pub fn gdn_wy4_fold(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
h_table: DevicePtr,
stash: DevicePtr,
na_tab: DevicePtr,
hi_tables: DevicePtr,
slab_entries: u32,
engaged_flag: DevicePtr,
k_rows: u32,
batch_size: u32,
num_k_heads: u32,
num_v_heads: u32,
k_dim: u32,
v_dim: u32,
stash_seq_floats: u32,
stream: u64,
) -> Result<()>Expand description
Post-verdict fold (gated_delta_rule_wy4_fold): applies rows 0..na_tab[b]
of the stashed updates to H, one read + one write per state.