fix:修复HCCL软件目录重构后遗留问题-hccl-AtomGit

Mexperimental/ops/all_reduce/example/README.md+6-6

	@@ -26,7 +26,7 @@ A5 注册方式时选用的载体，本身不是本目录的产出目标。
26	## 1. 动机	26	## 1. 动机
27		27
28	- 验证 A5 注册方式在 experimental 下的可用性（主动机）：主链路算法（如	28	- 验证 A5 注册方式在 experimental 下的可用性（主动机）：主链路算法（如
29	- `src/ops/all_reduce/executor/ins_v2_all_reduce_sole_executor.cc` 的 `CcuMSAllReduceSoleMesh`）通过	29	+ `src/ops/all_reduce/algorithm/executor/ins_v2_all_reduce_sole_executor.cc` 的 `CcuMSAllReduceSoleMesh`）通过
30	`REGISTER_EXEC_V2` 宏注册进 `CollAlgExecRegistryV2`，运行时由 `GetAlgExec` 按算法名查找执行器。	30	`REGISTER_EXEC_V2` 宏注册进 `CollAlgExecRegistryV2`，运行时由 `GetAlgExec` 按算法名查找执行器。
31	本目录在experimental/ 下新增一个同样走 A5 注册的算法，以测试 A5 注册方式是否可用。	31	本目录在experimental/ 下新增一个同样走 A5 注册的算法，以测试 A5 注册方式是否可用。
32	- 以最小样例验证全链路：不修改主链路算法，而是在experimental 下复制同构的 executor/template/kernel 三层	32	- 以最小样例验证全链路：不修改主链路算法，而是在experimental 下复制同构的 executor/template/kernel 三层
	@@ -37,7 +37,7 @@ A5 注册方式时选用的载体，本身不是本目录的产出目标。
37	## 2. 目录结构与设计	37	## 2. 目录结构与设计
38		38
39	下图为以 `hccl/` 仓库根为起点的完整结构，省略的无关文件/目录统一用 `...` 标注。本目录结构与主链路	39	下图为以 `hccl/` 仓库根为起点的完整结构，省略的无关文件/目录统一用 `...` 标注。本目录结构与主链路
40	-`src/ops/all_reduce/`（`executor/` + `template/ccu/kernel/`）同构，证明相同的 A5 注册代码在	40	+`src/ops/all_reduce/`（`algorithm/executor/` + `algorithm/template/ccu/kernel/`）同构，证明相同的 A5 注册代码在
41	experimental 目录下可以原样工作；图中同时标注了与本目录直接相关的关键上下文（注册表、主链路	41	experimental 目录下可以原样工作；图中同时标注了与本目录直接相关的关键上下文（注册表、主链路
42	对照样例、父级 CMakeLists）。	42	对照样例、父级 CMakeLists）。
43		43
	@@ -46,10 +46,10 @@ hccl/ # 仓库根
46	├── CMakeLists.txt # 顶层：option(ENABLE_EXPERIMENTAL)；条件 add_subdirectory(experimental/ops/)	46	├── CMakeLists.txt # 顶层：option(ENABLE_EXPERIMENTAL)；条件 add_subdirectory(experimental/ops/)
47	├── build.sh # 编包入口：--experimental → -DENABLE_EXPERIMENTAL=ON	47	├── build.sh # 编包入口：--experimental → -DENABLE_EXPERIMENTAL=ON
48	├── src/ # 主链路（商用代码，位于 experimental 之外）	48	├── src/ # 主链路（商用代码，位于 experimental 之外）
49	-│ ├── ops/op_common/executor/registry/	49	+│ ├── ops/op_common/algorithm/executor/registry/
50	│ │ └── coll_alg_v2_exec_registry.* # 执行器注册表：REGISTER_EXEC_V2 写入 / GetAlgExec 查找	50	│ │ └── coll_alg_v2_exec_registry.* # 执行器注册表：REGISTER_EXEC_V2 写入 / GetAlgExec 查找
51	│ ├── ops/all_reduce/	51	│ ├── ops/all_reduce/
52	-│ │ ├── executor/ins_v2_all_reduce_sole_executor.cc # 主链路对照样例（注册	52	+│ │ ├── algorithm/executor/ins_v2_all_reduce_sole_executor.cc # 主链路对照样例（注册
53	│ │ └── ... # 其余 executor/template/kernel 文件	53	│ │ └── ... # 其余 executor/template/kernel 文件
54	│ └── ... # 其余 src 子模块	54	│ └── ... # 其余 src 子模块
55	├── include/ # 对外头文件（hccl.h / hccl_mc2.h 等）	55	├── include/ # 对外头文件（hccl.h / hccl_mc2.h 等）
	@@ -207,11 +207,11 @@ template 继承自引擎模板基类（以 `CcuAlgTemplateBase` 为例），须
207	\| `CalcScratchMultiple` \| 返回 scratch 倍率，供 executor 计算单 loop 数据上界 \| `inBuffType`/`outBuffType` 输入输出 buffer 类型 \|	207	\| `CalcScratchMultiple` \| 返回 scratch 倍率，供 executor 计算单 loop 数据上界 \| `inBuffType`/`outBuffType` 输入输出 buffer 类型 \|
208	\| `FastLaunch` \| 快速下发：用预存 `submitInfos` 改写地址后直发内核 \| `param`；`tempFastLaunchCtx` threads/ccuKernelSubmitInfos/buffInfo \|	208	\| `FastLaunch` \| 快速下发：用预存 `submitInfos` 改写地址后直发内核 \| `param`；`tempFastLaunchCtx` threads/ccuKernelSubmitInfos/buffInfo \|
209		209
210	-`CalcCostCoeff`（executor 与 template）对应基类 `CalcCostCoeffParam` 字段的完整说明可见 `src/ops/op_common/template/` 下模板头文件注释。	210	+`CalcCostCoeff`（executor 与 template）对应基类 `CalcCostCoeffParam` 字段的完整说明可见 `src/ops/op_common/algorithm/template/` 下模板头文件注释。
211		211
212	#### REGISTER_EXEC_V2 宏参数	212	#### REGISTER_EXEC_V2 宏参数
213		213
214	-宏定义见 `src/ops/op_common/executor/registry/coll_alg_v2_exec_registry.h`，签名：	214	+宏定义见 `src/ops/op_common/algorithm/executor/registry/coll_alg_v2_exec_registry.h`，签名：
215		215
216	```cpp	216	```cpp
217	REGISTER_EXEC_V2(type, name, insCollAlgBase, AlgTopoMatch, InsAlgTemplate)	217	REGISTER_EXEC_V2(type, name, insCollAlgBase, AlgTopoMatch, InsAlgTemplate)

Mexperimental/ops/all_reduce/example/README_en.md+6-6

Msrc/ops/op_common/algorithm/topo_match/CMakeLists.txt+0-3

	@@ -13,9 +13,6 @@ set(src_list
13	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_base.cc	13	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_base.cc
14	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_base_v2.cc	14	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_base_v2.cc
15	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_concurrent.cc	15	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_concurrent.cc
16	- # PhysicalLevel标准化: 不依赖版本宏, 因此放在无条件块
17	- ${CMAKE_CURRENT_SOURCE_DIR}/physical_level_build.cc
18	- ${CMAKE_CURRENT_SOURCE_DIR}/physical_level_normalize.cc
19	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_one_level.cc	16	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_one_level.cc
20	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_two_level.cc	17	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_two_level.cc
21	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_three_level.cc	18	${CMAKE_CURRENT_SOURCE_DIR}/topo_match_three_level.cc

Msrc/ops/op_common/topo_info/CMakeLists.txt+3-0

	@@ -11,6 +11,9 @@
11	set(src_list	11	set(src_list
12	${CMAKE_CURRENT_SOURCE_DIR}/topo.cc	12	${CMAKE_CURRENT_SOURCE_DIR}/topo.cc
13	${CMAKE_CURRENT_SOURCE_DIR}/topo_host.cc	13	${CMAKE_CURRENT_SOURCE_DIR}/topo_host.cc
		14	+ # PhysicalLevel标准化: 不依赖版本宏, 因此放在无条件块
		15	+ ${CMAKE_CURRENT_SOURCE_DIR}/physical_level_build.cc
		16	+ ${CMAKE_CURRENT_SOURCE_DIR}/physical_level_normalize.cc
14	)	17	)
15		18
16	if(TARGET hccl)	19	if(TARGET hccl)

Rsrc/ops/op_common/algorithm/topo_match/physical_level.h→src/ops/op_common/topo_info/physical_level.h+0-0

文件重命名但无更改。

Rsrc/ops/op_common/algorithm/topo_match/physical_level_build.cc→src/ops/op_common/topo_info/physical_level_build.cc+0-0

文件重命名但无更改。

Rsrc/ops/op_common/algorithm/topo_match/physical_level_normalize.cc→src/ops/op_common/topo_info/physical_level_normalize.cc+0-0

文件重命名但无更改。

	@@ -14,24 +14,24 @@ Besides this algorithm itself, Section 2.5 provides the general guide on "implem
14		14
15	## 1. Motivation	15	## 1. Motivation
16		16
17	-- Verify the usability of the A5 registration approach under experimental (primary motivation): Main-pathway algorithms (e.g., `CcuMSAllReduceSoleMesh` in `src/ops/all_reduce/executor/ins_v2_all_reduce_sole_executor.cc`) register into `CollAlgExecRegistryV2` via the `REGISTER_EXEC_V2` macro, and are looked up at runtime by `GetAlgExec` based on the algorithm name. This directory adds an algorithm under experimental/ that also follows A5 registration, to test whether the A5 registration approach is usable.	17	+- Verify the usability of the A5 registration approach under experimental (primary motivation): Main-pathway algorithms (e.g., `CcuMSAllReduceSoleMesh` in `src/ops/all_reduce/algorithm/executor/ins_v2_all_reduce_sole_executor.cc`) register into `CollAlgExecRegistryV2` via the `REGISTER_EXEC_V2` macro, and are looked up at runtime by `GetAlgExec` based on the algorithm name. This directory adds an algorithm under experimental/ that also follows A5 registration, to test whether the A5 registration approach is usable.
18	- Verify the full pathway with a minimal sample: Without modifying main-pathway algorithms, replicate the isomorphic executor/template/kernel three-layer structure under experimental/ as a carrier, verifying that every step from build to checker is usable.	18	- Verify the full pathway with a minimal sample: Without modifying main-pathway algorithms, replicate the isomorphic executor/template/kernel three-layer structure under experimental/ as a carrier, verifying that every step from build to checker is usable.
19		19
20	---	20	---
21		21
22	## 2. Directory Structure and Design	22	## 2. Directory Structure and Design
23		23
24	-The diagram below shows the complete structure starting from the `hccl/` repository root; unrelated files/directories are uniformly marked with `...`. This directory's structure is isomorphic to the main pathway `src/ops/all_reduce/` (`executor/` + `template/ccu/kernel/`), demonstrating that the same A5 registration code works unchanged under the experimental directory; the diagram also annotates key context directly related to this directory (the registry, the main-pathway reference sample, and the parent CMakeLists).	24	+The diagram below shows the complete structure starting from the `hccl/` repository root; unrelated files/directories are uniformly marked with `...`. This directory's structure is isomorphic to the main pathway `src/ops/all_reduce/` (`algorithm/executor/` + `algorithm/template/ccu/kernel/`), demonstrating that the same A5 registration code works unchanged under the experimental directory; the diagram also annotates key context directly related to this directory (the registry, the main-pathway reference sample, and the parent CMakeLists).
25		25
26	```plaintext	26	```plaintext
27	hccl/ # Repository root	27	hccl/ # Repository root
28	├── CMakeLists.txt # Top-level: option(ENABLE_EXPERIMENTAL); conditional add_subdirectory(experimental/ops/)	28	├── CMakeLists.txt # Top-level: option(ENABLE_EXPERIMENTAL); conditional add_subdirectory(experimental/ops/)
29	├── build.sh # Build entry: --experimental → -DENABLE_EXPERIMENTAL=ON	29	├── build.sh # Build entry: --experimental → -DENABLE_EXPERIMENTAL=ON
30	├── src/ # Main pathway (commercial code, outside experimental)	30	├── src/ # Main pathway (commercial code, outside experimental)
31	-│ ├── ops/op_common/executor/registry/	31	+│ ├── ops/op_common/algorithm/executor/registry/
32	│ │ └── coll_alg_v2_exec_registry.* # Executor registry: REGISTER_EXEC_V2 writes / GetAlgExec looks up	32	│ │ └── coll_alg_v2_exec_registry.* # Executor registry: REGISTER_EXEC_V2 writes / GetAlgExec looks up
33	│ ├── ops/all_reduce/	33	│ ├── ops/all_reduce/
34	-│ │ ├── executor/ins_v2_all_reduce_sole_executor.cc # Main-pathway reference sample (registers CcuMSAllReduceSoleMesh)	34	+│ │ ├── algorithm/executor/ins_v2_all_reduce_sole_executor.cc # Main-pathway reference sample (registers CcuMSAllReduceSoleMesh)
35	│ │ └── ... # Other executor/template/kernel files	35	│ │ └── ... # Other executor/template/kernel files
36	│ └── ... # Other src submodules	36	│ └── ... # Other src submodules
37	├── include/ # Public headers (hccl.h / hccl_mc2.h, etc.)	37	├── include/ # Public headers (hccl.h / hccl_mc2.h, etc.)
	@@ -189,11 +189,11 @@ The template inherits from the engine template base class (using `CcuAlgTemplate
189	\| `CalcScratchMultiple` \| Returns the scratch multiplier for the executor to compute the per-loop data upper bound \| `inBuffType`/`outBuffType` input/output buffer types \|	189	\| `CalcScratchMultiple` \| Returns the scratch multiplier for the executor to compute the per-loop data upper bound \| `inBuffType`/`outBuffType` input/output buffer types \|
190	\| `FastLaunch` \| Fast dispatch: rewrites addresses using pre-stored `submitInfos` and dispatches the kernel directly \| `param`; `tempFastLaunchCtx` threads/ccuKernelSubmitInfos/buffInfo \|	190	\| `FastLaunch` \| Fast dispatch: rewrites addresses using pre-stored `submitInfos` and dispatches the kernel directly \| `param`; `tempFastLaunchCtx` threads/ccuKernelSubmitInfos/buffInfo \|
191		191
192	-For the full description of the `CalcCostCoeffParam` fields of the base class used by `CalcCostCoeff` (both executor and template), see the template header file comments under `src/ops/op_common/template/`.	192	+For the full description of the `CalcCostCoeffParam` fields of the base class used by `CalcCostCoeff` (both executor and template), see the template header file comments under `src/ops/op_common/algorithm/template/`.
193		193
194	#### REGISTER_EXEC_V2 Macro Parameters	194	#### REGISTER_EXEC_V2 Macro Parameters
195		195
196	-The macro is defined in `src/ops/op_common/executor/registry/coll_alg_v2_exec_registry.h`, signature:	196	+The macro is defined in `src/ops/op_common/algorithm/executor/registry/coll_alg_v2_exec_registry.h`, signature:
197		197
198	```cpp	198	```cpp
199	REGISTER_EXEC_V2(type, name, insCollAlgBase, AlgTopoMatch, InsAlgTemplate)	199	REGISTER_EXEC_V2(type, name, insCollAlgBase, AlgTopoMatch, InsAlgTemplate)