菜单
Start here to profile a FlyDSL kernel on AMD CDNA (gfx950/MI350X). Router for the flyprof CLI: how to run the fixed capture->analyze->report workflow, read its JSON envelopes, and pick the right sub-skill. Use whenever asked to profile/analyze a kernel, find why a kernel is slow, build an optimization report, or capture an ATT trace for a FlyDSL kernel.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Analyze a captured FlyDSL kernel bundle with flyprof: read the roofline/occupancy (counters), tile advice, stall/bubble taxonomy (bubbles), and the ISA<->Python mapping (map). Use after `flyprof capture`, when you need to understand where a kernel spends its stall cycles and how the ISA lines up with the .py source.
Capture a rocprofv3 ATT (Advanced Thread Trace) for a FlyDSL kernel with flyprof, including kernel discovery, tile sizing, and handling empty/out-of-range dispatches. Use when you need to record a kernel's per-instruction trace before analysis, or when `flyprof capture` returned EMPTY_ATT / DISPATCH_OUT_OF_RANGE / NO_KERNEL_DISCOVERED.
Produce and act on a FlyDSL kernel optimization report with flyprof: synthesize the ranked bubble->FlyDSL-knob recommendations, read the ISA<->Python evidence, optionally apply one knob, and verify the change re-profiles better. Use when asked what to change to make a kernel faster, or to turn a captured bundle into an optimization plan.
Diagnose a failing flyprof command by its error code. Use when any flyprof command returns ok:false / a non-zero exit, to decide whether to stop, fix config, retry, or treat the result as a finding. Maps each error.code to a hard action so you don't loop.
基于 SOC 职业分类
| name | flyprof-usage |
| description | Start here to profile a FlyDSL kernel on AMD CDNA (gfx950/MI350X). Router for the flyprof CLI: how to run the fixed capture->analyze->report workflow, read its JSON envelopes, and pick the right sub-skill. Use whenever asked to profile/analyze a kernel, find why a kernel is slow, build an optimization report, or capture an ATT trace for a FlyDSL kernel. |
| allowed-tools | Bash(flyprof:*), Read, Write, Edit |
flyprof is the front half of FlyDSL kernel optimization: given a kernel name it
produces a complete, agent-readable diagnosis — every stall bubble, where it lives
in the .py, and which FlyDSL thread-level knob addresses it. It does not
optimize the kernel; it gives you the evidence to. Drive it the same way every time.
-f json (it's the default when
piped). Shape: {"ok": true, "command": ..., "bundle": ..., "data": {...}, "warnings": [...]}
or {"ok": false, "command": ..., "error": {"code", "exitCode", "message", "help", ...}}.
Read data/error — never scrape prose.runs/<kernel>-<ts> by default, or pass --bundle DIR); every later command takes
--bundle DIR and reads the prior artifacts (capture.json, bubbles.json, …).0 ok · 66 empty/
no-match (a result, not a bug) · 69 rocprofv3/GPU missing (STOP) · 75 timeout
(retry) · 78 bad config/missing build tree · 70 internal. See flyprof-triage.flyprof list -f json is the source
of truth and changes as the worktree moves.flyprof list # what can I profile?
flyprof doctor # can this box capture? (verdict: ready)
flyprof run <kernel> --gpu 0 # the whole pipeline, or step through it:
flyprof tile <k> --shape … --dtype … (GPU-free: tile/occupancy/roofline advice)
flyprof capture <k> --tag both --with-pmc (rocprofv3 ATT + PMC -> bundle)
flyprof counters <k> (roofline / occupancy / L2 / bound)
flyprof bubbles --bundle <dir> (stall taxonomy)
flyprof map --bundle <dir> (ISA <-> .py:line)
flyprof report --bundle <dir> (ranked bubble->knob recommendations)
flyprof bundle --bundle <dir> (canonical examples/<k>/ for the hub)
flyprof run is the convenience path; the per-command path is for when you want to
inspect or re-run one stage.
| You want to… | Sub-skill |
|---|---|
| capture an ATT trace / it returned empty or out-of-range | flyprof-capture |
| read counters / tile / bubbles / map on a captured bundle | flyprof-analyze |
| produce the optimization report and act on a recommendation | flyprof-report |
| a command failed — what does the error code mean | flyprof-triage |
Don't guess. Lean on the envelopes. If a number looks wrong, re-run the command
that produced it and read its confidence field before making a claim.