git:20260911.9f92063 to git:20260912.5d83694
8 added, 4 removed. Audit A to A.
---
name: ascend-triton-kernel-optimization
description: Profile and iteratively optimize a correctness-passed Ascend Triton kernel with explicit NPU baselines, per-shape measurements, UB live-set and physical-core reasoning, MTE/Vector/Scalar bottleneck attribution, one-hypothesis rounds, noise-aware KEEP/DISCARD decisions, and Run Manifest evidence. Use for single-kernel latency or throughput improvement after all planned correctness cases pass. Do not use to create or migrate the first correct kernel, bypass failed validation, assess whole-model serving regressions, attribute model HBM, or diagnose a non-Triton operator.
---
# ascend-triton-kernel-optimization
Optimize an already validated Ascend Triton kernel using measured bottlenecks and repeatable latency evidence.
Choose one bottleneck hypothesis per round. Consider UB live set, physical cores and MTE/Vector/Scalar overlap. Compare repeated per-shape measurements with a reference baseline; retain a change only when the gain exceeds noise and correctness still covers the candidate.
## Agent entry
- Run from the repository root using the platform's Python launcher. The workspace
- selects its installed platform environment automatically.
+ Run from the repository root. The entry reuses the installed platform environment.
```text
- python .agents/skills/ascend-triton-kernel-optimization/scripts/triton_optimization.py --config optimization.json --results round-results.json
+ uv run --no-project python .agents/skills/ascend-triton-kernel-optimization/scripts/triton_optimization.py --config optimization.json --results round-results.json
```
The config contains op_name, kernel and its validation evidence, target, cases, baseline measurements and objective. Round results carry candidate measurements and validation. The report computes KEEP/DISCARD and verifies kernel lineage and case coverage.
Use ascend-triton-kernel-validation when correctness is incomplete. Use profiling-analysis for whole-model performance attribution.
+ The report aggregates supplied evidence. Imported numerical results and latency
+ measurements do not by themselves prove candidate NPU execution; this remains
+ unknown in the aggregate status. Keep the measured outcomes and use existing
+ runner or profiler evidence in the task assessment. That report limitation adds
+ no prerequisite for work already supported by valid execution evidence.
+
Read the relevant detail only when needed:
- - [behavior](references/behavior.md)
- [profiling decision tree](references/profiling-decision-tree.md)
- [ascend techniques](references/ascend-techniques.md)