[Docs] Update Agent Skills for CV fusion operators 🤖#966
[Docs] Update Agent Skills for CV fusion operators 🤖#966liggest wants to merge 3 commits intotile-ai:ascendc_ptofrom
Conversation
|
👋 Hi! Thank you for contributing to the TileLang project. Please remember to run We appreciate you taking this step! Our team will review your contribution, and we look forward to your awesome work! 🚀 |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request significantly expands the TileLang-Ascend documentation, introducing best practices for CV fusion operators, matrix fractal limits, and workspace memory management. It also adds a comprehensive guide for PR readiness and troubleshooting. Review feedback pointed out a variable name inconsistency in a code example and a typo in the hardware error description.
|
/re-test |
|
❌ Task Failed Summary: No summary available Workflow run: View details |
|
/re-test |
|
❌ Task Failed Summary: No summary available Workflow run: View details |
|
/re-test |
|
✅ Task Completed successfully Summary: Total: 118 | Passed: 118 | Failed: 0 (Pass rate: 100%) Workflow run: View details |
|
/re-test |
|
❌ Task Failed Summary: No summary available Workflow run: View details |
liggest
force-pushed
the
example_dequant_gemm_w4a8_skills
branch
from
c3f1fe5 to
969d0a6
Compare
| - ✅ **1D 并行**：`for j in T.Parallel(N)` | ||
| - ✅ **2D 并行**：`for i, j in T.Parallel(M, N)` | ||
| - ✅ **serial + parallel 组合**：`for i in range(M): for j in T.Parallel(N)` | ||
| - ❌ **3D 或更高维并行**：不支持，会触发编译错误 |
There was a problem hiding this comment.
Support the 3D double buffer scenario. Actually, the computation is still performed only on 2D tiles; the highest dimension merely indicates which stream to operate on. Such as https://github.com/tile-ai/tilelang-ascend/blob/ascendc_pto/examples/elementwise/elementwise_add_pipeline.py#L71
There was a problem hiding this comment.
I assume the "3D 或更高维并行" in the document means for i, j, k ... in T.Parallel(M, N, K, ...).
It seems that the use case in elementwise_add_pipeline.py#L71 still fits the "2D 并行" case.
Perhaps the description is OK here I think 🤔
liggest
force-pushed
the
example_dequant_gemm_w4a8_skills
branch
from
969d0a6 to
085f1c7
Compare
- Add CV fusion operator development guidance - Add V-core parallelization best practices - Add workspace index consistency rules - Add GEMM fractal limits documentation - Add pr-ready-guide.md for pre-commit checklist - Update .gitignore for opencode artifacts
liggest
force-pushed
the
example_dequant_gemm_w4a8_skills
branch
from
af275e5 to
e9270b7
Compare
4 participants
Summary
本次 PR 更新 Agent Skills 文档，增强 CV 融合算子开发支持，补充 workspace 索引一致性、V 核并行化等最佳实践。
Changes
新增文档：
tilelang-op-generate/references/pr-ready-guide.md：上库前收尾工作完整指南（Golden 实现一致性检查、代码格式、注释转英文等）tilelang-api-best-practices 更新：
tilelang-expert-to-developer 更新：
tilelang-op-design 更新：
tilelang-op-generate 更新：
.gitignore 更新：
.arts/、.agents/reportsTesting
文档变更，无需运行测试。相关示例代码已在
examples/dequantize_gemm/中验证通过。