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).

45 stars 4 forksProductivity & Notes

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add pjt222/agent-almanac --skill analyze-kernel-bottleneck
Manual install
git clone --depth 1 https://github.com/pjt222/agent-almanac.git /tmp/agent-almanac
mkdir -p ~/.claude/skills
cp -r /tmp/agent-almanac/i18n/caveman-lite/skills/analyze-kernel-bottleneck ~/.claude/skills/analyze-kernel-bottleneck

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# Analyze Kernel Bottleneck

Systematically identify whether a GPU kernel is compute-bound, memory-bound, or latency-bound by measuring baseline performance, classifying on the roofline, computing occupancy and compute/load ratio per tile, inspecting SASS instruction mix and stall codes, checking the shared memory cliff, and applying a decision matrix to select the right optimization strategy.

## When to Use

- Before optimizing any CUDA kernel -- establish baseline and classify bottleneck type
- After writing a first working version of a kernel to identify the optimization path
- When a kernel underperforms expectations relative to theoretical peak
- When deciding between cp.async, larger tiles, or algorithmic restructuring

## Inputs

- **Required**: Compiled kernel (`.cubin` or `.cu` source with build command)
- **Required**: Benchmark harness that launches the kernel with CUDA event timing
- **Required**: Problem dimensions (e.g., M, N, K for GEMM; seq_len, heads, head_dim for attention)
- **Optional**: Target GPU architecture (default: GA104 / sm_86 / RTX 3070 Ti)
- **Optional**: Expected peak utilization percentage for comparison
- **Optional**: Prior profiling data (Nsight Compute reports)

## Procedure

### Step 1: Measure Baseline Performance

Run the kernel with CUDA events (`BenchTimer`), record time in milliseconds. Calculate effective throughput metrics:

1. **Compile** the kernel if not already built:
Read the full skill on GitHub

Details

Skill name
analyze-kernel-bottleneck
Path
i18n/caveman-lite/skills/analyze-kernel-bottleneck/SKILL.md
License
MIT
Repo updated
18 November 2026

FAQ

What does the Analyze Kernel Bottleneck skill do?

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).

How do I install the Analyze Kernel Bottleneck skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add pjt222/agent-almanac --skill analyze-kernel-bottleneck. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Analyze Kernel Bottleneck popular?

It lives in pjt222/agent-almanac, which has 45 GitHub stars and 4 forks, and is included in 1 curated skill list.

Related skills

Adapt Architecture

45

pjt222/agent-almanac

Execute structural metamorphosis using strangler fig migration, chrysalis phases, and interface preservation. Covers transformation planning, parallel running, progressive cutover, rollback design, and post-metamorphosis stabilization for system architecture evolution. Use when assess-form has classified the system as READY for transformation, when migrating from monolith to microservices, when replacing a core subsystem while dependents continue operating, or when any architectural change must be gradual rather than big-bang.

Software DevelopmentIn 1 list

Adaptic

45

pjt222/agent-almanac

Master skill composing the 5-step synoptic cycle for panoramic synthesis across multiple domains. Orchestrates meditate, expand-awareness, observe, awareness, integrate-gestalt, and express-insight into a coherent process that produces unified understanding rather than sequential compromise. Use when a problem genuinely spans 3+ domains and the interactions between domains matter more than depth in any one, when sequential analysis feels like compromise rather than integration, or before major architectural decisions affecting multiple stakeholders.

OtherIn 1 list

Add Puzzle Type

45

pjt222/agent-almanac

Scaffold a new puzzle type across all 10+ pipeline integration points in jigsawR. Creates the core puzzle module, wires it into the unified pipeline (generation, positioning, rendering, adjacency), adds ggpuzzle geom/stat layers, updates DESCRIPTION and config.yml, extends the Shiny app, and creates a comprehensive test suite. Use when adding a completely new puzzle type to the package or following the 10-point integration checklist to ensure nothing is missed end-to-end.

Testing & QAIn 1 list

Assess Context

45

pjt222/agent-almanac

AI context assessment — evaluating problem malleability, mapping structural rigidity versus flexibility, analyzing transformation pressure, and estimating capacity to adapt. Use when a complex task feels stuck and it is unclear whether to push through or pivot, before a significant approach change to assess whether the current reasoning structure can support it, when accumulated workarounds suggest the underlying approach may be wrong, or as a periodic structural health check during extended multi-step tasks.

Productivity & NotesIn 1 list

Worker Integration

48

frankxai/claude-skills-library

Worker–agent integration for intelligent task dispatch and performance tracking. Use when wiring agents to a worker pool, routing tasks to workers, or instrumenting worker performance.

Productivity & Notes

Attune

45

pjt222/agent-almanac

AI relational calibration — reading and adapting to the specific person you are working with. Goes beyond user-intent alignment (solving the right problem) to genuine attunement (meeting the person where they are). Maps communication style, expertise depth, emotional register, and implicit preferences from conversational evidence. Use at the start of a new session, when communication feels mismatched, after receiving unexpected feedback, or when transitioning between very different users or contexts.

Productivity & NotesIn 1 list

Browse the most popular Claude skills, all productivity & notes skills, or write your own with the skills generator.