arith.cu | arith_add_*, arith_sub_*, arith_mul_*, arith_div_*, arith_mod_*, arith_abs_*, arith_neg_* | Arithmetic operations for is expressions (the operator that evaluates a numeric expression and binds the result) |
join.cu | hash_join_bucket_count_v2, hash_join_scatter_v2, hash_join_probe_v2, hash_join_semi, hash_join_anti, compute_composite_hash | Hash joins (the v2 kernels use a bucketed hash-table layout) plus composite hashing, which combines several key columns into one hash value |
pack.cu | pack_keys, pack_and_hash_keys, hash_packed_keys, gather_packed_rows, compare_packed_keys | Key packing (encoding a multi-column key into one contiguous byte string), hashing of packed keys, and utilities that operate on the packed rows |
dedup.cu | mark_unique_*, compact_rows | Removing duplicate rows, done by sorting first and then keeping one row per group |
filter.cu | filter_compare_*, compact_*_by_mask, mask_{and,or,not} | Filtering rows against a condition and stream compaction (dropping the rows a boolean mask rejects and closing the gaps) |
sort.cu | radix_histogram, radix_scatter_*, init_indices, apply_permutation_*, gather_keys_* | Stable radix sort (a sort that preserves the order of equal keys) and applying the resulting permutation to reorder rows |
groupby.cu | detect_group_boundaries, extract_group_keys, groupby_*, groupby_logsumexp_* | Aggregation over sorted rows; the logsumexp variant is the numerically stable way to sum values held in log space |
scan.cu | exclusive_scan_mask, count_mask, multiblock_scan_* | Prefix-sum operations (each output element is the running total of the inputs before it), used to compute output positions |
set_ops.cu | concat_{u32,bytes}, sorted_diff_mark | Set union and difference over sorted row sets |
circuit.cu | xgcf_forward_level, xgcf_backward_level_* | Evaluating a probabilistic program’s compiled arithmetic circuit (XGCF, xlog’s circuit form of the program) and computing gradients back through it with reverse-mode differentiation (the backpropagation used to train weights) |
sat.cu | sat_cdcl_solve, sat_check_model, sat_proof_check, sat_assert_*, sat_xgcf_cnf_*, sat_emit_not_phi | A GPU Boolean-satisfiability (SAT) verifier using CDCL (conflict-driven clause learning, the standard modern SAT search algorithm), plus helpers that build the formulas for checking whether two programs are equivalent |
mc_sample.cu | mc_sample_bernoulli | Drawing Bernoulli (biased-coin) samples for Monte Carlo inference, which estimates probabilities by averaging over many random samples |
