bench

Unified benchmarks for HVM and Bend.

Each subdirectory under bench/ contains a single benchmark case. Files are named main.hvm (HVM), main.bend (Bend), or main.ts (native TS).

Benchmarks

• cnot_04 — Applies λ-encoded NOT 2^4 times via cloned composition (HVM)
• cnot_16 — Applies λ-encoded NOT 2^16 times via cloned composition (HVM)
• cnot_24 — Applies λ-encoded NOT 2^24 times via cloned composition (HVM)
• bignat_mul — Schoolbook big-integer multiplication over immutable U32 limb lists (Bend)
• edit_dist — Row-DP Levenshtein distance over immutable U32 sequences (Bend)
• gab_boyer — Tautology checking with term rewriting over 106 lemmas (Bend)
• gab_deriv — Symbolic differentiation of a polynomial, 5000 iterations (Bend)
• gab_tak — Takeuchi triply-recursive function on Nat (18, 12, 6) (Bend)
• gab_takl — Takeuchi variant using list length comparison (18, 12, 6) (Bend)
• fib_nat — Naive Fibonacci over unary naturals with unary addition and unary-to-U32 conversion (HVM/TS)
• gen_easy — Type-directed proof search generating a sort function, small inputs (HVM)
• gen_hard — Type-directed proof search generating a sort function, large inputs (HVM)
• gen_mul4k — Proof search generating a multiplication function (HVM)
• lambda_eval — Lambda-calculus normalization with de Bruijn terms and Church arithmetic (Bend)
• mat_mul — Nat-driven nested loops with U32 pseudo-random mixing (Bend)
• nqueens — N-Queens backtracking over U32 bitset masks (Bend)
• sort_merge — Merge sort over immutable U32 lists with pseudo-random inputs (Bend)
• trie_bitwise — Persistent bitwise trie insert/query/fold over U32 keys (Bend)
• seq_pingpong — Mutual recursion via boxed continuations, 5M bounces (HVM)
• spin_tree — Binary tree to depth 8 with spin loops at 256 leaves (HVM)
• tree_fold — Builds and folds a binary tree to depth 24 (Bend)
• u32_fib — Naive Fibonacci on U32 to n=44 (Bend)

Runner

Run benchmarks with the unified script:

./bench.ts --hvm-interpreted
./bench.ts --hvm-compiled
./bench.ts --ts-via-bunjs --ts-via-nodejs
./bench.ts --bend-via-bunjs --bend-via-nodejs
./bench.ts --bend-interpreted-via-bunjs --bend-interpreted-via-hvm-compiled

Available options:

• --bend-via-bunjs (bend-bun)
• --bend-via-nodejs (bend-node)
• --ts-via-bunjs (ts-bun)
• --ts-via-nodejs (ts-node)
• --bend-via-hvm-interpreted[-TN] (bend-hvmi[N])
• --bend-via-hvm-compiled[-TN] (bend-hvmc[N])
• --bend-interpreted-via-bunjs (bendi-bun)
• --bend-interpreted-via-nodejs (bendi-node)
• --bend-interpreted-via-hvm-interpreted[-TN] (bendi-hvmi[N])
• --bend-interpreted-via-hvm-compiled[-TN] (bendi-hvmc[N])
• --hvm-interpreted[-TN] (hvmi[N])
• --hvm-compiled[-TN] (hvmc[N])

Examples:

• ./bench.ts hvmi hvmi4 hvmc8 bend-bun
• ./bench.ts --hvm-interpreted-T4 --bend-via-hvm-compiled-T8

Command overrides:

• BEND_CMD (default: bend)
• HVM_CMD (default: hvm)
• BUN_CMD (default: bun)
• NODE_CMD (default: node)
• BEND_NODE_ENTRY (optional explicit script path for bendi-node)