cccl-benchmarking
Use when asked to build/run a CCCL benchmark (CUB, Thrust, or libcu++), measure performance, or compare benchmark results between branches/commits.
Source facts
- Repository
- NVIDIA/cccl
- Last source activity
- October 7, 2026 at 11:47
- Detected SKILL.md language
- English
- Stars
- 2,559
- Forks
- 540
Install options
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Review the source files
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Showing SKILL.md
SKILL.md
Source instructions · Read-only preview- name
- cccl-benchmarking
- description
- Use when asked to build/run a CCCL benchmark (CUB, Thrust, or libcu++), measure performance, or compare benchmark results between branches/commits.
Read `docs/cub/benchmarking.rst` before building, running, or comparing benchmarks. Despite its location under
`docs/cub/`, it applies to Thrust and libcu++ benchmarks too.
Always run benchmark binaries with `-d 0 --stopping-criterion entropy --md <filename>.md --json <filename>.json`, and
keep both the markdown and JSON files for later processing (e.g. with `nvbench_compare.py`).
View on GitHub