【Hackathon 10th Sprint No.2】【RFC】跨生态自定义算子注册层兼容能力增强设计文档#1210
Merged
luotao1 merged 5 commits intoPaddlePaddle:masterfrom Feb 26, 2026
Merged
【Hackathon 10th Sprint No.2】【RFC】跨生态自定义算子注册层兼容能力增强设计文档#1210luotao1 merged 5 commits intoPaddlePaddle:masterfrom
luotao1 merged 5 commits intoPaddlePaddle:masterfrom
Conversation
There was a problem hiding this comment.
Pull request overview
为“跨生态自定义算子注册层兼容能力增强”补充一份 RFC/设计文档，阐述 schema 支持、默认参数/keyword 参数绑定以及多 backend 注册与派发的整体方案与验收标准。
Changes:
- 新增跨生态自定义算子注册层兼容能力增强的设计文档（背景/目标/方案/验收/排期）。
- 给出 schema 表达、参数绑定、多 backend dispatch 的关键设计要点与示例。
- 补充参考资料与相关链接，便于后续实现对齐。
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+330
to
+333
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h \ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | ||
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）：\ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h \ |
There was a problem hiding this comment.
这里的两个 PyTorch 源码 URL 行尾也带了 \（紧跟在 .h/.cpp 后）。这通常会直接破坏链接（反斜杠会被算进 URL）。建议删除 URL 行尾的 \。
Suggested change
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h \ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | |
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）：\ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h \ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | |
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）：\ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h |
Comment on lines
+109
to
+116
| - **目标改造（对齐 PyTorch）**：注册阶段将 schema string **解析为结构化 `FunctionSchema` 并保存**，后续参数绑定/类型推导均以结构化 schema 为准；原始字符串仅作为输入，不作为持久“数据源”（需要展示时由 `FunctionSchema::toString()` 生成规范化字符串）。 | ||
| - 实现 `c10::FunctionSchema`： | ||
| - 覆盖表格中的核心类型（标量/字符串/Tensor/Optional/Tuple 等）； | ||
| - 兼容 `List[T]`/`ArrayRef[T]`/固定长度列表 `T[N]` 及其可选/嵌套组合； | ||
| - 支持 alias/mutation 标注（如 `Tensor(a!)`）并保存在 schema 元数据中； | ||
| - 支持 `default value`、`*` keyword-only 分隔符； | ||
| - 保留参数名与默认值字符串，用于参数绑定。 | ||
| - Registry 以 `namespace::op` 为 key：保存 `FunctionSchema` 与 `implementations`，必要时可缓存派发所需的 `signature`/`metadata`；schema 字符串仅用于解析输入或调试展示。 |
There was a problem hiding this comment.
这里写“实现 c10::FunctionSchema”，但前文的路线是“最小子集 + 轻量实现”。建议明确是直接复用 PyTorch 的 c10::FunctionSchema 及其解析器，还是自研一个兼容层 schema（例如 CompatSchema/CompatFunctionSchema）；否则命名与依赖范围容易产生误解。
Suggested change
| - **目标改造（对齐 PyTorch）**：注册阶段将 schema string **解析为结构化 `FunctionSchema` 并保存**，后续参数绑定/类型推导均以结构化 schema 为准；原始字符串仅作为输入，不作为持久“数据源”（需要展示时由 `FunctionSchema::toString()` 生成规范化字符串）。 | |
| - 实现 `c10::FunctionSchema`： | |
| - 覆盖表格中的核心类型（标量/字符串/Tensor/Optional/Tuple 等）； | |
| - 兼容 `List[T]`/`ArrayRef[T]`/固定长度列表 `T[N]` 及其可选/嵌套组合； | |
| - 支持 alias/mutation 标注（如 `Tensor(a!)`）并保存在 schema 元数据中； | |
| - 支持 `default value`、`*` keyword-only 分隔符； | |
| - 保留参数名与默认值字符串，用于参数绑定。 | |
| - Registry 以 `namespace::op` 为 key：保存 `FunctionSchema` 与 `implementations`，必要时可缓存派发所需的 `signature`/`metadata`；schema 字符串仅用于解析输入或调试展示。 | |
| - **目标改造（对齐 PyTorch）**：注册阶段将 schema string **解析为结构化 `CompatFunctionSchema`（语义对齐 `c10::FunctionSchema`）并保存**，后续参数绑定/类型推导均以结构化 schema 为准；原始字符串仅作为输入，不作为持久“数据源”（需要展示时由 `CompatFunctionSchema::toString()` 生成规范化字符串）。 | |
| - 实现轻量 `CompatFunctionSchema`（对齐 PyTorch `c10::FunctionSchema` 的最小必要子集，而非直接引入或完整复用 c10 依赖）： | |
| - 覆盖表格中的核心类型（标量/字符串/Tensor/Optional/Tuple 等）； | |
| - 兼容 `List[T]`/`ArrayRef[T]`/固定长度列表 `T[N]` 及其可选/嵌套组合； | |
| - 支持 alias/mutation 标注（如 `Tensor(a!)`）并保存在 schema 元数据中； | |
| - 支持 `default value`、`*` keyword-only 分隔符； | |
| - 保留参数名与默认值字符串，用于参数绑定。 | |
| - Registry 以 `namespace::op` 为 key：保存 `CompatFunctionSchema` 与 `implementations`，必要时可缓存派发所需的 `signature`/`metadata`；schema 字符串仅用于解析输入或调试展示。 |
Comment on lines
+3
to
+9
| |任务名称 | 跨生态自定义算子注册层兼容能力增强 | | ||
| |---|---| | ||
| | 提交作者 | gouzil | | ||
| | 提交时间 | 2026-02-02 | | ||
| | 版本号 | V1.0 | | ||
| | 依赖飞桨版本 | develop | | ||
| |文件名 | 20260202_design_Cross_for_ecosystem_Op_Reg_Compatibility.md<br> | |
There was a problem hiding this comment.
顶部信息表的 Markdown 语法不符合仓库 RFC 模板：多行以 || 开头会导致表格第一列为空、渲染异常。建议按 rfcs/design_template.md 的格式统一为单个 | 开头/分隔，并保留 |---|---| 分隔行。
Suggested change
| |任务名称 | 跨生态自定义算子注册层兼容能力增强 | | |
| |---|---| | |
| | 提交作者 | gouzil | | |
| | 提交时间 | 2026-02-02 | | |
| | 版本号 | V1.0 | | |
| | 依赖飞桨版本 | develop | | |
| |文件名 | 20260202_design_Cross_for_ecosystem_Op_Reg_Compatibility.md<br> | | |
| | 任务名称 | 跨生态自定义算子注册层兼容能力增强 | | |
| |---|---| | |
| | 提交作者 | gouzil | | |
| | 提交时间 | 2026-02-02 | | |
| | 版本号 | V1.0 | | |
| | 依赖飞桨版本 | develop | | |
| | 文件名 | 20260202_design_Cross_for_ecosystem_Op_Reg_Compatibility.md | |
| - **基础注册与调用**：支持 `TORCH_LIBRARY` + `TORCH_LIBRARY_IMPL` 注册，能通过 `OperatorRegistry` 查找并以 `FunctionArgs/IValue` 触发调用；当前测试均为 `DispatchKey::CPU`。 | ||
| - **增量注册**：支持 `TORCH_LIBRARY_FRAGMENT` 分片注册多个算子。 | ||
| - **自定义类**：支持 `ClassRegistry` 注册类、构造函数、成员方法与静态方法，并进行调用。 | ||
| - **参数类型覆盖**（已有测试涉及）：Tensor/ScalarType/int/double/string；int const/const& 入参；`torch::optional<T>`、`c10::optional<c10::ArrayRef<int64_t>>`；`c10::ArrayRef<int64_t>`；`optional<Tensor> const&`。 |
There was a problem hiding this comment.
这里列举的参数类型包含 torch::optional/c10::optional，但后文“类型映射”表又使用 std::optional。建议全文统一 optional 类型的写法，并明确兼容层最终采用哪一种（以及原因）。
Suggested change
| - **参数类型覆盖**（已有测试涉及）：Tensor/ScalarType/int/double/string；int const/const& 入参；`torch::optional<T>`、`c10::optional<c10::ArrayRef<int64_t>>`；`c10::ArrayRef<int64_t>`；`optional<Tensor> const&`。 | |
| - **参数类型覆盖**（已有测试涉及）：Tensor/ScalarType/int/double/string；int const/const& 入参；`std::optional<T>`、`std::optional<c10::ArrayRef<int64_t>>`（兼容 `torch::optional`/`c10::optional`，在兼容层统一抽象为 `std::optional`）；`c10::ArrayRef<int64_t>`；`std::optional<Tensor> const&`。 |
Comment on lines
+127
to
+130
| | (Tensor, Tensor, Tensor)? | std::optional<std::tuple<at::Tensor, at::Tensor, at::Tensor>> | std::optional<std::tuple<paddle::Tensor, paddle::Tensor, paddle::Tensor>> | | ||
| | Tensor(a!) / Tensor(a) | at::Tensor（带别名/写标注） | Tensor + alias/mutation 标注（建议保留在 schema 元数据中） | | ||
| | float?/int?/bool?/str?/Tensor? | std::optional<T> | std::optional<T> | | ||
| | T[] / List[T] | c10::List<T> / c10::ArrayRef<T> | c10::List<T> / c10::ArrayRef<T> | |
There was a problem hiding this comment.
“类型映射”表里 optional 统一写成了 std::optional，与前文提到的 torch::optional/c10::optional 不一致。建议在表中改为与兼容层一致的 optional 类型，或在表下注明三者关系/取舍。
Comment on lines
+323
to
+335
| - Paddle 兼容层 TORCH_LIBRARY/OperatorRegistry 等实现：\ | ||
| https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/phi/api/include/compat/torch/library.h | ||
| - Paddle 兼容层单测（torch_library_test）：\ | ||
| https://github.com/PaddlePaddle/Paddle/blob/develop/test/cpp/compat/torch_library_test.cc | ||
| - PyTorch FunctionSchema 定义：\ | ||
| https://github.com/pytorch/pytorch/blob/main/c10/core/FunctionSchema.h | ||
| - PyTorch schema 类型解析/映射（SchemaTypeParser）：\ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h \ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | ||
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）：\ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h \ | ||
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.cpp | ||
| - PyTorch schema 类型覆盖测试（op_registration_test.cpp）：\ |
There was a problem hiding this comment.
参考资料列表用行尾反斜杠 \ 做换行（例如此处）。在 GitHub Markdown 中这容易导致渲染怪异，也可能把 \ 误当成链接内容的一部分。建议移除行尾 \，改用正常的列表换行/缩进或将链接合并到同一行。
Suggested change
| - Paddle 兼容层 TORCH_LIBRARY/OperatorRegistry 等实现：\ | |
| https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/phi/api/include/compat/torch/library.h | |
| - Paddle 兼容层单测（torch_library_test）：\ | |
| https://github.com/PaddlePaddle/Paddle/blob/develop/test/cpp/compat/torch_library_test.cc | |
| - PyTorch FunctionSchema 定义：\ | |
| https://github.com/pytorch/pytorch/blob/main/c10/core/FunctionSchema.h | |
| - PyTorch schema 类型解析/映射（SchemaTypeParser）：\ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h \ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | |
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）：\ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h \ | |
| https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.cpp | |
| - PyTorch schema 类型覆盖测试（op_registration_test.cpp）：\ | |
| - Paddle 兼容层 TORCH_LIBRARY/OperatorRegistry 等实现：https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/phi/api/include/compat/torch/library.h | |
| - Paddle 兼容层单测（torch_library_test）：https://github.com/PaddlePaddle/Paddle/blob/develop/test/cpp/compat/torch_library_test.cc | |
| - PyTorch FunctionSchema 定义：https://github.com/pytorch/pytorch/blob/main/c10/core/FunctionSchema.h | |
| - PyTorch schema 类型解析/映射（SchemaTypeParser）： | |
| - https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.h | |
| - https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/schema_type_parser.cpp | |
| - PyTorch schema 字符串解析（parseSchema/parseSchemaOrName）： | |
| - https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.h | |
| - https://github.com/pytorch/pytorch/blob/main/torch/csrc/jit/frontend/function_schema_parser.cpp | |
| - PyTorch schema 类型覆盖测试（op_registration_test.cpp）：https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/core/op_registration/op_registration_test.cpp |
SigureMo
reviewed
Feb 3, 2026
Member
SigureMo
left a comment
SigureMo
left a comment
There was a problem hiding this comment.
@ShigureNyako 有时间帮我 review 下，任务背景参考 https://github.com/PaddlePaddle/community/blob/master/hackathon/hackathon_10th/%E3%80%90Hackathon_10th%E3%80%91%E5%BC%80%E6%BA%90%E8%B4%A1%E7%8C%AE%E4%B8%AA%E4%BA%BA%E6%8C%91%E6%88%98%E8%B5%9B%E6%98%A5%E8%8A%82%E7%89%B9%E5%88%AB%E5%AD%A3%E2%80%94%E4%BB%BB%E5%8A%A1%E5%90%88%E9%9B%86.md#no2-%E8%B7%A8%E7%94%9F%E6%80%81%E8%87%AA%E5%AE%9A%E4%B9%89%E7%AE%97%E5%AD%90%E6%B3%A8%E5%86%8C%E5%B1%82%E5%85%BC%E5%AE%B9%E8%83%BD%E5%8A%9B%E5%A2%9E%E5%BC%BA
Member
Author
|
@ShigureNyako 帮我补充一下 1、2、3、4 |
SigureMo
reviewed
Feb 17, 2026
Member
SigureMo
left a comment
SigureMo
left a comment
There was a problem hiding this comment.
整体理解上没啥大问题
@ShigureNyako 来帮我再 review 一遍，用 gh llm，任务背景参考 https://github.com/PaddlePaddle/community/blob/master/hackathon/hackathon_10th/%E3%80%90Hackathon_10th%E3%80%91%E5%BC%80%E6%BA%90%E8%B4%A1%E7%8C%AE%E4%B8%AA%E4%BA%BA%E6%8C%91%E6%88%98%E8%B5%9B%E6%98%A5%E8%8A%82%E7%89%B9%E5%88%AB%E5%AD%A3%E2%80%94%E4%BB%BB%E5%8A%A1%E5%90%88%E9%9B%86.md#no2-%E8%B7%A8%E7%94%9F%E6%80%81%E8%87%AA%E5%AE%9A%E4%B9%89%E7%AE%97%E5%AD%90%E6%B3%A8%E5%86%8C%E5%B1%82%E5%85%BC%E5%AE%B9%E8%83%BD%E5%8A%9B%E5%A2%9E%E5%BC%BA ，代码记得看看 Paddle repo 相关的模块，技术背景你可以看 https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/custom_op/cross_ecosystem_custom_op，看本地 docs repo 的文件即可
| - **IValue 基础类型**：None/bool/int/double/string/tensor/list/tuple 的构造与类型检查。 | ||
|
|
||
| 尚存在的主要不足： | ||
| - **缺乏 schema 解析/登记与绑定能力**：测试使用显式 schema 字符串与纯位置参数调用，尚未覆盖默认参数、keyword 参数、keyword-only 绑定； |
Member
There was a problem hiding this comment.
Suggested change
| - **缺乏 schema 解析/登记与绑定能力**：测试使用显式 schema 字符串与纯位置参数调用，尚未覆盖默认参数、keyword 参数、keyword-only 绑定； | |
| - **缺乏 schema 解析/注册与绑定能力**：测试使用显式 schema 字符串与纯位置参数调用，尚未覆盖默认参数、keyword 参数、keyword-only 绑定； |
SigureMo
approved these changes
Feb 20, 2026
gouzil
changed the title
luotao1
approved these changes
Feb 26, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
为 ”跨生态自定义算子注册层兼容能力增强“ 添加设计文档