Benchmarks

Benchmarked against equivalent C++ (g++ -O2) and Python 3.14. Best of 3 runs.

Visuall is within 1.1–2.8× of C++ on integer compute and loops, and 1.4× on deeply recursive code (Ackermann). The full benchmark suite runs ~1.9× slower than C++ overall — 20–50× faster than Python on numeric workloads. The escape analysis pass avoids GC heap allocation for non-escaping list, dict, tuple, and closure objects, eliminating GC overhead on allocation-heavy paths. Recent v1.3.2 GC optimizations (O(1) interior pointer resolution via chunk-indexed hash table) cut deep-stack GC pause times 86–133×.

Macro Benchmarks

Test C++ (ms) Visuall (ms) Python (ms) V / C++ Py / C++
Primes (trial div, 100K ×3) 6.9 12.7 241.6 1.8× 35×
TreeSum (recursive, depth 22) 8.2 35.6 645.9 4.3× 79×
Collatz (1..100K) 15.5 17.3 822.8 1.1× 53×
Strings (200K f-strings) 32.3 66.4 36.7 2.1× 1.1×
Pi (Leibniz, 10M terms) 10.7 15.1 781.0 1.4× 73×
Nested loops (2000×2000) 3.2 9.1 236.0 2.8× 74×
Ackermann (3,11) 851.6 1,192 26,918.0 1.4× 32×
GCD sum (1..2000) 45.8 51.1 863.8 1.1× 19×
Fibonacci (fib(35) ×500K) 5.7 7.3 728.4 1.3× 128×
Float distance (1M) 1.8 7.7 198.2 4.3× 110×

Micro Benchmarks

Test Iterations C++ Visuall vs C++ Python
Match dispatch 1M 7.5 ms 6.6 ms 0.9× 269 ms
Function call 1M 4.8 ms 11.0 ms 2.3× 252 ms
Tight loop sum 100K 5.9 ms 11.0 ms 1.9× 144 ms
Methodology

All tests run against equivalent C++ compiled with g++ -O2 and Python 3.14. Best of 3 runs reported. Test machine: Thinkpad E16 with Windows 11; these are README benchmarks. Benchmarks updated for Visuall v1.3.2.