Consolidate gale CPU and GPU optimization into a refactor plan.
npx skills add https://github.com/ianrgraham/gale --skill gale-v2-refactor
Please help me install this Agent Skill. Skill: gale-v2-refactor Source: https://github.com/ianrgraham/gale/tree/main/.claude/skills/gale-v2-refactor Command: npx skills add https://github.com/ianrgraham/gale --skill gale-v2-refactor
This Skill helps you consolidate gale's accumulated CPU and GPU optimization work into cleaner building blocks without sacrificing the measured performance edge or the discipline needed to keep it.
Use the gale v2 refactor skill to review the current design, align the refactor plan with docs/gale-v2-refactor.md, and propose the next implementation step without regressing the benchmark baseline.
High-intent search queries and answers about installing and using this skill.
To refactor GPU code without losing CUDA performance, consolidate CPU and GPU optimization work into cleaner building blocks while keeping benchmark baselines, profiling gates, and correctness checks intact throughout the solver migration.
The best way to split a dense multi-GPU solver is organizing the codebase around clearer Sim and Device abstractions while applying regression baselines and algorithm-matched comparisons to prevent misleading performance conclusions during the separation.
Profiling discipline matters when refactoring multigrid simulation code because it enforces regression baselines and algorithm-matched comparisons, preventing misleading performance measurements and incomplete validation during the transition from prototype to production structure.
Yes, you can restructure production-path validation workflows for multi-GPU solvers without regressing benchmarks by freezing benchmark baselines and applying profiling gates before proposing the next implementation step in the refactor plan.
Comparing mismatched algorithms during CUDA optimization refactoring produces misleading performance conclusions and invalidates correctness checks, which is why applying strict profiling discipline and algorithm-matched comparisons is required to maintain stable solver results.
pilotspaceBenchmark GPU-accelerated Moon workloads against CPU baselines in CUDA environments.
anishs1207Run automated browser benchmarks and compare timing metrics against baselines.
deandumProfile, benchmark, and detect N+1 queries to identify performance bottlenecks.
Quill-AgentPerforms fast k-NN search and clustering on dense vectors using FAISS index types.
