engine::fa_bucket_key): every t_kv that maps to the same key reproduces eager’s split geometry, so one captured graph replays bit-identically for the whole bucket. A new key triggers a re-capture (n_splits changes ~every 64 tokens).

Fields

token_d: CudaSlice<u32>

pos_d: CudaSlice<i32>

graphs: HashMap<(bool, usize), CudaGraph>

bucket_max: HashMap<(bool, usize), usize>

captures: usize

Implementations

impl GraphDecodeState

pub fn new(e: &Engine) -> Result<Self, Box<dyn Error>>