Struct GpuCnf 

pub struct GpuCnf {
    pub var_cap: u32,
    pub clause_cap: u32,
    pub lit_cap: u32,
    pub num_vars: TrackedCudaSlice<u32>,
    pub num_clauses: TrackedCudaSlice<u32>,
    pub num_lits: TrackedCudaSlice<u32>,
    pub clause_offsets: TrackedCudaSlice<u32>,
    pub literals: TrackedCudaSlice<i32>,
}

GPU-resident CNF in CSR form (DIMACS literals, 1-based variable ids).

This is the solver-facing CNF representation used by the GPU CDCL verifier.

Fields

var_cap: u32

Variable capacity (>= num_vars).

clause_cap: u32

Clause capacity (>= num_clauses).

lit_cap: u32

Literal capacity (>= num_lits).

num_vars: TrackedCudaSlice<u32>

Device-resident num_vars (len = 1).

num_clauses: TrackedCudaSlice<u32>

Device-resident num_clauses (len = 1).

num_lits: TrackedCudaSlice<u32>

Device-resident num_lits (len = 1).

clause_offsets: TrackedCudaSlice<u32>

CSR offsets (len = clause_cap + 1).

literals: TrackedCudaSlice<i32>

Flattened CSR literal array (len = lit_cap).

Implementations

impl GpuCnf

pub fn num_literals_cap(&self) -> usize

pub fn from_host( instance: &SolveInstance, provider: &Arc<CudaKernelProvider>, ) -> Result<Self>

Host -> device upload helper for tests and tooling.

Production GPU-native paths should build GpuCnf directly on device.