CMOC optimization workloads
Authored by Brian Holman. Designed by Brian Holman · Sponsored by Dezota LLC. Code is GPL-3.0-or-later; see LICENSE. CMOC and other upstream tools retain their own notices.
The source supplies eight bounded CPU workloads and independent Python output models. The retained study compared Debug, Release and explicit frame-pointer omission using three seeds, two repeats, two host compiler slices and three guest modes. The measured numerical records describe the original execution; these public runner adaptations have not been represented as another study.
Supply the public VCCPlusOSS source tree, its matching built libVCCCore.a, a source-built supported CMOC runtime and your own read-only system ROM. The adapted runner hashes source files before and after execution, without requiring any private Git repository or history. It writes a new result directory.
python3 benchmark.py --runtime /path/to/Toolchain \
--vcc-source /path/to/VCCPlusOSS --core /path/to/libVCCCore.a \
--rom /path/to/coco3.rom --host-arches arm64 x86_64 \
--output /path/to/new-benchmark-runThe driver selects the actual compiler executable slice and retains its data alias, optimization flags, independent full-byte output checks and three-NOP cycle calibration. An arm64 runner executes guests serially. On Apple Silicon, x86_64 compiler execution is Rosetta translation, not physical Intel testing. Device-themed cases use synthetic buffers, not actual device I/O or whole-game frame rate. Generated payload size includes harness and linked runtime code.
Run python3 -m unittest test_benchmark.py for the
evidence-accounting tests. That check validates the accounting rules; it
does not execute guest programs. No ROM, compiler executable,
application binary, archived program or raw log is distributed in this
source selection. The public main branch evolves; recorded input hashes,
rather than a mutable URL alone, identify measured bytes.