kobold-bench vs gnucobol-rs vs kobold-cuda

Benchmark harness for the gnucobol-rs hot path (cob_move COMP-3 <-> display) and kobold-data-shim record decode, with a parity re-check after every run. Performance never alters sealed semantics.

Faithful Rust port of GnuCOBOL packed-decimal (COMP-3), zoned, and display numeric byte semantics and the MOVE conversions between them, proven byte-identical against the GnuCOBOL 3.2 libcob oracle. Plus a PIC->field-model parser , DATA DIVISION record-layout engine, , COPY+REPLACING copybook expander, pure-Rust decimal arithmetic, VALUE initial-record images, PIC P-scaling, OCCURS DEPENDING ON physical-max layout, LEVEL-88 condition-name predicates, and SET-TO-TRUE byte construction. Not a compiler; not arithmetic.

GPU/CUDA-accelerated batch decoding of COBOL numeric fields (COMP-3 packed-decimal and zoned-decimal DISPLAY) for high-throughput legacy-data ingest. Scaffold; the CUDA kernels land later.