We compare three implementations of GenDB's generated query code on TPC-H (SF10): the original C++ generated by GenDB, optimized C++ refined by Claude Code (Opus 4.6), and a full Rust rewrite also by Claude Code (Opus 4.6). The process: select the best-performing C++ binary for each query from the GenDB run, then give Claude Code 5 iterations to analyze, profile, and improve each implementation — first for optimized C++ (aggressive compiler flags, madvise tuning, parallelized joins, thread count optimization), then for Rust (rayon parallelism, unsafe bounds-check elimination, memmap2 zero-copy I/O).
| Original C++ | Optimized C++ | Rust | |
|---|---|---|---|
| Q1 | 49.8 ms | 39.2 ms | 71.7 ms |
| Q3 | 25.0 ms | 26.0 ms | 52.5 ms |
| Q6 | 31.8 ms | 35.5 ms | 23.7 ms |
| Q9 | 85.4 ms | 64.4 ms | 101.9 ms |
| Q18 | 49.2 ms | 20.1 ms | 32.8 ms |
| Total | 241.2 ms | 185.2 ms (1.30x) | 282.6 ms |
Optimized C++ achieves a 1.30x speedup over the original, with Q18 showing the largest gain (2.44x) from parallelized join building. Rust wins on Q6 (zone-map scan with get_unchecked) but carries ~30ms per-query overhead from mmap page table setup, penalizing short queries. The Rust main_scan compute times are competitive with C++, suggesting the overhead is structural rather than algorithmic. We plan to introduce a dedicated Code Refiner agent to the pipeline, responsible for low-level, implementation-level optimizations — compiler flag tuning, memory access patterns, SIMD utilization, cross-language code generation — to automatically achieve these gains as part of the standard GenDB workflow.
Source code for all three implementations is available in output/deprecated/tpc-h/language-comparison/.