skills.rest
igamenovoer
igamenovoerCOMMUNITY
Explore all skills by this author →
GitHub

krnopt-cuda-structural-optimization

Plan large-scale CUDA kernel redesigns by addressing structural bottlenecks.

Updated May 27, 2026
Search Tags:#profiling#cuda#performance-analysis#kernel-design#structural-optimization#redesign-patterns
One-click install
npx skills add https://github.com/igamenovoer/domain-skills --skill krnopt-cuda-structural-optimization
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: krnopt-cuda-structural-optimization
Source: https://github.com/igamenovoer/domain-skills/tree/main/domain/cuda/krnopt-cuda-structural-optimization
Command: npx skills add https://github.com/igamenovoer/domain-skills --skill krnopt-cuda-structural-optimization

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Replan and redesign CUDA kernels from a big-picture view. Use when the user wants a structural change to a kernel or hot path, such as rethinking boundaries, scheduling, primitives, pipeline stages, or metadata flow, instead of pursuing another local optimization. Do targeted idea search or source research before guessing when deeper knowledge is needed, and route workload-domain pattern selection to krnopt-cuda-domain-optimization. Do not use this skill for hotspot finding, raw profiler interpretation, or low-level CUDA coding.

Core Features & Use Cases

  • Big-picture structural assessment: identify whether the hot path's problem is boundary, scheduler, primitive, pipeline, or metadata.
  • Structural redesign guidance: select an appropriate redesign family (boundary elimination, persistent scheduling, grouped GEMM, fusion, etc.) based on the mismatch taxonomy.
  • Integrated workflow alignment: connect profiling insights to a planned redesign and ensure target architecture considerations are accounted for in the plan.

Quick Start

Analyze the current kernel hot path to decide if a structural redesign is warranted and specify the redesign family to pursue in the next step.

Frequently Asked Questions about krnopt-cuda-structural-optimization

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I plan a large-scale CUDA kernel redesign?▼

Plan large-scale CUDA kernel redesigns by identifying structural bottlenecks like boundaries, scheduling, primitives, or pipeline stages rather than pursuing local optimizations. This approach requires a defined problem, workload shape, and measurements to determine the appropriate redesign family and validation criteria.

What is structural optimization in CUDA kernel design?▼

Structural optimization in CUDA kernel design rethinks big-picture architecture, targeting boundary elimination, persistent scheduling, grouped GEMM, or fusion. It resolves mismatches in irregular workloads, scheduling concerns, or data flow bottlenecks across prototypes and production code.

When do I need a CUDA kernel structural redesign instead of local optimization?▼

You need a CUDA kernel structural redesign when facing hot paths with irregular workloads, scheduling concerns, or data flow bottlenecks that local optimizations cannot resolve. This applies when rethinking boundaries, primitives, pipeline stages, or metadata flow is necessary.

What do I need to assess a CUDA hot path for a structural redesign?▼

Assessing a CUDA hot path for a structural redesign requires a defined problem, workload shape, and measurements. You must identify whether the issue is boundary, scheduler, primitive, pipeline, or metadata related to select the correct redesign family.

Can I use this approach for low-level CUDA coding and hotspot finding?▼

No, this approach is not for low-level CUDA coding, hotspot finding, or raw profiler interpretation. It focuses exclusively on planning structural changes and selecting redesign families for hot paths, routing workload-domain pattern selection to specialized domain optimization.

Related Skills

Explore More Software Engineering Skills→
igamenovoerigamenovoer

krnopt-hw-aware-optimization

Identify GPU compute capability and workload shape for CUDA kernel optimization.

Community
Advanced
KohlJaryKohlJary

mind-palace

Query Mind Palace YAML entities for codebase subsystem knowledge.

Community
Intermediate
garethdainegarethdaine

laravel-architecture

Generates a Laravel 12 architecture with Actions, Inertia.js, Octane, Redis, and PostgreSQL.

Community
Intermediate
dyyounggdyyoungg

sglang-diffusion-benchmark-profile

Profile diffusion denoise latency and hotspots using the SGLang backend.

Community
Advanced