Struct GpuWeightSlots 

pub struct GpuWeightSlots { /* private fields */ }

Device-resident mapping from neural output slots to CNF variable ids.

Slots are grouped (one group per neural predicate instance). Each slot is a CNF var id (DIMACS, 1-based) whose log-weights should be updated from the group’s probability vector.

Implementations

impl GpuWeightSlots

pub fn upload( provider: &CudaKernelProvider, groups: &[Vec<u32>], ) -> Result<Self>

Upload a slot mapping from host vectors.

groups[g][i] is the CNF variable id corresponding to label/slot i of group g.

pub fn num_groups(&self) -> u32

Returns the number of neural-output groups as a GPU-sized count.

pub fn num_groups_usize(&self) -> usize

Returns the number of neural-output groups as a host index.

pub fn total_slots(&self) -> u32

Returns the total number of mapped neural output slots.

pub fn group_offsets(&self) -> &TrackedCudaSlice<u32>

Returns the device-resident CSR offsets for all groups.

pub fn slot_cnf_var(&self) -> &TrackedCudaSlice<u32>

Returns the device-resident CNF variable identifier for every slot.

pub fn group_slot_cnf_var( &self, group_idx: usize, ) -> Result<DeviceMemoryView<u32>>

Device view over slot_cnf_var for a single group.