triton-npu-optimize-knowledge-v3
Generic reference-only optimize knowledge (v3 working copy) for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts; prefer this tree for ongoing knowledge updates.
- License 许可
- Apache-2.0
- 所属仓库
- Piagari/helix
- 最近更新
- 2026-07-16
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- triton-npu-optimize-knowledge-v3
- description
- Generic reference-only optimize knowledge (v3 working copy) for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts; prefer this tree for ongoing knowledge updates.
02 资源文件 29
- 参考文档references/pattern_index.md29.2 KB
- 参考文档references/patterns/algebraic-optimization.md11.5 KB
- 参考文档references/patterns/attention-cv-pipeline.md6.2 KB
- 参考文档references/patterns/autotune.md5.8 KB
- 参考文档references/patterns/cache_use.md3.8 KB
- 参考文档references/patterns/classic-matmul.md5.3 KB
- 参考文档references/patterns/compile_hint.md5.4 KB
- 参考文档references/patterns/diagonal.md7.9 KB
- 参考文档references/patterns/discrete_memory_access.md3.3 KB
- 参考文档references/patterns/gather-load.md4.0 KB
- 参考文档references/patterns/grid-flatten-and-ub-buffering.md4.9 KB
- 参考文档references/patterns/layout-store-and-block-pointers.md4.0 KB
- 参考文档references/patterns/loop-invariant-hoisting.md3.5 KB
- 参考文档references/patterns/padded_row_col_copy.md3.8 KB
- 参考文档references/patterns/parallel.md3.6 KB
- 参考文档references/patterns/program-multiple-rows.md4.1 KB
- 参考文档references/patterns/remove-implicit-transpose.md2.3 KB
- 参考文档references/patterns/reorder-load.md2.6 KB
- 参考文档references/patterns/scalar-latency-traps.md3.7 KB
- 参考文档references/patterns/slice_coalesce.md3.1 KB
- 参考文档references/patterns/slice_intermediate.md2.7 KB
- 参考文档references/patterns/software-pipeline.md4.1 KB
- 参考文档references/patterns/tiling.md5.9 KB
- 参考文档references/patterns/vec-cmp.md4.6 KB
- 参考文档references/symptom_index.md4.0 KB
- 参考文档references/symptoms/high-scalar-overhead.md928 B
- 参考文档references/symptoms/high-transfer-pressure.md886 B
- 参考文档references/symptoms/poor-locality.md862 B
- 参考文档references/symptoms/weak-pipeline-overlap.md847 B
03 SKILL.md 正文
Optimize Knowledge (v3)
Purpose
This skill is the generic optimize knowledge library for reusable pattern and symptom references. It is a copy of triton-npu-optimize-knowledge intended as the v3 update target; keep triton-npu-optimize-knowledge unchanged unless you are intentionally aligning the bases.
Scope
- This skill is reference-only.
- This skill does not define optimize workflow behavior.
- This skill does not own
opt-round-N/perf-analysis.md,attempts.md,summary.md, oropt-note.md. triton-npu-optimizeowns optimize workflow and validation rules.triton-npu-analyze-round-performanceowns round-level performance diagnosis.- For evidence-to-pattern distillation, follow the
ascend-npu-distill-patternsskill rather than inventing a local process. - During synthesis rewrites, preserve valid pre-existing pattern knowledge (examples, technique catalogs, implementation notes) unless round evidence clearly invalidates it.
Reading Order
- For code-structure-first triage, read
references/pattern_index.md. - For profile- or IR-backed routing, read
references/symptom_index.md. - Read only the one or two most relevant detailed cards after the index narrows the candidate set.
Reasoning Rules
- Treat pattern cards and symptom cards as routing aids, not a hard rule engine.
- Return to the caller skill for diagnosis, optimization choice, and recordkeeping.
- Keep specialized packs such as
triton-npu-cann-ext-api-patternsseparate unless the caller explicitly needs them.
04 同仓库更多 Skills
triton-npu-repair-guide
Heuristic fixes for Ascend Triton compile/JIT/kernel errors and numerical precision mismatches when editing or converting the operator. Does not replace ascend-npu-gen-test, ascend-npu-gen-bench, or harness specs.
triton-npu-optimize
Iteratively optimize a Triton Ascend NPU operator with correctness and performance gates. Use for operator optimization tasks that need repeated correctness validation, benchmark validation, multi-round experiment tracking, reusable optimization notes, and profiler-backed performance analysis when benchmark results need deeper explanation.
triton-npu-optimize-knowledge
Generic reference-only optimize knowledge for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts.
triton-npu-optimize-knowledge
Generic reference-only optimize knowledge for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts.
triton-npu-convert-pytorch-operator
Convert one PyTorch operator into a Triton NPU-backed PyTorch operator, preserve the trailing input-helper block, and validate the converted output through standalone or differential testing.
triton-npu-cann-ext-api-patterns
A5-only specialized optimization pattern references for Triton Ascend NPU kernels that use CANN Triton extension APIs. This skill does not define the optimize workflow; it only provides pattern material for optimize to consume.