llvm-clang-build-tuning

name	llvm-clang-build-tuning
description	Use when demonstrating or tuning a compile-heavy LLVM/clang build workload with sched-claw. Covers the demo launcher, direct build metrics, and the fallback path to IPC/CPI proxies.
aliases	["llvm-build-demo","clang-build-tuning"]
tags	["sched-claw","llvm","clang","build","scheduler"]

LLVM/Clang Build Tuning

When to use

• Demonstrate sched-claw on a compile-heavy LLVM/clang workload.
• Tune scheduler behavior for a large parallel build where overall wall-clock build time is the main objective.

Role

This is a workload profile, not a foundational workflow skill. It should supply workload-specific selectors, direct metrics, and likely scheduler levers, then compose the generic skills that do the actual contract, collection, analysis, policy mapping, codegen, and rollout work.

Compose with

• sched-workload-contract
• sched-perf-collection
• sched-perf-analysis
• sched-policy-mapping
• sched-ext-codegen
• sched-ext-run-evaluation
• sched-ext-rollout-safety

Workload-specific facts

• Primary direct metric: build_seconds:minimize
• Secondary direct metric: configure_seconds:minimize
• Preferred fallback proxies: ipc:maximize, cpi:minimize
• Common policy starting points: locality, migration control, and queue balance

Entrypoints

• Demo wrapper:
  • apps/sched-claw/scripts/demos/llvm-clang-autotune.sh
• Workload launcher:
  • apps/sched-claw/scripts/workloads/run-llvm-clang-build.sh

Guardrails

• Keep direct build timing primary unless the workload context explicitly says it is contaminated.
• Keep rollout windows bounded because large LLVM builds can run for a long time.

Rules

• Do not pretend IPC/CPI are direct business metrics. They are fallback proxies here.
• Keep the workload replayable through the launcher script instead of ad hoc shell commands.
• Preserve the artifact path used by the launcher so the agent can read build.log, configure.log, and metrics.env.

Reference Material

• references/demo-contract.md