Intel VTune & AMD uProf — Microarchitecture Profiling

What it does

This skill provides a practical, step-by-step guide to profiling CPU-bound applications using Intel VTune Profiler and AMD uProf. It explains how to collect hotspot and microarchitecture data, interpret IPC/CPI metrics, identify front-end/back-end bounds, measure memory bandwidth, and construct a roofline analysis. The instructions include CLI commands, recommended build flags for symbol visibility, and mapping between VTune and uProf metrics.

When to use it

Use this skill when diagnosing performance bottlenecks in native code (C/C++, Rust, Zig) or high-performance numerical workloads. It's appropriate for developers who need to find hotspots, understand pipeline stalls, reason about cache/memory behavior, or evaluate whether code is compute- or memory-bound. Also useful for preparing data for roofline models and for tuning multi-threaded or vectorized code.

What's included

• Scripts: none bundled with the SKILL.md, but extensive CLI snippets for VTune and uProf commands are provided in the body.
• References: links and suggested tools (vtune, AMD uProf, likwid-perfctr) and workflows for collecting and reporting profiling data.
• Instructions: clear workflows for setup, hotspot collection, microarchitecture exploration, memory-access analysis, and roofline construction, plus recommended metrics to inspect.

Compatible agents

Best used by agents with knowledge of native toolchains and CLI execution (agents that can recommend commands, parse profiler outputs, and suggest code-level optimizations). Works well with developer-focused assistants and CI automation that can run profiling steps.