cuda

CUDA C++ (.cu) kernel reference — TVM-FFI direct-export vs Python-binding entry points, the chevron-launch null-stream pitfall under CUDA-graph capture, the sm_100 fp8→bf16 cvt PTX gap, the load_inline name-cache trap, and the __launch_bounds__ register-spill lever. Also the single-source for generic per-call-overhead / CUDA-graph-capture / PDL theory (Waves-Per-SM decision table) regardless of DSL. Use when writing or debugging hand-written CUDA C++ kernels, or whenever you need generic CUDA-graph / sync-audit / PDL reasoning.

跳到安装

仓库: TongmingLAIC/AKO4X
最近来源活动: 2026年5月31日 08:57
检测到的 SKILL.md 语言: 英语
星标: 67
分支: 12

安装方式

默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。

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。

直接命令不会经过审查 Prompt；运行前请先检查来源。

npx skills add https://github.com/TongmingLAIC/AKO4X --skill cuda

命令会保持在同一行。复制前请横向滚动并检查完整内容。

想先保存到本地？可下载 SkillsMP 当前能够提供的文件。

检查来源文件

决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。

文件资源管理器

2 个文件

SKILL.md

来源说明 · 只读预览

返回搜索返回安装

bench

Run performance benchmarks to get a verdict on whether a kernel change actually helped. Use --ab-compare for sub-1x deltas (drift cancels in same container), --variance-check for noise floor, and the subset modes (--first, --smoke; --extremes is modal-backend only) for compile-correctness probes only — those are NOT performance verdicts. The only authoritative source for "is this change real". Benchmark specifics (config.toml schema, status enum, scoring formula, baseline rule, fresh-inputs contract) live in the `benchmark` skill.

2026-05-31

benchmark

Reference for the active benchmark harness — what it IS and how it behaves (the active benchmark is flashinfer-bench). Covers config.toml structure (`[solution]`/`[build]`/`[benchmark]` tables), the status enum, the workload / fresh-inputs model, reference-baseline + scoring mechanics, tolerance keys, TVM-FFI builder link limits, and the silent-skip cascade. Invoke whenever you decode a bench status string, hit an unfamiliar config.toml field, suspect "correctness passed but the headline is implausible", or need to know what's frozen before proposing a bench edit — do NOT guess field or status semantics. Bench commands and noise methodology (A/B compare, variance check, drift cancellation) live in the `bench` skill.

2026-05-31

cpp

Host-side C++ kernels (`[build].language = cpp`) — the `solution/{kernel.cpp, binding.py}` two-file TVM-FFI binding pattern, and when to reach for it (host-side orchestration of a kernel you write, or composing CUTLASS headers into it). Distinct from the `cuda` skill which covers `.cu` device kernels compiled by nvcc. Use when writing a host-side C++ kernel or deciding cpp-vs-cuda.

2026-05-31

查看这个仓库的全部 Skills

name

cuda

description

CUDA

Reference for CUDA C++ kernel writers. Detailed guide: cuda.md.

When to consult

Writing a CUDA .cu kernel (TVM-FFI direct export or Python binding).
Resolving a RUNTIME_ERROR traceable to chevron-launch stream binding (legacy null-stream is NOT capture-aware — see cuda.md "Kernel-launch stream is NOT optional under CUDA graph capture").
Reducing per-call overhead: GPU↔CPU sync audit, CUDA-graph capture, or PDL kernel→kernel overlap — generic theory + the Waves-Per-SM decision table live here; per-DSL PDL bindings live in the DSL skills.

COUPLED references

None directly to runtime. Per-operator CUDA wins / traps live under docs/prior/ (when your operator has a prior archive).

cuda

来源信息

安装方式

检查来源文件

同仓库更多 Skills

同仓库更多 Skills

CUDA

When to consult

COUPLED references

CUDA

When to consult

COUPLED references