s-tir-tuning
Guides TVM S-TIR scheduling, transformations, dlight rules, meta-schedule tuning, builders, runners, and CPU/GPU tuning decisions.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- s-tir-tuning
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides TVM S-TIR scheduling, transformations, dlight rules, meta-schedule tuning, builders, runners, and CPU/GPU tuning decisions.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/meta-schedule.md3.3 KB
- 参考文档references/schedule-workflows.md2.9 KB
- 参考文档references/troubleshooting.md2.5 KB
- 脚本scripts/meta_schedule_import_probe.py1.3 KB
03 SKILL.md 正文
S-TIR Scheduling and Tuning
Use this route when a task involves tvm.s_tir.Schedule, schedule primitives,
transform passes, dlight/default schedules, meta-schedule, tune_tir, or
bounded tuning of TIR generated from Relax or TIRx.
Route
- Confirm TVM imports and the selected target is available. Use
../install-build/SKILL.mdfor package/build failures. - Decide whether the user needs manual scheduling, a transform pass, dlight, or meta-schedule tuning.
- For manual schedules and debug traces, read
references/schedule-workflows.md. - For
tune_tir, builder/runner/database/cost-model decisions, or bounded tuning runs, readreferences/meta-schedule.md. - Run
scripts/meta_schedule_import_probe.pybefore a tuning run to verify the API and optional dependencies. - Use
references/troubleshooting.mdforScheduleError, invalid traces, no-candidate tuning, database, XGBoost, runner/RPC, and backend issues.
API anchors
from tvm.s_tir import Schedule
from tvm.s_tir import meta_schedule as ms
sch = Schedule(mod, seed=None, debug_mask="none", error_render_level="detail", enable_check=True)
ms.tune_tir(mod, target, work_dir, max_trials_global, runner="local", database="json")
tune_tir returns a meta-schedule Database. Use tiny max_trials_global
values for smoke checks, a disposable work_dir, and a CPU target unless the
user has explicitly prepared a GPU/RPC backend.
Boundaries
- Relax model import and high-level pipelines:
../relax-compile/SKILL.md. - TIRx tile primitives and native kernel dispatch:
../tirx-kernels/SKILL.md. - RPC tracker/server setup for remote runners:
../rpc-deployment/SKILL.md. - Package/toolchain failures:
../install-build/SKILL.md.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).