Benchmarking, profiling, performance experiments, datasets, runners, and analysis tooling for TotalCross across platforms.
This repository is the home for repeatable TotalCross performance workloads and the tooling used to run, package, validate, and analyze them. Product code and correctness regressions remain in the main TotalCross repository.
The lab may contain:
- benchmark applications and workload fixtures;
- dataset descriptors and manifests;
- cross-platform runners and packaging tools;
- result schemas and parsers;
- profiling and diagnostics helpers;
- durable performance-analysis documentation.
Generated benchmark results, logs, binaries, deployed packages, and temporary evidence should normally stay outside Git unless a specific artifact is intentionally maintained as a fixture.
benchmarks/— benchmark workloads and applications.datasets/— versioned dataset descriptors and usage contracts.runners/— cross-platform benchmark orchestration.schemas/— machine-readable result and dataset contracts.tools/— packaging, validation, download, and analysis utilities.
Datasets are versioned independently from benchmark code. Large payloads may be hosted outside Git while this repository keeps their descriptors, integrity metadata, and usage contract.
The first registered dataset is image-scroll/v1, used by the TotalCross image-rendering scroll/decode workloads.
See DATASETS.md and datasets/image-scroll/v1/README.md.
The first image-rendering investigation (P12) is complete. The lab captured the causal path from the recurring raster regression through the production-candidate validation that supported TotalCross PR #488.
The final production candidate was TotalCross
bdd27273ead29bab320cba43b9ebad27be9b87ad, measured with lab source
264b9fba8581f08f3c0e4f31529831b7270d2c95. Its static warm run recorded
18/18 cached-final hits, zero generic/smooth destination draws, and a 5.509 ms
paint tree. The repeated-scroll run recorded no interval above 100 ms across
9,720 measured frame intervals.
See reports/p12-production-candidate-final.md for the concise closure and the individual P12 reports for causal evidence. Generated packages and raw run outputs remain intentionally outside Git under ignored local-data paths.
Source code in this repository is licensed under the MIT License unless a file says otherwise.
Datasets are not automatically covered by the MIT License. Their provenance, access, and redistribution terms are documented separately in DATASETS.md.