Menu
AI Kernel 算子生成与优化工作流程。当用户需要生成、验证或优化 kernel 算子时使用此 Skill。支持 Triton、CUDA C、C++、TileLang等多种后端 DSL。
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/mindspore-ai/akg --skill kernel-workflowThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
矩阵乘法矩阵乘法 A[M, K] @ B[K, N] = C[M, N]中，大K维度矩阵乘法(K>>M,N)优化：针对M/N较小但K极大(如M=N=256,K=131072)的场景，Split-K切分K维度并行化、Workspace+Reduce替代全局同步，实现显著性能提升
Triton Ascend hard API restrictions and forbidden syntax. MUST-follow rules that apply to every kernel: forbidden control flow (return/break/continue/lambda/while), tensor slice/index restrictions, scalar conversion rules, BLOCK_SIZE upper bound. Violating any of these produces a compile or runtime error on Ascend.
Triton Ascend 性能优化通用策略: BLOCK_SIZE 选择 (1024-2048 for elementwise, must be <65536), grid configuration (use VEC_CORE_NUM / CUBE_CORE_NUM, 2D/3D grid for matmul / conv / reduce, 1D grid + inner loop for elementwise / pointwise), 256B alignment for memory transfers, autotune block-size patterns, fp16 / fp32 precision conversion. Bind via keywords like matmul, elementwise, reduce, block_size, grid, autotune, alignment, fp16, fp32, tile, interleaved-loop, cube-core, vec-core.
Based on SOC occupation classification
Showing SKILL.md
| name | kernel-workflow |
| description | AI Kernel 算子生成与优化工作流程。当用户需要生成、验证或优化 kernel 算子时使用此 Skill。支持 Triton、CUDA C、C++、TileLang等多种后端 DSL。 |
references/torch_task_verify.mdreferences/tool-selection.mduser_requirements）finish注意：user_requirements 是可选参数。
call_op_task_builderclass Model(nn.Module)）→ 加载 references/torch_task_verify.md 进行验证@triton.jit / __global__ 等）→ 验证 task 代码后加载 references/tool-selection.md用户: 生成一个 ReLU 算子
Think: 用户没有提供代码，直接生成 task。
Action: call_op_task_builder(user_request="生成一个 ReLU 算子")
Observation: {status: "READY", task_code: "...", op_name: "relu"}
Think: task 已生成，加载 tool-selection 指导选择生成方式。
Action: read_file(file_path="resources/skills/kernel-workflow/references/tool-selection.md")
...
用户: 帮我生成这段代码的 kernel: class Model(nn.Module)...
Think: 用户提供了代码，加载验证指南。
Action: read_file(file_path="resources/skills/kernel-workflow/references/torch_task_verify.md")
Observation: <验证指南>
Think: 按指南验证代码格式。
Action: execute_script(script_path="resources/skills/kernel-workflow/scripts/check_torch_code.py", args="--stdin --json", stdin_input="<代码>")
...
用户: 帮我优化这段 Triton kernel: @triton.jit def kernel(...): ...
对应的 task 是: class Model(nn.Module)...
Think: 用户提供了 kernel 和 task，先验证 task。
Action: read_file(file_path="resources/skills/kernel-workflow/references/torch_task_verify.md")
...验证通过后...
Action: read_file(file_path="resources/skills/kernel-workflow/references/tool-selection.md")
...
references/torch_task_verify.md - task 代码验证与补全流程references/tool-selection.md - 子Agent 选择与二次确认流程scripts/check_torch_code.py - 验证 task 代码格式（参数：--stdin --json）call_op_task_builder