What is iket-profiling?

A profiler for CuTe DSL GPU kernels that records named events and time ranges inside the running kernel. CuTe DSL is NVIDIA's Python-based way to describe GPU kernels; the results can be opened as a Perfetto trace or JSON.

Use it while developing or tuning a CuTe DSL kernel to inspect per-warp timelines and the timing of setup, main computation, finalization, and pipeline events.

It shows timing and activity within GPU phases, including work shared between producers and consumers, rather than only showing total kernel time. This helps locate time spent in pipeline and synchronization work.

iket-profiling — Skill for Claude Code and Codex by humanfia

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent

npx skills add humanfia/iket-profiling-skill --skill iket-profiling

Clone the repo

git clone --depth 1 https://github.com/humanfia/iket-profiling-skill

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

README.md

[![agentmods](https://agentmods.dev/badge/skills/humanfia/iket-profiling-skill/iket-profiling/github.svg)](https://agentmods.dev/skills/humanfia/iket-profiling-skill/iket-profiling)

Your own site

<a href="https://agentmods.dev/skills/humanfia/iket-profiling-skill/iket-profiling"><img src="https://agentmods.dev/badge/skills/humanfia/iket-profiling-skill/iket-profiling/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

Your own site · 80×15

<a href="https://agentmods.dev/skills/humanfia/iket-profiling-skill/iket-profiling"><img src="https://agentmods.dev/badge/skills/humanfia/iket-profiling-skill/iket-profiling.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 134 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 2,425 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5.1	$0.00134	$0.02425
Opus 5.5	$0.00054	$0.00970
Sonnet 5.5	$0.00027	$0.00485
Haiku 4.5	$0.00013	$0.00243

Measured 2d ago against content hash 9a920370c055, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

iket-profiling scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/iket-profiling/SKILL.md · 139 lines

How it starts

The opening of the file, as written. The whole thing — 139 lines — stays where its author put it; the contents beside it link to each section on GitHub.

IKET profiling for CuTe DSL kernels

IKET lets CuTe DSL kernels emit named markers and ranges from device code, like NVTX but inside the kernel. The run-iket profiler, released with nvidia-cutlass-dsl, collects them into a Perfetto trace (*.pftrace, open at https://ui.perfetto.dev/) and JSON. IKET is experimental: the API, output format, and overhead may change. The source of truth is the official guide: https://docs.nvidia.com/cutlass/latest/media/docs/pythonDSL/guides/iket_profiling.html

Workflow

Check the profiler is available: run-iket --help.

Instrument the @cute.kernel function. Calls in host-side code (@cute.jit functions, launch wrappers) do not emit in-kernel events.

import cutlass
import cutlass.cute as cute

@cute.kernel
def kernel(gA: cute.Tensor, gB: cute.Tensor, gC: cute.Tensor):
    bidx, _, _ = cute.arch.block_idx()
    cute.experimental.iket.mark("kernel_start", bidx)

    load_token = cute.experimental.iket.range_start("load")
    # Load data from gA and gB.
    cute.experimental.iket.range_end(load_token)

    cute.experimental.iket.range_push("compute")
    # Compute and store results.
    cute.experimental.iket.range_pop()

Run the workload under run-iket. The workload command goes after --. The kernel must be JIT-compiled inside this process; a kernel that was already compiled and is reused gets no instrumentation.
```
run-iket --output-dir ./iket_output --clobber profile --postprocess all -- python my_kernel.py
```
--postprocess perfetto|json|all selects the output. For large grids, add --enabled-cluster X,Y,Z to dump only one cluster (a non-cluster kernel counts as a 1x1x1 cluster, so this selects one thread block). Use --enabled-cluster-config FILE.json to choose per kernel: a JSON object mapping a substring of the mangled kernel name to [x, y, z], null, or "all".
Open the *.pftrace in Perfetto. Pan and zoom with W/A/S/D. Tracks are grouped by GPU location, then CTA and warp. WarpLifeTime tracks are automatic; each token-range name has its own track; push/pop ranges are on StackedRanges; markers are on Marker.
Start with a few coarse ranges (one whole-kernel range_start/range_end, plus setup, mainloop, epilogue), then add detail only where the trace shows it is needed.

Read the full file on GitHub · 139 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

2d ago First seen · 139 lines · 134 tokens per session scan A 9a920370c055

Subscribe to this mod's changes

iket-profiling is a skill published in the GitHub repository humanfia/iket-profiling-skill (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 134 tokens to every session and 2,425 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-05.

Analytics and metrics GPU Scientific agent-skills cuda cute-dsl cutlass

kernel-profile

Standalone kernel profiling skill for cuda-cpp, cute-dsl, cutlass, and triton implementations. Checks CUDA/PyTorch/Triton/CuTe DSL/CUTLASS/NCU/nsight-python readiness, optionally locks GPU clocks, validates correctness, collects Nsight Compute metrics with nsight-python, produces envcheck.md, correctness.md…

fmh66/kernel-opt-agent · 168 tokens

iket-profiling

Getting it into your agent

Wrote this? Show the measurements

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

IKET profiling for CuTe DSL kernels

Workflow

What this file has done since we first saw it

Other skills, from other repositories

kernel-profile

cuopt-developer

cuopt-user-rules

pytorch-profile-analysis

nsys-profile-analysis

cuopt-multi-objective-exploration