analyze-kernel-bottleneck
Systematically identify whether a GPU kernel is compute-bound, memory-bound, or latency-bound using roofline analysis, occupancy calculations, compute/load ratio per tile, and SASS instruction inspection. Produces a decision matrix for optimization strategy selection (cp.async, warp interleaving, tiling, double-buffering, or CuAssembler hand-tuning).
By pjt222, quoted from its SKILL.md as written.
What it is
- Repository
- pjt222/agent-almanac
- Folder
- skills/analyze-kernel-bottleneck
- Category
- Data and analysis
- Licence
- MIT
- Repository last updated
- Nov 6, 2026
- Copies found in other repositories
- 9
- Scripts beside the SKILL.md
- 0
What its code touches
This skill is instructions only. It ships no scripts, so there is no code to run.
Read from the text of its scripts, so this shows what the code can do, not what it does on any one run. Nothing was run.
Checks
- YesReads cleanly. Its header carries a name and a description.
- YesLicence known. A licence file names the terms: MIT.
- YesWhat it touches is recorded. Read from its scripts on Nov 8, 2026.
- Not yetInstalls. The install test in a clean, offline sandbox has not run on this skill yet.
- YesStill maintained. Its repository changed 2 days before it was read.
The Monitor indexes this skill and links to it. The files live in the author's repository, and the name and description are the author's own words. An author can correct or remove this page through Right of reply.