Star726
1126
Code
Code
Issues160
Pull Requests379
Actions
Discussion
Wiki
Analysis
Star726
1126
Merged
同步 CheckerL2 最新代码到 hccl_vm #5659
git_qkAdd on Sep 23
同步 CheckerL2 最新代码到 hccl_vm #5659
Merged
git_qkAdd on Sep 23
Merged fromgit_qk/hcommmasterintocann/hcommmaster
Total: 702 files changed+83338-74451
Atest/hccl_vm/.clang-format+13-0
@@ -0,0 +1,13 @@
1+# -----------------------------------------------------------------------------------------------------------
2+# Copyright (c) 2026 Huawei Technologies Co., Ltd.
3+# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4+# CANN Open Software License Agreement Version 2.0 (the "License").
5+# Please refer to the License for details. You may not use this file except in compliance with the License.
6+# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7+# INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8+# See LICENSE in the root of the software repository for the full text of the License.
9+# -----------------------------------------------------------------------------------------------------------
10+ 
11+BasedOnStyle: LLVM
12+IndentWidth: 4
13+UseTab: Never
Mtest/hccl_vm/README.md+14-26
@@ -67,7 +67,7 @@ bash build_pkg.sh
67 67 
68#### 3.2.1 环境配置68#### 3.2.1 环境配置
69 69 
70-请参照[hccl_rootinfo文件内容](#47-hccl_rootinfojson文件)，创建并配置hccl_rootinfo.json文件。70+若/etc路径下已存在hccl_rootinfo.json文件，请手动删除该文件。
71 71 
72#### 3.2.2 CCU模式72#### 3.2.2 CCU模式
73 73 
@@ -107,8 +107,8 @@ cd /home/workspace/hcomm/test/hccl_vm/hccl_vm_install/bin
107```107```
108 108 
1093. 验证hccl_test用例运行结果1093. 验证hccl_test用例运行结果
110-[Runner结果查看](#491-runner插件结果) 110+[Runner结果查看](#481-runner插件结果)
111-[Checker结果查看](#492-checker插件结果)111+[Checker结果查看](#482-checker插件结果)
112 112 
113#### 3.2.3 AICPU模式113#### 3.2.3 AICPU模式
114 114 
@@ -148,7 +148,7 @@ cd /home/workspace/hcomm/test/hccl_vm/hccl_vm_install/bin
148(hvm)$> exit148(hvm)$> exit
149```149```
150 150 
151-3. 验证hccl_test用例运行结果 [Runner结果查看](#491-runner插件结果) [Checker结果查看](#492-checker插件结果)151+3. 验证hccl_test用例运行结果 [Runner结果查看](#481-runner插件结果) [Checker结果查看](#482-checker插件结果)
152 152 
153#### 3.2.4 HostDPU模式153#### 3.2.4 HostDPU模式
154 154 
@@ -193,7 +193,7 @@ cd /home/workspace/hcomm/test/hccl_vm/hccl_vm_install/bin
193(hvm)$> exit193(hvm)$> exit
194```194```
195 195 
196-3. 验证hccl_test用例运行结果 [Runner结果查看](#491-runner插件结果) [Checker结果查看](#492-checker插件结果)196+3. 验证hccl_test用例运行结果 [Runner结果查看](#481-runner插件结果) [Checker结果查看](#482-checker插件结果)
197 197 
198### 3.3 Pytorch用例示例198### 3.3 Pytorch用例示例
199 199 
@@ -229,11 +229,11 @@ export HCOMM_CODE_HOME=/home/workspace/hcomm
229| ------------------------- | --------------------------------------------------------------------------------------------------------- | --------------------- |229| ------------------------- | --------------------------------------------------------------------------------------------------------- | --------------------- |
230| `HCCL_CODE_HOME` | HCCL-VM编译指定HCCL源码路径。默认未配置。 | `export HCCL_CODE_HOME=/home/workspace/hccl` |230| `HCCL_CODE_HOME` | HCCL-VM编译指定HCCL源码路径。默认未配置。 | `export HCCL_CODE_HOME=/home/workspace/hccl` |
231| `HCOMM_CODE_HOME` | HCCL-VM编译指定HCOMM源码路径。默认未配置。 | `export HCOMM_CODE_HOME=/home/workspace/hcomm` |231| `HCOMM_CODE_HOME` | HCCL-VM编译指定HCOMM源码路径。默认未配置。 | `export HCOMM_CODE_HOME=/home/workspace/hcomm` |
232-| `HCCLVM_ENABLE_DUMP_DATA` | 使能Runner插件dump input\&output数据。若使能，则在测试用例执行过程中，会将每个算子的input\&output数据dump到all\_rank\_input\_output.txt文件中 | `export HCCLVM_ENABLE_DUMP_DATA=1 使能，export HCCLVM_ENABLE_DUMP_DATA=0 禁用` |232+| `HCCLVM_ENABLE_DUMP_DATA` | 1. 使能性能仿真文件输出落盘。若使能，则在测试用例执行过程中输出3个runner_hcclvm_*_data.bin文件和1个ccu_channel_jetty_config.xml文件供性能仿真使用. <br> 2.使能Runner插件dump input\&output数据。若使能，则在测试用例执行过程中，会将每个算子的input\&output数据dump到all\_rank\_input\_output.txt文件中 | `export HCCLVM_ENABLE_DUMP_DATA=1 使能，export HCCLVM_ENABLE_DUMP_DATA=0 或不配置该环境变量 禁用` |
233 233 
234### 4.2 HCCL-Test用例构建234### 4.2 HCCL-Test用例构建
235 235 
236-hccl_test用例源码在CANN包安装目录下，支持OpenMPI和MPICH两种环境编译、运行，运行时差异详见[OpenMPI和MPICH环境用例执行差异](#48-openmpi和mpich环境用例运行差异)，本用例指导中以OpenMPI环境为例。236+hccl_test用例源码在CANN包安装目录下，支持OpenMPI和MPICH两种环境编译、运行，运行时差异详见[OpenMPI和MPICH环境用例执行差异](#47-openmpi和mpich环境用例运行差异)，本用例指导中以OpenMPI环境为例。
237 237 
238#### 4.2.1 OpenMPI环境编译238#### 4.2.1 OpenMPI环境编译
239 239 
@@ -711,23 +711,11 @@ Checker 插件当前使用 Checker V3 执行单算子校验。Checker V3 大图
711}711}
712```712```
713 713 
714-### 4.7 hccl_rootinfo.json文件714+### 4.7 OpenMPI和MPICH环境用例运行差异
715- 
716-目前工具初始化通信域使用的是ranktable.json文件，因此hccl_rootinfo.json文件的作用仅限于提供topo.json文件的路径。
717-若/etc路径下没有hccl_rootinfo.json文件，则用户需自行创建该文件，内容如下：
718- 
719-```json
720-{
721- "version": "2.0",
722- "topo_file_path": "/home/workspace/hcomm/test/hccl_vm/hccl_vm_install/data/topo.json"
723-}
724-```
725- 
726-### 4.8 OpenMPI和MPICH环境用例运行差异
727 715 
728运行hccl_test用例前，用户可以通过which命令判断当前环境使用的是哪个mpirun。716运行hccl_test用例前，用户可以通过which命令判断当前环境使用的是哪个mpirun。
729 717 
730-#### 4.8.1 环境变量配置差异718+#### 4.7.1 环境变量配置差异
731 719 
732环境中一般默认配置的是OpenMPI，若用户使用OpenMPI运行用例，一般不需要额外配置环境变量。720环境中一般默认配置的是OpenMPI，若用户使用OpenMPI运行用例，一般不需要额外配置环境变量。
733若用户使用MPICH环境运行用例，需要按照如下方式配置环境变量：721若用户使用MPICH环境运行用例，需要按照如下方式配置环境变量：
@@ -738,7 +726,7 @@ export LD_LIBRARY_PATH=/usr/lib/mpich/lib/:${ASCEND_HOME_PATH}/lib64/:${ASCEND_H
738export PATH=/usr/lib/mpich/bin:$PATH726export PATH=/usr/lib/mpich/bin:$PATH
739```727```
740 728 
741-#### 4.8.2 mpirun命令参数差异729+#### 4.7.2 mpirun命令参数差异
742 730 
743OpenMPI环境，用户按照如下命令运行hccl_test用例：731OpenMPI环境，用户按照如下命令运行hccl_test用例：
744 732 
@@ -764,9 +752,9 @@ mpirun -np 2 ${HCCL_TEST_PATH}/bin/reduce_scatter_test -b 64 -e 64 -d int32 -o s
764 752 
765 - -np 2：指定进程数为2，与节点数一致。753 - -np 2：指定进程数为2，与节点数一致。
766 754 
767-### 4.9 结果查看755+### 4.8 结果查看
768 756 
769-#### 4.9.1 Runner插件结果757+#### 4.8.1 Runner插件结果
770 758 
771如果在hccl-vm终端执行hccl-vm plugin install @runner安装插件后，算子流程执行完成后会自动触发runner插件的执行，最终结果依赖hccl_test进行校验，用户需关注在重定向的日志文件中是否存在[error]级别日志和最终校验结果：759如果在hccl-vm终端执行hccl-vm plugin install @runner安装插件后，算子流程执行完成后会自动触发runner插件的执行，最终结果依赖hccl_test进行校验，用户需关注在重定向的日志文件中是否存在[error]级别日志和最终校验结果：
772 760 
@@ -775,7 +763,7 @@ data_size(Bytes): | aveg_time(us): | alg_bandwidth(GB/s): | check_result:
77564 | 1000.00 | 0.00006 | success76364 | 1000.00 | 0.00006 | success
776```764```
777 765 
778-#### 4.9.2 Checker插件结果766+#### 4.8.2 Checker插件结果
779 767 
780在hccl-vm终端内执行hccl-vm plugin run @checker后，Checker校验流程及结果会打印在终端内，用于需关注是否存在[error]级别日志和最终校验结果：768在hccl-vm终端内执行hccl-vm plugin run @checker后，Checker校验流程及结果会打印在终端内，用于需关注是否存在[error]级别日志和最终校验结果：
781 769 
@@ -786,7 +774,7 @@ data_size(Bytes): | aveg_time(us): | alg_bandwidth(GB/s): | check_result:
786```774```
787 775 
788---776---
789-### 4.10 大块内存复用（仅校验模式）777+### 4.9 大块内存复用（仅校验模式）
790 778 
791仅校验模式用于大规模集群仅运行 Checker 校验的场景。开启后，单块 200MB 到 4GB 的大内存申请复用同一块 4GB 共享区 `HcclCommPool`，各 rank 共享、允许互相覆盖，以此大幅降低 `/dev/shm` 占用。此时大块内容不保证正确，仅适用于不读取缓冲区数据的 Checker V3 校验链路，需要数值正确的结果时请勿开启。779仅校验模式用于大规模集群仅运行 Checker 校验的场景。开启后，单块 200MB 到 4GB 的大内存申请复用同一块 4GB 共享区 `HcclCommPool`，各 rank 共享、允许互相覆盖，以此大幅降低 `/dev/shm` 占用。此时大块内容不保证正确，仅适用于不读取缓冲区数据的 Checker V3 校验链路，需要数值正确的结果时请勿开启。
792 780 
Mtest/hccl_vm/README_en.md+219-184
Mtest/hccl_vm/asset/cluster_model/config/cluster/ascend950_cluster_128_server_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19name: "ascend950_cluster_32_server_normal"11name: "ascend950_cluster_32_server_normal"
20description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"12description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"
21 13 
Mtest/hccl_vm/asset/cluster_model/config/cluster/ascend950_cluster_16_server_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19name: "ascend950_cluster_32_server_normal"11name: "ascend950_cluster_32_server_normal"
20description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"12description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"
21 13 
Mtest/hccl_vm/asset/cluster_model/config/cluster/ascend950_cluster_32_server_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19name: "ascend950_cluster_32_server_normal"11name: "ascend950_cluster_32_server_normal"
20description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"12description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"
21 13 
Mtest/hccl_vm/asset/cluster_model/config/cluster/ascend950_cluster_4_server_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19name: "ascend950_cluster_32_server_normal"11name: "ascend950_cluster_32_server_normal"
20description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"12description: "昇腾950 normal组网：32个超级节点，每个超级节点1个服务器"
21 13 
Mtest/hccl_vm/asset/cluster_model/config/cluster/ascend960_cluster_4_server_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19name: "ascend960_cluster_32_server_normal"11name: "ascend960_cluster_32_server_normal"
20description: "昇腾960 normal组网：32个超级节点，每个超级节点1个服务器"12description: "昇腾960 normal组网：32个超级节点，每个超级节点1个服务器"
21 13 
Mtest/hccl_vm/asset/cluster_model/config/server_or_pod/ascend950_server_topo_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19type: "server_intra_links"11type: "server_intra_links"
20name: "ascend950_links_topo_normal"12name: "ascend950_links_topo_normal"
21description: "ascend950芯片NORMAL类型拓扑连接关系描述文件"13description: "ascend950芯片NORMAL类型拓扑连接关系描述文件"
Mtest/hccl_vm/asset/cluster_model/config/server_or_pod/ascend960_server_topo_normal.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19type: "server_intra_links"11type: "server_intra_links"
20name: "ascend960_links_topo_normal"12name: "ascend960_links_topo_normal"
21description: "ascend960芯片NORMAL类型拓扑连接关系描述文件"13description: "ascend960芯片NORMAL类型拓扑连接关系描述文件"
Mtest/hccl_vm/asset/cluster_model/topo_meta/111.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 1 13 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/112.yaml+0-8
@@ -8,15 +8,7 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
19- 
20meta:12meta:
21 podNum: 1 13 podNum: 1
22 serNum: 1 14 serNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/114.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/118.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 1 13 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/121.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/122.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/124.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/128.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/12_2_4.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/141.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/asset/cluster_model/topo_meta/142.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19# 1. 全局统计信息 podNum, serNum, rankNum 都小于 102411# 1. 全局统计信息 podNum, serNum, rankNum 都小于 1024
20meta:12meta:
21 podNum: 113 podNum: 1
Mtest/hccl_vm/config/log_config.yaml+0-8
@@ -8,14 +8,6 @@
8# See LICENSE in the root of the software repository for the full text of the License.8# See LICENSE in the root of the software repository for the full text of the License.
9# -----------------------------------------------------------------------------------------------------------9# -----------------------------------------------------------------------------------------------------------
10 10 
11-# You may not use this file except in compliance with the License.
12-# You may obtain a copy of the License at
13-# http://www.apache.org/licenses/LICENSE-2.0
14-# Unless required by applicable law or agreed to in writing, software
15-# distributed under the License is distributed on an "AS IS" BASIS,
16-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17-# See the License for the specific language governing permissions and limitations under the License.
18- 
19hccl_vm:11hccl_vm:
20 console_level: 212 console_level: 2
21 file_level: 113 file_level: 1
Mtest/hccl_vm/docs/faq/generate_faq_html.py+0-9
@@ -10,15 +10,6 @@
10# See LICENSE in the root of the software repository for the full text of the License.10# See LICENSE in the root of the software repository for the full text of the License.
11# -----------------------------------------------------------------------------------------------------------11# -----------------------------------------------------------------------------------------------------------
12 12 
13-# You may not use this file except in compliance with the License.
14-# You may obtain a copy of the License at
15-# http://www.apache.org/licenses/LICENSE-2.0
16-# Unless required by applicable law or agreed to in writing, software
17-# distributed under the License is distributed on an "AS IS" BASIS,
18-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
19-# See the License for the specific language governing permissions and limitations under the License.
20- 
21-#!/usr/bin/env python3
22"""13"""
23FAQ Markdown 转 HTML 转换脚本14FAQ Markdown 转 HTML 转换脚本
24 15 
Mtest/hccl_vm/docs/faq/generate_faq_html_v2.py+0-9
@@ -10,15 +10,6 @@
10# See LICENSE in the root of the software repository for the full text of the License.10# See LICENSE in the root of the software repository for the full text of the License.
11# -----------------------------------------------------------------------------------------------------------11# -----------------------------------------------------------------------------------------------------------
12 12 
13-# You may not use this file except in compliance with the License.
14-# You may obtain a copy of the License at
15-# http://www.apache.org/licenses/LICENSE-2.0
16-# Unless required by applicable law or agreed to in writing, software
17-# distributed under the License is distributed on an "AS IS" BASIS,
18-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
19-# See the License for the specific language governing permissions and limitations under the License.
20- 
21-#!/usr/bin/env python3
22"""13"""
23FAQ Markdown 转 HTML 转换脚本14FAQ Markdown 转 HTML 转换脚本
24 15 
Mtest/hccl_vm/docs/hccl_vm_binary_file_format.md+53-39
@@ -4,11 +4,19 @@
4 4 
5| 项目 | 内容 |5| 项目 | 内容 |
6|----------|--------------------------------|6|----------|--------------------------------|
7-| 版本 | v1.0 |7+| 版本 | v2.0 |
8-| 更新日期 | 2025-01-16 |8+| 更新日期 | 2026-09-17 |
9| 适用场景 | HCCL VM 仿真数据导出与读取 |9| 适用场景 | HCCL VM 仿真数据导出与读取 |
10| 作者 | HCCL VM Team |10| 作者 | HCCL VM Team |
11 11 
12+> **v2.0 变更说明**：综合数据文件（`*_hcclvm_syn_data.bin`）**version 保持 1，文件头不做版本变更**
13+> （本特性不做新旧文件兼容性区分）。`ChannelData` 二进制布局与历史版本**一致（未新增字段，176 字节）**，
14+> 但 AICPU 模式 dump **语义**变更以支持多 channel（`HCCL_UB_MULTI_CHANNEL_NUM`）特性：
15+> jetty 精确归属到（本端 eid, 对端 eid）链路 + **每 eid 对一条聚合记录**
16+> （`jettyNum` 为该 eid 对的 jetty 个数，`jettyId[0..jettyNum)` 携带全部本端 jetty，
17+> 读取方需在数组内查找 `task.jettyId`，不能再假定 `jettyNum` 恒为 1 或直接取 `jettyId[0]`）。
18+> 任务/微码/flag 文件格式不变。
19+ 
12---20---
13 21 
14## 1. 概述22## 1. 概述
@@ -34,7 +42,7 @@
34| 偏移量 | 字节数 | 类型 | 字段名 | 说明 |42| 偏移量 | 字节数 | 类型 | 字段名 | 说明 |
35|:------:|:------:|---------|-------------|----------------------------|43|:------:|:------:|---------|-------------|----------------------------|
36| 0 | 4 | uint32 | magic | 魔数，标识文件类型 |44| 0 | 4 | uint32 | magic | 魔数，标识文件类型 |
37-| 4 | 2 | uint16 | version | 版本号（当前为 1） |45+| 4 | 2 | uint16 | version | 版本号（当前为 1） |
38| 6 | 2 | uint16 | header_size | 文件头大小（20 字节） |46| 6 | 2 | uint16 | header_size | 文件头大小（20 字节） |
39| 8 | 4 | uint32 | flags | 标志位（保留） |47| 8 | 4 | uint32 | flags | 标志位（保留） |
40| 12 | 4 | uint32 | count | 数据条目数量 |48| 12 | 4 | uint32 | count | 数据条目数量 |
@@ -65,7 +73,7 @@ struct FileHeader {
65|:----:|----------------------------------|----------|----------------------------------------------|73|:----:|----------------------------------|----------|----------------------------------------------|
66| 1 | FileHeader | 20 Bytes | 文件头 |74| 1 | FileHeader | 20 Bytes | 文件头 |
67| 2 | ModelInfo | 变长 | 模型信息（含 ModelInfoCommInner、VDataDesTag、All2AllDataDesTag） |75| 2 | ModelInfo | 变长 | 模型信息（含 ModelInfoCommInner、VDataDesTag、All2AllDataDesTag） |
68-| 3 | ChannelInfo 或 JettyInfo | 变长 | 二选一，由 `op_expansion_mode` 决定 |76+| 3 | ChannelInfo（CCU/AICPU 共用） | 变长 | AIV 模式无此数据块 |
69| 4 | MemLayoutInfo | 变长 | 内存布局信息 |77| 4 | MemLayoutInfo | 变长 | 内存布局信息 |
70 78 
71**op_expansion_mode 取值说明：**79**op_expansion_mode 取值说明：**
@@ -73,7 +81,7 @@ struct FileHeader {
73| 值 | 模式 | 使用的数据结构 |81| 值 | 模式 | 使用的数据结构 |
74|:--:|---------|----------------|82|:--:|---------|----------------|
75| 0 | CCU | ChannelInfo |83| 0 | CCU | ChannelInfo |
76-| 1 | AICPU | JettyInfo |84+| 1 | AICPU | ChannelInfo（多 channel 语义见 3.3.3） |
77 85 
78### 3.2 ModelInfo 结构86### 3.2 ModelInfo 结构
79 87 
@@ -126,7 +134,7 @@ struct FileHeader {
126 134 
127### 3.3 ChannelInfo 结构135### 3.3 ChannelInfo 结构
128 136 
129-**适用条件**：`op_expansion_mode = 0`（CCU 模式）137+**适用条件**：`op_expansion_mode = 0`（CCU 模式）或 `1`（AICPU 模式）——两种模式均使用 ChannelData 定长结构；AIV 模式不写。
130 138 
131#### 3.3.1 ChannelInfo 头部139#### 3.3.1 ChannelInfo 头部
132 140 
@@ -137,24 +145,40 @@ struct FileHeader {
137 145 
138#### 3.3.2 ChannelData146#### 3.3.2 ChannelData
139 147 
140-**大小**：152 字节。148+**大小**：176 字节（布局与历史版本一致）。
141 149 
142-| 偏移量 | 字节数 | 类型 | 字段名 | 说明 |150+| 偏移量 | 字节数 | 类型 | 字段名 | 说明 |
143|:------:|:------:|------------|-----------|---------------|151|:------:|:------:|------------|-----------|---------------|
144-| 0 | 2 | uint16 | channelId | 通道 ID |152+| 0 | 2 | uint16 | channelId | CCU 模式: 通道 ID；AICPU 模式: EndPointPair 主键（eid 对标识） |
145| 2 | 1 | uint8 | srcDieId | 源 Die ID |153| 2 | 1 | uint8 | srcDieId | 源 Die ID |
146| 3 | 1 | uint8 | dstDieId | 目标 Die ID |154| 3 | 1 | uint8 | dstDieId | 目标 Die ID |
147| 4 | 4 | uint32 | srcRank | 源 Rank ID |155| 4 | 4 | uint32 | srcRank | 源 Rank ID |
148| 8 | 4 | uint32 | dstRank | 目标 Rank ID |156| 8 | 4 | uint32 | dstRank | 目标 Rank ID |
149-| 12 | 16 | uint8[16] | leid | 本端 EID |157+| 12 | 16 | uint8[16] | leid | 本端 EID |
150-| 28 | 16 | uint8[16] | reid | 远端 EID |158+| 28 | 16 | uint8[16] | reid | 远端 EID |
151| 44 | 2 | uint16 | protocol | 协议类型 |159| 44 | 2 | uint16 | protocol | 协议类型 |
152| 46 | 2 | uint16 | jettyNum | Jetty 数量 |160| 46 | 2 | uint16 | jettyNum | Jetty 数量 |
153-| 48 | 128 | uint32[32] | jettyId | Jetty ID 数组 |161+| 48 | 128 | uint32[32] | jettyId | Jetty ID 数组 |
154 162 
155-### 3.4 JettyInfo 结构163+#### 3.3.3 多 channel 语义（AICPU 模式）
156 164 
157-**适用条件**：`op_expansion_mode = 1`（AICPU 模式）165+`HCCL_UB_MULTI_CHANNEL_NUM=N` 特性下，同一对 rank（同一对 eid）之间存在 N 条并行 channel，
166+AICPU 模式 dump 规则：
167+ 
168+1. **每 eid 对一条聚合 ChannelData 记录**：同一 `channelId`（eid 对）的 N 条 channel 输出 **1 条**记录，
169+ `jettyNum=N`，`jettyId[0..jettyNum)` 为该 eid 对上**全部本端（发送侧）jetty**，
170+ 按**本端 jetty 创建序**排列（数组下标即 pair 内 channel 序号）；
171+2. 读取方将任务元数据中的 `task.jettyId` 在 `jettyId[0..jettyNum)` **数组内查找**，
172+ 即得任务经由哪个 jetty 从本端哪个 eid（`leid`）发往对端哪个 eid（`reid`）；
173+3. `jettyId[0..jettyNum)` **精确归属**到该 eid 对（由 RA 层 `RaCtxQpImport` 回填归属信息），
174+ 不含该本端 eid 上通往其它对端的 jetty。
175+ 
176+CCU 模式维持"每 channel 一条记录、`jettyId[]` 携带该 channel 的 jetty 列表"的既有语义。
177+ 
178+### 3.4 JettyInfo 结构（未使用，保留历史说明）
179+ 
180+`JettyData` 为早期设计中 AICPU 模式的独立结构，**当前实现未使用**：AICPU 模式实际同样写
181+ChannelData（见 3.3）。本节保留仅作历史格式参考，读取方不应按 JettyData 解析 AICPU 数据。
158 182 
159#### 3.4.1 JettyInfo 头部183#### 3.4.1 JettyInfo 头部
160 184 
@@ -346,7 +370,7 @@ struct FileHeader {
346| 值 | 名称 | 说明 |370| 值 | 名称 | 说明 |
347|:--:|---------------------------------|---------------------------|371|:--:|---------------------------------|---------------------------|
348| 0 | SIM_OP_EXPANSION_MODE_CCU | CCU 模式，使用 ChannelInfo |372| 0 | SIM_OP_EXPANSION_MODE_CCU | CCU 模式，使用 ChannelInfo |
349-| 1 | SIM_OP_EXPANSION_MODE_AICPU | AICPU 模式，使用 JettyInfo |373+| 1 | SIM_OP_EXPANSION_MODE_AICPU | AICPU 模式，使用 ChannelInfo |
350 374 
351---375---
352 376 
@@ -392,10 +416,10 @@ struct ModelInfoCommInner {
392};416};
393 417 
394// ============================================================================418// ============================================================================
395-// 通道数据 (CCU 模式)419+// 通道数据 (CCU / AICPU 模式共用)
396// ============================================================================420// ============================================================================
397struct ChannelData {421struct ChannelData {
398- uint16_t channelId;422+ uint16_t channelId; // CCU: 通道ID; AICPU: EndPointPair主键(eid对标识)
399 uint8_t srcDieId;423 uint8_t srcDieId;
400 uint8_t dstDieId;424 uint8_t dstDieId;
401 uint32_t srcRank;425 uint32_t srcRank;
@@ -403,12 +427,12 @@ struct ChannelData {
403 uint8_t leid[16];427 uint8_t leid[16];
404 uint8_t reid[16];428 uint8_t reid[16];
405 uint16_t protocol;429 uint16_t protocol;
406- uint16_t jettyNum;430+ uint16_t jettyNum; // AICPU: 该eid对的本端jetty个数(多channel时可大于1); CCU: 该channel的jetty个数
407- uint32_t jettyId[32];431+ uint32_t jettyId[32]; // AICPU: 该eid对全部本端(发送)jetty, 按创建序排列, task.jettyId在数组内查找
408};432};
409 433 
410// ============================================================================434// ============================================================================
411-// Jetty 数据 (AICPU 模式)435+// Jetty 数据（未使用的历史结构；AICPU 实际使用 ChannelData，见上方）
412// ============================================================================436// ============================================================================
413struct JettyData {437struct JettyData {
414 uint32_t jettyId;438 uint32_t jettyId;
@@ -572,9 +596,8 @@ bool ReadSynthesisData(const char* filename) {
572 std::cout << "\n[All2All] Matrix Count: " << matrixCount << std::endl;596 std::cout << "\n[All2All] Matrix Count: " << matrixCount << std::endl;
573 }597 }
574 598 
575- // ===== Step 5: 根据 op_expansion_mode 读取通道或 Jetty 信息 =====599+ // ===== Step 5: 读取 ChannelInfo（CCU 与 AICPU 模式均为 ChannelData 定长结构；AIV 模式无此段） =====
576- if (modelComm.op_expansion_mode == 0) {600+ if (modelComm.op_expansion_mode != 2) {
577- // CCU 模式 - 读取 ChannelInfo
578 uint32_t channelCount;601 uint32_t channelCount;
579 fread(&channelCount, sizeof(uint32_t), 1, fp);602 fread(&channelCount, sizeof(uint32_t), 1, fp);
580 603
@@ -585,21 +608,12 @@ bool ReadSynthesisData(const char* filename) {
585 for (size_t i = 0; i < channels.size() && i < 3; ++i) {608 for (size_t i = 0; i < channels.size() && i < 3; ++i) {
586 std::cout << " Channel[" << i << "]: ID=" << channels[i].channelId609 std::cout << " Channel[" << i << "]: ID=" << channels[i].channelId
587 << ", srcRank=" << channels[i].srcRank610 << ", srcRank=" << channels[i].srcRank
588- << ", dstRank=" << channels[i].dstRank << std::endl;611+ << ", dstRank=" << channels[i].dstRank
589- }612+ << ", jettyNum=" << channels[i].jettyNum;
590- } else {613+ for (uint16_t j = 0; j < channels[i].jettyNum; ++j) {
591- // AICPU 模式 - 读取 JettyInfo614+ std::cout << ", jetty[" << j << "]=" << channels[i].jettyId[j];
592- uint32_t jettyCount;615+ }
593- fread(&jettyCount, sizeof(uint32_t), 1, fp);616+ std::cout << std::endl;
594-
595- std::vector<JettyData> jetties(jettyCount);
596- fread(jetties.data(), sizeof(JettyData), jettyCount, fp);
597-
598- std::cout << "\n[JettyInfo] Count: " << jettyCount << std::endl;
599- for (size_t i = 0; i < jetties.size() && i < 3; ++i) {
600- std::cout << " Jetty[" << i << "]: ID=" << jetties[i].jettyId
601- << ", srcRank=" << jetties[i].srcRank
602- << ", dstRank=" << jetties[i].dstRank << std::endl;
603 }617 }
604 }618 }
605 619 
@@ -704,7 +718,7 @@ bool ReadTaskMetaData(const char* filename) {
704| 1 | 字节序 | 所有多字节字段均采用**小端序**（Little-Endian） |718| 1 | 字节序 | 所有多字节字段均采用**小端序**（Little-Endian） |
705| 2 | 内存对齐 | 结构体使用 `#pragma pack(1)` 进行 **1 字节对齐** |719| 2 | 内存对齐 | 结构体使用 `#pragma pack(1)` 进行 **1 字节对齐** |
706| 3 | 变长字段读取 | 先读取 `count` 字段，再根据 count 值读取相应数量的数据 |720| 3 | 变长字段读取 | 先读取 `count` 字段，再根据 count 值读取相应数量的数据 |
707-| 4 | 模式判断 | 根据 `op_expansion_mode` 判断读取 ChannelInfo 还是 JettyInfo |721+| 4 | 模式判断 | CCU 与 AICPU 模式均读取 ChannelInfo（ChannelData 定长结构）；AIV 模式无此段；JettyData 为未使用的历史结构 |
708| 5 | 联合体解析 | 任务元数据中的 `taskData` 需根据 `taskType` 选择正确的结构体 |722| 5 | 联合体解析 | 任务元数据中的 `taskData` 需根据 `taskType` 选择正确的结构体 |
709| 6 | 魔数验证 | 读取文件时务必先验证魔数，确保文件类型正确 |723| 6 | 魔数验证 | 读取文件时务必先验证魔数，确保文件类型正确 |
710 724 
Dtest/hccl_vm/docs/header_dependency_en.md+0-355
@@ -1,355 +0,0 @@
1-# CheckerL2 Third-Party Header File Dependency List
2- 
3-> This list is based on the `CheckerL2` version and covers all third-party header file dependencies.
4- 
5-## Description
6- 
7-This document lists all third-party header files that the project references through `#include`. Each entry marks its origin (CANN / HCCL / HCOMM) and its actual path on disk.
8- 
9-**Path Conventions**:
10-- CANN root directory: `/home/teamserver/workspace/Ascend/cann-9.1.0/` (abbreviated as `CANN/` below)
11-- HCOMM root directory: `/home/teamserver/workspace/hcomm/` (abbreviated as `HCOMM/` below)
12-- HCCL root directory: `/home/teamserver/workspace/hccl/` (abbreviated as `HCCL/` below)
13-- Project root directory: `CheckerL2/` (abbreviated as `Project/` below)
14- 
15-**Resolution Principle**: The declaration order of `target_include_directories` in CMake determines the search priority. The compiler uses the first match found.
16- 
17----
18- 
19-## 1. Header Files from CANN
20- 
21-### 1.1 ACL Runtime Interfaces (`CANN/include/acl/`)
22- 
23-| #include | Actual File Path | Reference Locations |
24-|----------|------------|---------|
25-| `acl/acl_base.h` | `CANN/include/acl/acl_base.h` | all aclrt_*.cc under proxy, hccl_comm_stub.cc, hccp_stub.cc, hccp_ra_socket_stub.cc, ascend_hal_stub.cc, aclrt_exec_control.cc; include/runnerdb/db_sim_runner_common.h; test/proxy/ |
26-| `acl/acl_rt.h` | `CANN/include/acl/acl_rt.h` | all aclrt_*.cc under proxy, hccl_op_stub.cc; test/proxy/ |
27- 
28-### 1.2 HCCL Public Interfaces (`CANN/include/hccl/`)
29- 
30-| #include | Actual File Path | Reference Locations |
31-|----------|------------|---------|
32-| `hccl/hccl_types.h` | `CANN/include/hccl/hccl_types.h` | proxy/aclrt_kernel_stub.cc, hccl_op_stub.cc, hccl_proxy_common.h; plugin/runner/ccu_executor/ccu_fp16.h; common/sim_data_dump.cc; device_arm/level2/proxy/device_sqe_parse_stub.h; test/proxy/ |
33-| `hccl/hccl.h` | `CANN/include/hccl/hccl.h` | **proxy/hccl_op_stub.cc**; test/proxy/hccl_comm_stub_test.cc |
34-| `hccl/hcom.h` | `CANN/include/hccl/hcom.h` | proxy/hccl_inner_stub.cc; test/proxy/hccl_inner_stub_test.cc |
35-| `hccl/base.h` | `CANN/include/hccl/base.h` | proxy/aclrt_context_stub.cc; plugin/runner/ccu_executor/ccu_fp16.h; plugin/checker/header/external/task_param.h; include/sim_ip_address.h |
36-| `hccl/dtype_common.h` | `CANN/include/hccl/dtype_common.h` | proxy/scatter_stub.cc; test/proxy/scatter_stub_test.cc |
37-| `hccl/hccl_common.h` | `CANN/x86_64-linux/asc/include/adv_api/hccl/hccl_common.h` | test/proxy/hccl_proxy_common_test.cc |
38- 
39-### 1.3 HCCL Internal Interfaces (`CANN/pkg_inc/hccl/`)
40- 
41-| #include | Actual File Path | Reference Locations |
42-|----------|------------|---------|
43-| `"dtype_common.h"` | `CANN/pkg_inc/hccl/dtype_common.h` | proxy/aclrt_new_stub.cc, hccp_ccu_stub.cc, hccp_stub.cc, hccl_op_stub.cc, aclrt_device_stub.cc, hccl_proxy_common.cc; plugin/runner/runner_utils/storage_manager.h; plugin/checker/src/framework/task_graph_generator_v3/ccu_graph_generator_v3/ccu_all_rank_param_recorder_v3.h; plugin/checker/src/framework/task_graph_generator/ccu_all_rank_param_recorder.h; plugin/checker/src/utils/storage_manager.h |
44- 
45-### 1.4 CANN Runtime Interfaces (`CANN/pkg_inc/runtime/`)
46- 
47-| #include | Actual File Path | Reference Locations |
48-|----------|------------|---------|
49-| `"runtime/base.h"` | `CANN/pkg_inc/runtime/runtime/base.h` | proxy/hccl_comm_stub.cc, aclrt_runtime_config.cc, aclrt_notify_stub.cc, hccp_stub.cc, aclrt_device_stub.cc, hccp_ra_socket_stub.cc, aclrt_stream_stub.cc, aclrt_stub.cc; test/proxy/ |
50-| `"runtime/event.h"` | `CANN/pkg_inc/runtime/runtime/event.h` | proxy/aclrt_runtime_config.cc, aclrt_notify_stub.cc |
51-| `"runtime/rt.h"` | `CANN/pkg_inc/runtime/runtime/rt.h` | device_arm/level2/proxy/device_sqe_parse_stub.h |
52- 
53-### 1.5 CANN Profiling Interfaces (`CANN/pkg_inc/profiling/`)
54- 
55-| #include | Actual File Path | Reference Locations |
56-|----------|------------|---------|
57-| `"aprof_pub.h"` | `CANN/pkg_inc/profiling/aprof_pub.h` | proxy/aprofiling_stub.cc; device_arm/level2/proxy/aprofiling_stub.cc; test/proxy/aprofiling_stub_test.cc |
58- 
59-### 1.6 CANN Trace Interfaces (`CANN/pkg_inc/trace/`)
60- 
61-| #include | Actual File Path | Reference Locations |
62-|----------|------------|---------|
63-| `"atrace_pub.h"` | `CANN/pkg_inc/trace/atrace_pub.h` | proxy/adapter_rts_stub.cc |
64-| `"atrace_types.h"` | `CANN/pkg_inc/trace/atrace_types.h` | proxy/aclrt_new_stub.cc, hccp_stub.cc |
65- 
66-### 1.7 CANN Driver Interfaces (`CANN/include/driver/`)
67- 
68-| #include | Actual File Path | Reference Locations |
69-|----------|------------|---------|
70-| `"ascend_hal.h"` | `CANN/include/driver/ascend_hal.h` | proxy/adapter_rts_stub.cc, aclrt_new_stub.cc, ascend_hal_stub.cc; device_arm/level2/proxy/device_sqe_parse_stub.h, ascend_hal_stub.cc; test/device_arm/ |
71- 
72-### 1.8 CANN Base Interfaces (`CANN/pkg_inc/base/`)
73- 
74-| #include | Actual File Path | Reference Locations |
75-|----------|------------|---------|
76-| `"dlog_pub.h"` | `CANN/pkg_inc/base/dlog_pub.h` | plugin/checker/header/external/log.h (as fallback) |
77-| `"log_types.h"` | `CANN/pkg_inc/base/log_types.h` | plugin/checker/header/external/dlog_pub.h |
78- 
79-### 1.9 CANN CCU Microcode Interfaces (`CANN/pkg_inc/hcomm/ccu/`)
80- 
81-| #include | Actual File Path | Reference Locations |
82-|----------|------------|---------|
83-| `"ccu_microcode_v1.h"` | `CANN/pkg_inc/hcomm/ccu/ccu_microcode_v1.h` | proxy/hccp_ccu_stub.cc; plugin/runner/ccu_executor/ccu_resource_manager.cc, ccu_resource_common.h; plugin/checker/header/external/ccu_rep_base.h, ccu_instr_info.h; plugin/checker/header/internal/binary_data_type_pub.h; multiple files under plugin/checker/src/framework/task_graph_generator_v3/ccu_graph_generator_v3/; multiple files under plugin/checker/src/framework/task_graph_generator/; include/sim_binary_data_type_pub.h |
84-| `"ccu_common.h"` | `CANN/pkg_inc/hcomm/ccu/ccu_common.h` | proxy/hccp_ccu_stub.cc |
85- 
86-### 1.10 CANN RTS Device Interfaces (`CANN/pkg_inc/runtime/runtime/rts/`)
87- 
88-| #include | Actual File Path | Reference Locations |
89-|----------|------------|---------|
90-| `"rts_device.h"` | `CANN/pkg_inc/runtime/runtime/rts/rts_device.h` | proxy/hccp_stub.cc, adapter_rts_stub.cc, aclrt_new_stub.cc |
91- 
92-### 1.11 CANN Runtime Other Interfaces (`CANN/pkg_inc/runtime/`)
93- 
94-| #include | Actual File Path | Reference Locations |
95-|----------|------------|---------|
96-| `"mem.h"` | `CANN/pkg_inc/runtime/mem.h` | proxy/hccp_stub.cc, aclrt_new_stub.cc |
97-| `"rt_external_kernel.h"` | `CANN/pkg_inc/runtime/rt_external_kernel.h` | proxy/hccp_ccu_stub.cc |
98- 
99----
100- 
101-## 2. Header Files from HCOMM
102- 
103-### 2.1 HCCP Network Interfaces (`HCOMM/src/base_comm/resources/hccp/inc/network/`)
104- 
105-| #include | Actual File Path | Reference Locations |
106-|----------|------------|---------|
107-| `"hccp.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp.h` | proxy/adapter_rts_stub.cc |
108-| `"hccp_async.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp_async.h` | proxy/adapter_rts_stub.cc |
109-| `"hccp_ping.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp_ping.h` | proxy/adapter_rts_stub.cc |
110-| `"hccp_tlv.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp_tlv.h` | proxy/adapter_rts_stub.cc, hccp_stub.cc, **hccp_ccu_stub.cc** |
111-| `"hccp_ctx.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp_ctx.h` | proxy/hccp_ccu_stub.cc, hccp_stub.cc |
112-| `"hccp_common.h"` | `HCOMM/src/base_comm/resources/hccp/inc/network/hccp_common.h` | proxy/hccl_comm_stub.cc, aclrt_new_stub.cc, hccp_ccu_stub.cc, hccp_stub.cc, hccp_ra_socket_stub.cc |
113- 
114-### 2.2 HCCP CCU External Dependencies (`HCOMM/src/base_comm/resources/hccp/external_depends/ccu/`)
115- 
116-| #include | Actual File Path | Reference Locations |
117-|----------|------------|---------|
118-| `"ccu_u_comm.h"` | `HCOMM/src/base_comm/resources/hccp/external_depends/ccu/ccu_u_comm.h` | proxy/hccp_ccu_stub.cc |
119- 
120----
121- 
122-## 3. Header Files from HCCL (ARM Cross-Compilation Only)
123- 
124-### 3.1 Operator Common Interfaces (`HCCL/src/ops/op_common/inc/`)
125- 
126-| #include | Actual File Path | Reference Locations |
127-|----------|------------|---------|
128-| `"alg_param.h"` | `HCCL/src/ops/op_common/inc/alg_param.h` | device_arm/hccl_kernel_executor.cc; test/device_arm/ |
129- 
130-### 3.2 HCCL Common Code (`HCCL/src/common/`)
131- 
132-| #include | Actual File Path | Reference Locations |
133-|----------|------------|---------|
134-| (multiple headers exposed through include paths) | `HCCL/src/common/*.h` | the device_arm build module references these indirectly through include paths |
135- 
136-### 3.3 Dynamic Symbol Loading (`HCCL/src/common/hcomm_dlsym/` and `ccu/`)
137- 
138-| #include | Actual File Path | Reference Locations |
139-|----------|------------|---------|
140-| (multiple headers exposed through include paths) | `HCCL/src/common/hcomm_dlsym/*.h`, `HCCL/src/common/hcomm_dlsym/ccu/*.h`, `*.hpp` | the device_arm build module references these indirectly through include paths |
141- 
142----
143- 
144-## 4. Project Local External Header Files (Checker-Specific, HCCL Interface Definition Copies)
145- 
146-Path: `Project/src/plugin/checker/header/external/`
147- 
148-The project uses these local copies as a fallback when the CANN package does not contain a header file with the same name.
149- 
150-| File Name | Description | Reference Locations |
151-|-------|------|---------|
152-| `hccl_types.h` | HcclResult enumeration (24 error codes including HCCL_SUCCESS) | 50+ files under plugin/checker/ reference this through `#include "hccl_types.h"`; include/sim_binary_data_type_pub.h, sim_loader.h, sim_ccu_jetty_ctx.h, sim_ccu_channel_ctx.h, and so on |
153-| `base.h` | Basic type definitions | 19 references under plugin/checker/ (framework, semantics_check, checker, and so on); header/external/log.h |
154-| `data_type.h` | Data type definitions | 3 references under plugin/checker/src/framework/task_graph_generator_v3/, 4 references under task_graph_generator/ |
155-| `ccu_instr_info.h` | CCU instruction information | 8 references under plugin/checker/ (task_ccu.h, task_utils.cc, task_graph_generator, and so on) |
156-| `ccu_task_param.h` | CCU task parameters | plugin/checker/src/utils/task_ccu.h |
157-| `ccu_rep_base.h` | CCU Reporter base | the file exists; no direct `#include` reference currently |
158-| `ccu_rep_type.h` | CCU Reporter types | the file exists; no direct `#include` reference currently |
159-| `dlog_pub.h` | Logging public interface (CANN fallback) | plugin/checker/header/external/log.h |
160-| `enum_factory.h` | Enumeration factory | 5 internal references under plugin/checker/ (sim_task.h, check_utils.h, data_slice.h, data_type.h, task_param.h) |
161-| `const_val.h` | Constant value definitions | plugin/checker/header/external/task_param.h |
162-| `log.h` | Logging interface | internal to plugin/checker/ |
163-| `log_types.h` | Logging types (CANN fallback) | plugin/checker/header/external/dlog_pub.h |
164-| `string_util.h` | String utilities | plugin/checker/header/external/enum_factory.h; header/internal/task_def.h; src/utils/check_utils.h |
165-| `task_param.h` | Task parameters | the file exists; no direct `#include` reference currently (other header files include it internally) |
166- 
167----
168- 
169-## 5. AIV Dependencies (AIV Operator Simulation Only)
170- 
171-This section describes the dependencies for AIV (AI Vector Core) operator simulation in the tool. These dependencies originate from the HCOMM repository and the AscendC programming interfaces. The tool maintains local backups of key definitions to remove strong dependencies on external repositories.
172- 
173-### 5.1 AivOpArgs Data Structure (Local Backup in Project)
174- 
175-**Original Source**: the `hccl_aiv_utils.h` header file from the HCOMM repository
176- 
177-**Backup Location**: `src/proxy/level2/aclrt_kernel_stub.cc` lines 597-628
178- 
179-**Description**: `AivOpArgs` is the parameter structure for AIV collective communication operators. The structure passes AIV kernel execution parameters between the host side and the device side. The tool originally loaded the `ops_hccl::ExecuteKernelLaunch()` function from the HCOMM library through `dlsym` (symbol name: `_ZN8ops_hccl19ExecuteKernelLaunchERKNS_9AivOpArgsE`). That function accepts this structure as a parameter. The tool maintains a complete backup of the structure to remove the dependency on the HCOMM runtime library.
180- 
181-**Structure Definition**:
182-```cpp
183-struct AivOpArgs {
184- HcclCMDType cmdType = HcclCMDType::HCCL_CMD_MAX; // collective communication command type
185- std::string comm = {}; // communication domain identifier
186- HcclComm hcclComm = nullptr; // HCCL communicator handle
187- uint32_t numBlocks = AIV_STUB_MAX_NUM_BLOCKS; // AIV core count
188- void *stream = nullptr; // execution stream
189- uint64_t beginTime = 0; // start timestamp
190- OpCounterInfo counter = {}; // counter information
191- void *buffersIn = nullptr; // input buffer address
192- uint64_t input = 0; // input offset
193- uint64_t output = 0; // output offset
194- uint32_t rank = 0; // current rank
195- uint32_t sendRecvRemoteRank = 0; // SendRecv remote rank
196- uint32_t rankSize = 0; // total rank count
197- uint64_t xRankSize = 0, yRankSize = 0, zRankSize = 0; // topology dimensions
198- uint64_t count = 0; // element count
199- HcclDataType dataType = HcclDataType::HCCL_DATA_TYPE_INT32; // data type
200- HcclReduceOp op = HcclReduceOp::HCCL_REDUCE_SUM; // Reduce operation type
201- uint32_t root = 0; // Root rank
202- uint32_t sliceId = 0; // Slice ID
203- uint64_t inputSliceStride = 0; // input slice stride
204- uint64_t outputSliceStride = 0; // output slice stride
205- uint64_t repeatNum = 0; // repeat count
206- uint64_t inputRepeatStride = 0; // input repeat stride
207- uint64_t outputRepeatStride = 0; // output repeat stride
208- bool isOpBase = false; // whether this is a base operator
209- ExtraArgs extraArgs = {}; // extra parameters (AllToAllV, and so on)
210- uint64_t topo_[AIV_STUB_TOPO_LEN] = {0}; // topology information
211- KernelArgsType argsType = KernelArgsType::ARGS_TYPE_SERVER; // parameter type
212-};
213-```
214- 
215-**Dependency Risk**: Changes to the `AivOpArgs` structure on the HCOMM side (such as field additions, field removals, order changes, or type changes) may cause parameter misalignment during `dlsym` calls. This leads to runtime errors.
216- 
217-### 5.2 aiv_communication_base_v2.h Header File (Local Backup in Project)
218- 
219-**Original Source**: the header file with the same name from the HCOMM repository
220- 
221-**Backup Location**: `src/proxy/level2/aiv_kernel/hccl_op_stub/aiv_communication_base_v2.h`
222- 
223-**Description**: This header file defines the base class `AivCommBase` for AIV collective communication operators and the related kernel parameter macros (`KERNEL_ARGS_DEF`, `KERNEL_ARGS_CALL`, and so on). The tool originally included this file directly from the HCOMM repository. The tool maintains a complete backup to remove the dependency.
224- 
225-**Key Definitions in the File**:
226- 
227-| Definition | Description |
228-|------|------|
229-| `AivSuperKernelArgs` | SuperKernel parameter structure |
230-| `ExtraArgs` | Extra parameter structure (sendCounts/recvCounts, and so on, for AllToAllV) |
231-| `KERNEL_ARGS_DEF` / `KERNEL_ARGS_CALL` | AIV kernel parameter macros (approximately 25 parameters) |
232-| `SUPERKERNEL_ARGS_DEF` / `SUPERKERNEL_ARGS_CALL` | SuperKernel parameter macros |
233-| `AivCommBase` class | AIV communication base class that encapsulates core methods such as Barrier, Record/WaitFlag, and CpGM2GM |
234-| `EXPORT_AIV_META_INFO` | AIV kernel metadata export macro |
235-| `AIV_ATOMIC_DATA_TYPE_DEF` / `AIV_COPY_DATA_TYPE_DEF` | Supported data type macro definitions |
236- 
237-**Dependency Risk**: Changes to the AIV communication interfaces or kernel parameters on the HCOMM side require a synchronized update to this backup file.
238- 
239-### 5.3 AIV Operator Op Header Files (HCOMM / HCCL Dependencies)
240- 
241-Each AIV kernel implementation file for a collective communication operator in the tool (`src/proxy/level2/aiv_kernel/hccl_op_stub/*/aiv_communication_v2.cc`) includes the corresponding operator op header file through `#include`. These header files **do not exist within the project**. They come from the HCOMM and HCCL repositories respectively:
242- 
243-| #include | Source | Actual File Path | Reference Locations |
244-|----------|------|------------|---------|
245-| `"aiv_all_reduce_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/all_reduce/` | hccl_op_stub/all_reduce/aiv_communication_v2.cc |
246-| `"aiv_all_gather_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/all_gather/` | hccl_op_stub/all_gather/aiv_communication_v2.cc |
247-| `"aiv_reduce_scatter_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/reduce_scatter/` | hccl_op_stub/reduce_scatter/aiv_communication_v2.cc |
248-| `"aiv_broadcast_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/broadcast/` | hccl_op_stub/broadcast/aiv_communication_v2.cc |
249-| `"aiv_all_to_all_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/all_to_all/` | hccl_op_stub/all_to_all_v/aiv_communication_v2.cc |
250-| `"aiv_all_to_all_v_op.h"` | HCOMM | `hcomm/src/legacy/ascend910/algorithm/base/alg_aiv_template/all_to_all/` | hccl_op_stub/all_to_all_v/aiv_communication_v2.cc |
251-| `"aiv_scatter_op.h"` | HCCL | `hccl/src/ops/scatter/template/aiv/kernel/` | hccl_op_stub/scatter/aiv_communication_v2.cc |
252-| `"aiv_reduce_op.h"` | HCCL | `hccl/src/ops/reduce/template/aiv/kernel/` | hccl_op_stub/reduce/aiv_communication_v2.cc |
253- 
254-**Dependency Risk**: Changes to the interface signatures, template parameters, or member functions in the operator op header files on the HCOMM or HCCL side will cause AIV kernel compilation failures or lead to inconsistent runtime behavior.
255- 
256-### 5.4 AscendC Interface Dependencies
257- 
258-The tool depends on the AscendC programming interfaces provided by the CANN package to simulate AIV kernel execution. All AscendC interfaces in the project use local stub files under the `src/proxy/level2/aiv_kernel/ascendc_stub/` directory. The tool does not use the AscendC header files from the CANN SDK directly.
259- 
260-**AscendC Interfaces in Use** (implemented through stubs):
261- 
262-| Stub Header File | Encapsulated AscendC Interfaces | Reference Locations |
263-|------------|-------------------|---------|
264-| `ascendc_base_stub.h` | base classes such as `GlobalTensor`, `LocalTensor`, `TPipe`, `TBuf`, `TQueBind`, `pipe_barrier`, and `GetBlockIdx` | nearly all files under aiv_kernel/ |
265-| `ascendc_copy_stub.h` | data copy interfaces such as `DataCopy` and `DataCopyPad` | ascendc_stub/kernel_operator.h |
266-| `ascendc_math_stub.h` | math operation interfaces such as `Add`, `Sub`, `Mul`, and `Reduce` | ascendc_stub/kernel_operator.h |
267-| `ascendc_memory_stub.h` | atomic operation interfaces such as `SetAtomicAdd`, `SetAtomicMax`, and `SetAtomicMin` | ascendc_stub/kernel_operator.h |
268-| `ascendc_sync_stub.h` | synchronization interfaces such as `send_flag`, `recv_flag`, and `SyncAll` | ascendc_stub/kernel_operator.h |
269-| `ascendc_utils_stub.h` | utility functions such as `PipeBarrier` | ascendc_stub/kernel_operator.h |
270- 
271-**Dependency Risks**:
272- 
273-1. AscendC API changes (such as class member function signature changes or template parameter adjustments) may cause mismatches between stub implementations and the real interfaces
274-2. New AscendC data types (such as `fp8_e4m3fn_t`, `fp8_e5m2_t`, and `hifloat8_t`) require synchronized updates to stub support
275-3. The `AivCommBase` class uses AscendC template classes extensively (`GlobalTensor<T>`, `LocalTensor<T>`, `TQueBind`, and so on). Semantic changes to these types affect AIV simulation correctness
276- 
277----
278- 
279-## 6. AICPU Dependencies (AICPU Mode Only)
280- 
281-This section describes the dependencies for simulating host-device interaction in AICPU mode. The `ExecuteAicpuKernel` function in `src/device_arm/hccl_kernel_executor.cc` implements this mode. The function simulates the process of dispatching an AICPU kernel to the device side. This involves serialization and address translation of multiple data structures.
282- 
283-### 6.1 AICPU Interaction Data Structures
284- 
285-The tool maintains local backups of the following data structures in `src/device_arm/common/aicpu_args_stub.h` to remove the strong dependency on the HCOMM runtime library:
286- 
287-| Data Structure | Purpose | Corresponding Kernel Function |
288-|---------|------|-----------------|
289-| `CommAicpuParam` | Communication domain initialization parameters (hcomId, device ID, H2D/D2H transfer parameters) | `RunAicpuIndOpCommInit` |
290-| `HDCommunicateParams` | H2D/D2H control transfer parameters (deviceAddr, readCacheAddr) | referenced by `CommAicpuParam` |
291-| `ThreadMgrAicpuParam` | Thread management parameters (threadNum, serialized threadParam array, deviceHandle) | `RunAicpuIndOpThreadInit`, `RunAicpuThreadSupplementNotify` |
292-| `AicpuTsThread` | AICPU thread information (streamType, notifyLoadType, devId) | referenced by `ThreadMgrAicpuParam` |
293-| `InitTask` | Channel initialization task (context pointer, isCustom flag) | `RunAicpuIndOpChannelInitV2` |
294-| `HcclChannelUrmaRes` | Channel URMA resource description (channelList, uniqueIdAddr, remoteRankList) | `RunAicpuIndOpChannelInitV2` |
295-| `UniqueIdV2Header` | UniqueId V2 header (type, notifyNum, bufferNum, connNum) | channel link establishment serialized data parsing |
296-| `ConnUniqueBlock` | Connection unique block (flexible array that contains `ConnUniqueIds`) | channel link establishment serialized data parsing |
297- 
298-### 6.2 OpParam Data Structure (External Dependency)
299- 
300-**Source**: the `alg_param.h` header file from the HCCL repository (`HCCL/src/ops/op_common/inc/`), in the `ops_hccl` namespace
301- 
302-**Description**: `OpParam` is the general parameter structure for AICPU collective communication operators. The `HcclLaunchAicpuKernel` function uses this structure. Unlike the locally backed-up structures above, the tool **directly uses** the `OpParam` definition from the HCCL source code without a local backup.
303- 
304-**Dependency Risk**: Changes to the `OpParam` structure on the HCCL side (field additions, field removals, or type changes) may cause the tool to parse parameters incorrectly or crash at runtime.
305- 
306-### 6.3 AICPU Kernel Function Dependencies
307- 
308-The tool dynamically loads the following kernel functions from the HCOMM runtime library through `dlopen` + `dlsym` to simulate the actual execution on the AICPU side:
309- 
310-| dlsym Symbol Name | Library | Function |
311-|-------------|-------|------|
312-| `RunAicpuIndOpCommInit` | `libccl_kernel.so` | AICPU communication domain initialization |
313-| `RunAicpuIndOpThreadInit` | `libccl_kernel.so` | AICPU thread initialization |
314-| `RunAicpuIndOpChannelInitV2` | `libccl_kernel.so` | AICPU channel initialization V2 |
315-| `RunAicpuDfxOpInfoInitV2` | `libccl_kernel.so` | AICPU DFX operator information initialization V2 |
316-| `RunAicpuThreadSupplementNotify` | `libccl_kernel.so` | AICPU resource supplement notification |
317-| `HcclLaunchAicpuKernel` | `libscatter_aicpu_kernel.so` | AICPU collective communication kernel launch |
318- 
319-**Dependency Risks**:
320-1. Changes to the signatures or behavior of the above kernel functions on the HCOMM side may cause `dlsym` calls to fail or produce incorrect results
321-2. If `libccl_kernel.so` or `libscatter_aicpu_kernel.so` adds or removes kernel functions, the tool must update its function pointer list accordingly
322-3. Inconsistencies between the locally backed-up structures (`CommAicpuParam`, `ThreadMgrAicpuParam`, `HcclChannelUrmaRes`, and so on) and the actual definitions on the HCOMM side will cause address translation errors and memory out-of-bounds access
323- 
324-### 6.4 Kernel SO Library Name Dependencies
325- 
326-The `InitKernelFuncHandle` function hardcodes the following SO library names to load the execution entry points for AICPU kernels:
327- 
328-| Hardcoded SO Name | Loaded dlsym Symbols | Description |
329-|--------------|-----------------|------|
330-| `libccl_kernel.so` | `RunAicpuIndOpCommInit`, `RunAicpuIndOpThreadInit`, `RunAicpuIndOpChannelInitV2`, `RunAicpuDfxOpInfoInitV2`, `RunAicpuThreadSupplementNotify` | HCOMM communication framework core library |
331-| `libscatter_aicpu_kernel.so` | `HcclLaunchAicpuKernel` | AICPU collective communication operator kernel library |
332-| `libslog.so` | (logging library dependency) | CANN secure logging library |
333-| `libc_sec.so` | (security library dependency) | CANN security function library |
334- 
335-**Dependency Risks**:
336-1. **SO Name Changes**: If the HCOMM side renames or splits the above SOs (for example, renaming `libccl_kernel.so`), the `dlopen` call in the tool will fail. This causes all AICPU kernels to become non-executable
337-2. **Operator Kernel SO Name Ambiguity**: The operator kernel SO name is fixed as `libscatter_aicpu_kernel.so` regardless of the operator type (AllReduce, AllGather, ReduceScatter, and so on). The "scatter" in the name suggests applicability only to the Scatter operator, but the library actually hosts kernels for all collective communication operators. If the HCOMM side later splits SOs by operator type (for example, `liballreduce_aicpu_kernel.so`), the tool will fail to adapt
338-3. **User-Defined Operator Scenarios**: The HCCL business layer does not strictly define SO names for user-defined operators. Users may compile private SOs (for example, `libcustom_alltoall_kernel.so`). The tool currently recognizes only the fixed SO names listed above. The tool cannot load or simulate private SOs for user-defined operators
339- 
340----
341- 
342-## 7. Dependency Overview
343- 
344-| Source | Header File Count | Modules in Use |
345-|------|---------|---------|
346-| **CANN** | 23 distinct header files | nearly all modules (proxy, checker, runner, common, store, topo, utils, device_arm) |
347-| **HCOMM** | 13 distinct header files | proxy module (hccp network interfaces + AIV operator ops) |
348-| **HCCL** | 3 direct references + indirect references | device_arm module (ARM cross-compilation) + proxy (AIV scatter/reduce ops) |
349-| **Project Local External** | 14 header files | checker module only (as a fallback when CANN headers are missing) |
350-| **AIV Backup** | 2 key definitions | proxy/aiv_kernel module only (AIV operator simulation) |
351-| **AIV Operator Ops** | 8 header files (6 from HCOMM + 2 from HCCL) | proxy/aiv_kernel module only (AIV operator simulation) |
352-| **AscendC Interfaces** | 6 stub header files | proxy/aiv_kernel module only (AIV operator simulation) |
353-| **AICPU Backup** | 8 local structures + 6 kernel functions + 4 SO library names | device_arm module only (AICPU mode host-device interaction) |
354- 
355- 
Dtest/hccl_vm/docs/insightV3_guide_en.md+0-343
@@ -1,343 +0,0 @@
1-# HVRM Insight V3 User Guide
2- 
3-## 1. Overview
4- 
5-HVRM Insight V3 supports the `DAGView` task graph viewer. This document covers data preparation, page access, dataset selection, DAG graph browsing, and node search.
6- 
7-Other pages and advanced linkage features are not yet available.
8- 
9----
10- 
11-## 2. Prerequisites
12- 
13-Before using Insight V3, confirm that the Checker has completed execution and has generated Insight data.
14- 
15-To configure Checker data output, confirm that the Checker configuration file has enabled Insight dump:
16- 
17-```json
18-# The configuration file is located at /pathto/hccl_vm_install/plugin/checker/manifest.json
19- 
20-{
21- ...
22- "setting": { // Checker plugin configuration items
23- ...
24- "enable_insight_dump": true, // Whether to output Checker V3 Insight data (disabled by default)
25- "enable_memory_snapshot_dump": false // Whether to output memory snapshot data (not yet supported, disabled by default, requires enabling Insight data output first)
26- }
27-}
28-```
29- 
30-DAGView requires at least the following data files:
31- 
32-```text
33-<dataset_name>/
34-├── manifest.json
35-└── graph/
36- ├── graph.msgpack
37- └── layout.msgpack
38-```
39- 
40-The `manifest.json` file provides dataset metadata. The `graph.msgpack` and `layout.msgpack` files render the DAG task graph.
41- 
42----
43- 
44-## 3. Compile and Install the Insight Plugin
45- 
46-The Insight V3 frontend source code is located at:
47- 
48-```text
49-{hccl_vm directory}/src/plugin/insight/frontend_v3
50-```
51- 
52-Before installing the Insight plugin, compile the frontend first, then run the HCCL VM build and installation process.
53- 
54-Note: Compiling the Insight visualization frontend requires `Node.js` and `npm`. Use `Node.js 20.19.0` or later and `npm 10.x` or later.
55- 
56-The recommended steps are as follows:
57- 
58-1. Navigate to the frontend directory:
59- 
60-```bash
61-cd {hccl_vm directory}/src/plugin/insight/frontend_v3
62-```
63- 
64-2. Install frontend dependencies:
65- 
66-```bash
67-npm install
68-```
69- 
70-3. Build the frontend output:
71- 
72-```bash
73-npm run build
74-```
75- 
76-After the frontend compilation completes, the system generates the `src/plugin/insight/dist/` directory.
77- 
78-4. Return to the HCCL_VM root directory, and run `build.sh` to compile and install hccl-vm:
79- 
80-```bash
81-cd {HCCL_VM directory}
82-bash build.sh --package-path <ASCEND_CANN_PATH> --hcomm-path <HCOMM_CODE_PATH>
83-```
84- 
85-To package the installation directory, append `--pkg`. If the current build scenario requires AICPU / AIV / FULL mode, append the `--aicpu`, `--aiv`, or `--full` parameter following the standard project procedure.
86- 
87----
88- 
89-## 4. Insight Plugin Configuration and Start/Stop
90- 
91-The Insight plugin configuration file is located at:
92- 
93-```text
94-hccl_vm_install/plugin/visualization/insight/manifest.json
95-```
96- 
97-The default configuration is as follows:
98- 
99-```json
100-{
101- "name": "insight",
102- "version": "1.0.0",
103- "entry": "python3 server.py",
104- "dependency": {
105- "min_core_version": "1.0.0"
106- },
107- "setting": {
108- "dist_path": "./dist",
109- "data_path": "../../checker/data/insight",
110- "topo_config_path": "../../../../../asset/cluster_model/config/cluster",
111- "port": 8080
112- }
113-}
114-```
115- 
116-The common fields are described below:
117- 
118-| Field | Description |
119-|------|------|
120-| `entry` | The Insight plugin startup command. The system starts the service using `python3 server.py`. |
121-| `setting.dist_path` | The frontend static resource directory. The default value is `./dist`. |
122-| `setting.data_path` | The Insight data directory. The default points to the Checker output directory `data/insight`. |
123-| `setting.topo_config_path` | The cluster topology configuration directory. |
124-| `setting.port` | The Insight service port. The default value is `8080`. |
125- 
126-To modify the port or data directory, edit the `setting` field in `manifest.json` directly.
127- 
128-> Warning: Insight binds to `127.0.0.1` by default using `python3 server.py`. To listen on a specific IP address instead of the default `localhost(127.0.0.1)`, navigate to the `hccl_vm_install/plugin/visualization/insight` directory and run the following command manually:
129->
130-> ```bash
131-> python3 server.py --host <ip> --port <port>
132-> ```
133->
134-> For example:
135->
136-> ```bash
137-> cd hccl_vm_install/plugin/visualization/insight
138-> python3 server.py --host 0.0.0.0 --port 8080
139-> ```
140- 
141-The plugin installation and uninstall commands are as follows:
142- 
143-```bash
144-# Install and start the Insight plugin
145-hccl-vm plugin install @insight
146- 
147-# Uninstall the Insight plugin
148-hccl-vm plugin uninstall @insight
149-```
150- 
151----
152- 
153-## 5. Open Insight
154- 
155-If the Insight service is running, open the service address in the browser. For example:
156- 
157-```text
158-http://localhost:8080
159-```
160- 
161-To start Insight through the plugin, run the following command:
162- 
163-```bash
164-hccl-vm plugin install @insight
165-```
166- 
167-After the plugin starts, the terminal outputs the access address. Open that address in the browser to access the Insight page.
168- 
169-![Start Insight and open the page](insight_image/V3_image/launch.gif)
170- 
171----
172- 
173-## 6. Select a Dataset
174- 
175-After opening Insight, the system displays the `Overview` page by default. Follow these steps to select a dataset for analysis:
176- 
177-1. View the dataset list in the middle dataset table.
178-2. Click the target dataset.
179-3. Confirm the target Rank in the left Rank tree.
180-4. If the DAG graph is large, reduce the Rank selection range first.
181-5. Click `Enter Correlation View` on the right side.
182- 
183-![Select a dataset and enter the correlation view](insight_image/V3_image/dagView_V3.gif)
184- 
185-Select a small number of Ranks during the first analysis. Confirm the analysis direction, then expand the Rank range gradually.
186- 
187----
188- 
189-## 7. Enter DAGView
190- 
191-After entering the `Correlation` page, focus on the `DAGView Task View` in the lower section.
192- 
193-The main page areas are as follows:
194- 
195-| Area | Function |
196-|------|------|
197-| Left panel | Select Ranks and search for nodes |
198-| Center bottom | Display the DAG task graph |
199-| Right detail panel | View the current node details |
200- 
201-If the upper memory view is empty, the DAGView below remains fully functional.
202- 
203----
204- 
205-## 8. View the DAG Task Graph
206- 
207-The DAG task graph consists of swim lanes, nodes, and arrows:
208- 
209-| Element | Description |
210-|------|------|
211-| Swim lane | Represents an execution queue, displayed as `Rank / Stream / Queue` |
212-| Node | Represents a task, for example data transfer, Reduce, Record, or Wait |
213-| Arrow | Represents a dependency between tasks |
214-| Loop dashed box | Represents a Loop region. A dashed box encloses the related nodes within the Loop, and a `Loop` label appears outside the box. |
215- 
216-The common node types are as follows:
217- 
218-| Node Type | Meaning |
219-|----------|------|
220-| `TRANS_MEM` | Data transfer task |
221-| `REDUCE` | Reduce task |
222-| `RECORD` | Notify record task |
223-| `WAIT` | Notify wait task |
224-| `CCU_GRAPH` | CCU graph task |
225-| `AIV_GRAPH` | AIV graph task |
226- 
227-The recommended viewing approach is as follows:
228- 
229-1. Follow the arrow direction to review the task dependency order.
230-2. Click the task node of interest.
231-3. View the Rank, Stream, Queue, and detailed node information in the right detail panel.
232-4. Examine the parent nodes and child nodes to trace upstream and downstream dependencies.
233- 
234-For Loop display, pay attention to the following information:
235- 
236-1. If the DAG contains a Loop, the page automatically overlays a Loop dashed box on the task graph.
237-2. The dashed box covers the Loop start point, end point, and internal loop body nodes for quick identification of loop boundaries.
238-3. Click the Loop capsule to navigate directly to the corresponding Loop Start node.
239-4. After selecting a node inside a Loop, the right detail panel displays the Loop information and the nesting chain for that node.
240- 
241-![Browse the task graph and view node details in DAGView](insight_image/V3_image/dagView_V3_2.gif)
242- 
243----
244- 
245-## 9. Canvas Operations
246- 
247-The DAGView canvas supports the following operations:
248- 
249-| Operation | Description |
250-|------|------|
251-| Pan canvas | Drag an empty area |
252-| Zoom in / Zoom out | Use the mouse scroll wheel, or click `+` / `-` in the bottom-right corner |
253-| Reset zoom | Click the percentage button in the bottom-right corner |
254-| Select node | Click the target node |
255- 
256-When the graph is large, reduce the Rank selection range first, then zoom into specific areas to examine dependencies.
257- 
258----
259- 
260-## 10. View Node Details
261- 
262-After clicking a DAG node, the right detail panel displays node information. The common sections are as follows:
263- 
264-| Section | Description |
265-|------|------|
266-| Node overview | Node ID, task type, Rank, Stream, Queue |
267-| Loop information | The Loop that contains the current node, Loop count, instruction range, Loop boundaries |
268-| Parent nodes | Upstream nodes that the current node depends on |
269-| Child nodes | Downstream nodes that depend on the current node |
270-| Node semantics | Notify, Task metadata, Memory Slices |
271-| Raw JSON | The complete raw information of the current node |
272- 
273-When troubleshooting dependency relationships, check `Parent nodes` and `Child nodes` first. When troubleshooting loop structures, focus on `Loop information`. When troubleshooting data transfer issues, focus on `Memory Slices`.
274- 
275-The `Memory Slices` section follows these display rules:
276- 
277-1. A regular memory Slice displays `rank / type / offset / size`.
278-2. If the Slice belongs to `MS_CCU`, Insight V3 displays the `MSID` instead of the underlying abstract offset.
279-3. For batch tasks, if multiple `MS_CCU` Slices exist, the system merges them into a single summary card automatically.
280-4. The summary card displays a description such as `Total 8 MSIDs used`.
281-5. After expanding `MSID Details`, inspect the `id` and `size` of each `MSID` individually.
282- 
283----
284- 
285-## 11. Search for Nodes
286- 
287-The left `Search` panel locates DAG nodes quickly.
288- 
289-The supported search fields are as follows:
290- 
291-| Field | Use Case |
292-|------|----------|
293-| `taskId` | Use when the node ID is known |
294-| `taskType` | Search by task type, for example `TRANS_MEM` |
295-| `notifyId` | Search by notify id |
296- 
297-Steps:
298- 
299-1. Select a search field in the left search panel.
300-2. Enter the complete keyword.
301-3. Click the search result.
302-4. DAGView locates and selects the corresponding node automatically.
303- 
304-The current search uses exact matching. If the system finds no results, confirm that the keyword is complete and that the target Rank is selected.
305- 
306----
307- 
308-## 12. Recommended Analysis Workflow
309- 
310-Follow this workflow for DAGView analysis:
311- 
312-1. Select a dataset on the `Overview` page.
313-2. Select a small number of Ranks in the left Rank tree.
314-3. Click `Enter Correlation View`.
315-4. View the overall nodes and dependency arrows in DAGView.
316-5. Click the node of interest and view the details on the right.
317-6. Trace upstream and downstream dependencies through parent nodes and child nodes.
318-7. Use the search function to locate known nodes quickly.
319- 
320----
321- 
322-## 13. Frequently Asked Questions
323- 
324-### 13.1 The Page Shows No Data
325- 
326-Confirm that the Insight service address is correct, and confirm that the Checker has generated Insight data.
327- 
328-### 13.2 The DAG Is Empty After Entering the Correlation Page
329- 
330-Confirm that the dataset contains the following files:
331- 
332-```text
333-graph/graph.msgpack
334-graph/layout.msgpack
335-```
336- 
337-### 13.3 The DAG Graph Is Too Dense
338- 
339-Reduce the number of selected Ranks on the left side first. View a subset of Ranks, then zoom into specific areas for analysis.
340- 
341-### 13.4 The Search Returns No Node
342- 
343-Confirm that the search keyword is complete, and confirm that the Rank containing the target node is selected.
Dtest/hccl_vm/docs/runner_trace/debug_tool_design_errors_en.md+0-423
@@ -1,423 +0,0 @@
1-# CCU Executor Debug Tool Design Document — Feasibility Review and Optimization Report
2- 
3-A comprehensive review based on the code in `src/plugin/solver/virtual_runtime/ccu_executor/` and the design proposal in `runner_debug/debug_tool_design_doc.md`.
4- 
5----
6- 
7-## I. Feasibility Issues (Must Fix)
8- 
9-### P0-1: CKE Wait Flow Does Not Match Instrumentation Design
10- 
11-**Severity**: P0 (Approach feasibility error)
12- 
13-**Problem Description**:
14- 
15-The instrumentation pseudocode in Section 4.1 assumes that the `waitCKE_` flag can be detected before `ExecuteInstr()` returns, and the delta from `snapshotBefore → snapshotAfter` is computed afterward. However, the actual code flow is:
16- 
17-```
18-WaitCkeProcess() → CKE not met → ccuSimulator_->SetWaitCKEFlag(true) → return (Process() not called)
19-→ UpdateLoopStatus() returns false → ExecuteInstr() returns false → Execute() returns false
20-```
21- 
22-Key point: When CKE is not met, `WaitCkeProcess` inside `Run()` sets the waitCKE flag and returns directly, **without calling Process() and without modifying any resources**. Therefore:
23-- Pre-execution snapshot = post-execution snapshot (delta is empty)
24-- snapshotAfter should not be captured and delta should not be computed
25- 
26-**Code Evidence**:
27- 
28-`CcuExecutorBase.cc` lines 162-172:
29-```cpp
30-if ((waitCKE & waitCKEMask) == waitCKEMask) {
31- Process(ccuResMgr); // CKE met → execute Process
32-} else {
33- ccuSimulator_->SetWaitCKEFlag(true); // CKE not met → set flag, do not execute Process
34- return;
35-}
36-```
37- 
38-**Fix Suggestion**:
39- 
40-When waitCKE is not met, only record key spin fields; do not compute delta or create a full trace entry:
41-```cpp
42-// During waitCKE, only record spin information
43-CcuTraceCollector::RecordWaitSpin(rankId_, dieId_, curInstrId,
44- waitCKEId, waitCKEMask, actualCKEValue);
45-// Do not capture snapshotAfter, do not compute delta
46-// Do not create full CcuTraceEntry
47-return false;
48-```
49- 
50-Only create a full entry after CKE passes (merging previous spin information).
51- 
52----
53- 
54-### P0-2: globalSeqId/execRound Maintenance Should Be in SequentialExecutor
55- 
56-**Severity**: P0 (Approach feasibility error)
57- 
58-**Problem Description**:
59- 
60-The document assumes `CcuTraceCollector` maintains `globalSeqId/execRound/currentSqeTaskId` inside `CcuSimulator::ExecuteInstr()`. However, `CcuSimulator` can only see execution within its own CCU and cannot perceive global scheduling rounds or CCU switching.
61- 
62-`CcuSimulator::ExecuteInstr()` executes within a single CCU and does not know:
63-- Which round of the outer `while(HasTask())` it is (`execRound`)
64-- The global execution sequence number (`globalSeqId`)
65-- Which SQE task is currently being executed (`sqeTaskId`)
66- 
67-This information is only available in the upper call chain of `SequentialExecutor::Execute()` and `TaskCcuGraph()`.
68- 
69-**Code Evidence**:
70- 
71-`hccl_task_sequential_execute.cc` lines 51-72 — Global scheduling loop:
72-```cpp
73-while (HasTask()) {
74- uint32_t rankId = 0;
75- for (auto& rankTasks : allRankTaskQueues_) { // Iterate all ranks
76- for (auto& streamTasks : rankTasks) { // Iterate each stream
77- while (!streamTasks.empty()) {
78- auto task = streamTasks.front();
79- auto ret = ExecuteOneTask(task); // ← globalSeqId incremented here
80- if (ret == HCCL_SIM_VRT_HOLD_CMD) break;
81- streamTasks.pop();
82- }
83- }
84- }
85-}
86-```
87- 
88-`hccl_task_thread.cc` lines 135-148 — TaskCcuGraph:
89-```cpp
90-auto simulator = ccuResMgr.InitSimulator(rankId, dieId, instrStartId, endInstrId, instCnt);
91-// sqeTaskId is known here (obtained from task parameters)
92-if (simulator->Execute() == false) {
93- return HCCL_SIM_VRT_HOLD_CMD;
94-}
95-```
96- 
97-**Fix Suggestion**:
98- 
99-Global context is managed by `SequentialExecutor` and passed via parameters:
100-```cpp
101-// In SequentialExecutor::Execute()
102-while (HasTask()) {
103- uint32_t execRound = 0;
104- for (rankTasks : allRankTaskQueues_) {
105- for (streamTasks : rankTasks) {
106- while (!streamTasks.empty()) {
107- CcuTraceCollector::BeginGlobalStep(execRound);
108- // globalSeqId incremented inside ExecuteOneTask
109- auto ret = ExecuteOneTask(task);
110- ...
111- }
112- }
113- }
114- execRound++;
115-}
116-```
117- 
118-Pass sqeTaskId to `CcuSimulator` in `TaskCcuGraph()`:
119-```cpp
120-uint32_t sqeTaskId = CcuTraceCollector::RegisterSqeTask(rankId, dieId, task);
121-CcuSimulator::SetCurrentSqeTaskId(sqeTaskId); // Used internally by Simulator
122-```
123- 
124----
125- 
126-### P0-3: Full Snapshot Collection Has Excessive Performance Overhead
127- 
128-**Severity**: P0 (Performance bottleneck)
129- 
130-**Problem Description**:
131- 
132-In Section 4.1, a `CaptureResourceSnapshot()` (full read of all resources) is performed before and after each instruction, with significant overhead:
133- 
134-- Single snapshot: 4096×8(XN) + 4096×8(GSA) + 1024×2(CKE) ≈ 67KB
135-- Before and after each instruction ≈ 134KB
136-- 10000 instructions ≈ 1.34GB (snapshot data only)
137- 
138-Section 4.5 mentions incremental collection optimization, but the pseudocode in 4.1 still follows the full snapshot + diff computation flow, which contradicts the optimization approach.
139- 
140-**Industry Reference**:
141- 
142-- **rr debugger**: Checkpoint + incremental recording model; full snapshots only at key nodes
143-- **Chrome DevTools Timeline**: Event stream format; each event contains only changed args
144-- **Perfetto (Android trace)**: Protobuf binary incremental format
145- 
146-**Fix Suggestion**:
147- 
148-Change to ResourceManager change interception + initial snapshot + periodic checkpoints:
149- 
150-1. **Initial Snapshot**: Full snapshot at run start (stored in `CcuIdentity.initialSnapshot`)
151-2. **Change Interception**: Intercept changes directly in `CcuResourceManager::UpdateXnValue/UpdateGsaValue/UpdateCkeValue` etc., recording delta:
152-```cpp
153-void CcuResourceManager::UpdateXnValue(int rankId, int dieId, uint16_t xnId, uint64_t value) {
154- uint64_t oldValue = GetXnValue(rankId, dieId, xnId);
155- if (oldValue != value) {
156- CcuTraceCollector::RecordXnDelta(rankId, dieId, xnId, oldValue, value);
157- }
158- xnData_[rankId][dieId][xnId] = value;
159-}
160-```
161-3. **Periodic Checkpoint**: Full snapshot every N instructions (N configurable, default 1000) for quick intermediate state recovery
162-4. **Restore Arbitrary Point State**: From nearest checkpoint + accumulate subsequent deltas
163- 
164----
165- 
166-## II. Design Optimization Points (Recommended Improvements)
167- 
168-### P1-1: Loop Offset Formula Is Incomplete
169- 
170-**Severity**: P1 (Information gap)
171- 
172-**Problem Description**:
173- 
174-The offset calculation in `UpdateAddress()` in the code is more complex than the `CcuExecutionContext.gsaOffset` in the document:
175-```cpp
176-uint64_t CcuExecutorBase::UpdateAddress(uint64_t addr, uint16_t addrExpandCoef) {
177- return addr + ((ccuSimulator_->GetLoopExtendNum() * ccuSimulator_->GetGSAOffset()) << addrExpandCoef)
178- + ((ccuSimulator_->GetCurLoopCnt() * ccuSimulator_->GetLoopIterStepGSA()) << addrExpandCoef);
179-}
180-```
181- 
182-This involves 5 parameters: `addr` (original address), `addrExpandCoef` (expansion coefficient), `loopExtendNum`, `gsaOffset`, `curLoopCnt`, `loopIterStepGSA`.
183- 
184-The current `CcuExecutionContext` has only a simple `gsaOffset` field and cannot fully reproduce the Loop GSA address offset calculation process.
185- 
186-**Fix Suggestion**:
187- 
188-Add complete offset parameters to `CcuExecutionContext`:
189-```cpp
190-struct CcuExecutionContext {
191- bool inLoop;
192- uint16_t loopRound; // Current iteration round
193- uint16_t loopExtendIndex; // Expand index
194- 
195- // Loop offset parameters (valid only when inLoop=true)
196- uint64_t gsaAddrOffset; // = extendIndex * gsaOffset + curLoopCnt * iterStepGSA (computed final offset)
197- uint64_t gsaOffset; // GSA base offset coefficient (GetGSAOffset)
198- uint64_t iterStepGSA; // GSA step per iteration round (GetLoopIterStepGSA)
199- uint32_t curLoopCnt; // Current iteration count (GetCurLoopCnt)
200- uint32_t loopExtendNum; // Expand count (GetLoopExtendNum)
201- uint16_t addrExpandCoef; // Address expansion coefficient
202- 
203- uint16_t msOffset; // MS ID offset (GetLoopMsOffset)
204- uint16_t ckeOffset; // CKE ID offset (GetLoopCKEOffset)
205- uint16_t xnIdOffset; // XN ID offset (GetLoopXnIdOffset)
206-};
207-```
208- 
209----
210- 
211-### P1-2: Cross-CCU Changes Should Be Intercepted in ResourceManager
212- 
213-**Severity**: P1 (Collection accuracy)
214- 
215-**Problem Description**:
216- 
217-In Section 4.1, the cross-CCU change collection point is a post-hoc call to `CaptureCrossCcuChanges()` in `CcuSimulator::ExecuteInstr()`. However, cross-CCU operations actually occur in the Executor's `Process()` method (e.g., `SyncCkeExecutor::Process()` calls `SetRmtCKESignal()`).
218- 
219-The post-hoc collection approach requires comparing the remote CCU's before/after state, meaning a remote CCU snapshot would be needed before and after instruction execution — high overhead and potentially missing changes (the remote CCU could be simultaneously modified by other code).
220- 
221-**Fix Suggestion**:
222- 
223-Intercept cross-CCU changes directly in `CcuResourceManager::UpdateCkeValue()`:
224-```cpp
225-void CcuResourceManager::UpdateCkeValue(int rankId, int dieId, uint16_t ckeId, uint16_t value) {
226- uint16_t oldValue = GetCkeValue(rankId, dieId, ckeId);
227- // Detect whether this is a cross-CCU operation
228- auto [execRank, execDie] = CcuTraceCollector::GetCurrentExecutingCcu();
229- if (rankId != execRank || dieId != execDie) {
230- CcuTraceCollector::RecordCrossCcuCkeChange(
231- rankId, dieId, ckeId, oldValue, value, execRank, execDie);
232- }
233- ckeData_[rankId][dieId][ckeId] = value;
234-}
235-```
236- 
237-Similarly applicable to `UpdateXnValue()` (TransXnToRmt operations), `UpdateGsaValue()`, `TransMSToMS()`, etc.
238- 
239----
240- 
241-### P1-3: CcuInstrTraceDetail Polymorphism Is Over-engineered
242- 
243-**Severity**: P1 (Architecture optimization)
244- 
245-**Problem Description**:
246- 
247-The current design has **13** `CcuInstrTraceDetail` subclasses (Load/Arith/Trans/Sync/Loop/LoopGroup/Cke/Jump/Wait/Fence/Reduce/...), but there are the following issues:
248- 
249-1. **Duplication with common layer**: `CcuTransTraceDetail`'s `waitCKEId/setCKEId` duplicates `CcuTraceEntry.waitInfo`; `CcuCkeTraceDetail`'s `ckeId/ckeValueBefore/ckeValueAfter` duplicates `resourceDelta.ckeChanges`
250-2. **Subclass proliferation**: The code has 30+ instruction types (~20 in V1 + ~10+ in V2); each requires a subclass
251-3. **Complex frontend rendering**: Requires dispatching different rendering templates based on typeName
252-4. **High cost of new instructions**: Each new instruction type requires a new subclass struct + CollectTraceDetail implementation
253- 
254-**Industry Reference**:
255- 
256-- **Chrome DevTools Trace Event Format**: Flat `args` field (key-value map), no polymorphic inheritance
257-- **LLVM Execution Trace**: Flat "additional info" field
258-- **GDB MI (Machine Interface)**: Key-value property list
259- 
260-**Fix Suggestion**:
261- 
262-Simplify `CcuInstrTraceDetail` to a flat structure:
263-```cpp
264-struct CcuInstrTraceDetail {
265- std::string typeName; // e.g., "TransLocMemToLocMem"
266- std::map<std::string, std::string> args; // Instruction-specific parameter key-value table
267- // Examples:
268- // Trans type: {"srcAddr": "0x7f0000", "dstAddr": "0x7f1000", "length": "4KB", "channelId": "0"}
269- // Reduce type: {"op": "Add", "msList": "[0,1,2,3]", "dataType": "FP32", "castEn": "0"}
270- // CKE type: {"ckeOp": "Set", "ckeId": "3", "ckeMask": "0x0001", "isRemote": "true"}
271-};
272-```
273- 
274-Advantages:
275-- No need to define a subclass for each instruction type
276-- New instructions only need to populate args in `CollectTraceDetail()`
277-- Unified frontend rendering logic (key-value table)
278-- `Describe()` output can be used directly as args source
279- 
280----
281- 
282-### P1-4: JSON Serialization Format vs Large Data Volume
283- 
284-**Severity**: P1 (Performance optimization)
285- 
286-**Problem Description**:
287- 
288-Section 3.4 adopts JSON as the serialization format, but JSON has weaknesses:
289-- Each uint64_t XN/GSA value requires string representation (`"0x7f3a0000"`), 8 bytes becomes ~16 bytes
290-- Large amounts of redundant key name repetition (`"xnChanges"` repeated in every entry)
291-- 10000 instructions' JSON file can reach 50MB+
292-- Frontend loading and parsing of large JSON files is seriously slow
293- 
294-**Industry Reference**:
295- 
296-| Tool | Format | Characteristics |
297-|------|------|------|
298-| Chrome DevTools | JSON streaming (one event per line) | Supports line-by-line parsing |
299-| Perfetto (Android) | Protobuf binary | Frontend WASM parsing, small size |
300-| rr debugger | Custom binary | Efficient recording/playback |
301-| Intel VTune | Custom binary + SQLite | On-demand querying |
302- 
303-**Fix Suggestion**:
304- 
305-Layered serialization strategy:
306-- **Collection side**: msgpack/flatbuffers binary format, 5-10x smaller, faster writes
307-- **Frontend loading**: Decode binary into frontend index structures via Worker
308-- **Export/debug**: Optional export to JSON (for human readability and debugging)
309-- JSON Schema `metadata.rankId/dieId` should be removed (changed to `rankSize/diePerRank`, consistent with global model)
310- 
311----
312- 
313-## III. Design Optimization Points (Recommended Improvements, Lower Priority)
314- 
315-### P2-1: Describe() Can Serve as MVP Trace Information Source
316- 
317-**Severity**: P2 (Implementation strategy)
318- 
319-**Problem Description**:
320- 
321-Every Executor in the code implements `Describe()`, returning formatted strings like:
322-```
323-"[Simulation Execute] Wait CKE[3:0001], Sync LocCKE[5:00ff] To rmtCKE[8:00ff]
324- Use Channel[0], Set CKE[10:0001], clearType[1]"
325-```
326- 
327-These strings already contain key instruction parameter information. The current design requires each Executor to add a `CollectTraceDetail()` virtual method and implement it, which is a significant effort (30+ instruction types).
328- 
329-**Fix Suggestion**:
330- 
331-Phased implementation strategy:
332- 
333-**Phase 1 (MVP)**:
334-- Collect only `Describe()` output + resource delta
335-- `instrDescribe` field directly uses `executor->Describe()` return value
336-- Do not implement `CollectTraceDetail()`; detail field is empty or contains only typeName
337- 
338-**Phase 2 (Enhancement)**:
339-- Implement `CollectTraceDetail()` for high-frequency/critical instruction types (e.g., Trans, Reduce, SyncCke)
340-- Detail uses flat key-value map (see P1-3)
341-- Progressively cover all instruction types
342- 
343-This enables rapid MVP delivery, reducing initial development cost.
344- 
345----
346- 
347-### P2-2: Notify and Other Non-CCU Tasks Not in Trace
348- 
349-**Severity**: P2 (Information gap)
350- 
351-**Problem Description**:
352- 
353-`SequentialExecutor::Execute()` includes task types beyond `CCU_GRAPH`, such as `NOTIFY_RECORD/NOTIFY_WAIT/REDUCE/MEM_CPY/AIV_GRAPH`. `NotifyWait` can also return `HOLD_CMD`, with similar blocking behavior to CCU's waitCKE.
354- 
355-The current trace only covers `CCU_GRAPH` type tasks. But Notify's Record/Wait affects execution flow:
356-- `NotifyWait` blocking causes the outer loop to skip that stream
357-- After `NotifyRecord` releases the block, the next round allows that stream's CCU task to continue
358- 
359-If not reflected in trace, the frontend replay will show "CCU instructions suddenly jumping to the next round" with no explanation of what happened in between.
360- 
361-**Fix Suggestion**:
362- 
363-Add simplified records for non-CCU task types in `globalEntries[]`:
364-```cpp
365-struct CcuTraceNonCcuEntry {
366- uint32_t globalSeqId;
367- uint32_t execRound;
368- int32_t rankId;
369- HccLTaskMetaType taskType; // NOTIFY_RECORD, NOTIFY_WAIT, REDUCE, MEM_CPY...
370- HcclVmResult execResult; // SUCCESS or HOLD_CMD
371- std::string description; // Task description (for frontend display)
372-};
373-```
374- 
375-During CCU switch notifications, the frontend can display: "Round 1: Rank1 NotifyWait blocked, skipped".
376- 
377----
378- 
379-### P2-3: JSON Schema Metadata Model Inconsistency
380- 
381-**Severity**: P2 (Model consistency)
382- 
383-**Problem Description**:
384- 
385-The JSON Schema example in Section 3.4 still includes `rankId`/`dieId` fields in `metadata` (per-CCU model), but Section 3.3.1's `CcuRunMetadata` has already been changed to `rankSize`/`diePerRank` (global model). The two are inconsistent.
386- 
387-Additionally, the JSON Schema example uses `"seqId"` instead of the new design's `"globalSeqId"`, and is missing `"execRound"`/`"sqeTaskId"`/`"ccuRegistry"`/`"sqeTaskRegistry"` and other new fields.
388- 
389-**Fix Suggestion**:
390- 
391-Update the JSON Schema example to be consistent with the data structure definitions in 3.3.1-3.3.2:
392-- Remove `metadata.rankId`/`metadata.dieId`
393-- Add `metadata.rankSize`/`metadata.diePerRank`/`metadata.totalExecRounds`
394-- Add `ccuRegistry[]` and `sqeTaskRegistry[]`
395-- Change `seqId` to `globalSeqId` in entries
396-- Add `execRound`/`sqeTaskId`/`rankId`/`dieId` fields
397- 
398----
399- 
400-## IV. Summary Table
401- 
402-| Priority | ID | Issue | Type | Fix Summary |
403-|--------|------|------|------|---------|
404-| P0 | P0-1 | Delta should not be computed during CKE Wait | Feasibility error | Record only key fields during waitCKE; create full entry only after CKE passes |
405-| P0 | P0-2 | globalSeqId/execRound maintenance location is wrong | Feasibility error | Global context managed by SequentialExecutor, passed to Simulator via parameters |
406-| P0 | P0-3 | Full snapshot performance bottleneck | Performance bottleneck | ResourceManager change interception + initial snapshot + periodic checkpoints |
407-| P1 | P1-1 | Loop offset formula incomplete | Information gap | Add iterStepGSA/curLoopCnt/extendNum/addrExpandCoef |
408-| P1 | P1-2 | Cross-CCU change collection timing | Collection accuracy | Intercept cross-CCU changes inside ResourceManager |
409-| P1 | P1-3 | Detail polymorphism over-engineered | Architecture optimization | Simplify to flat key-value map |
410-| P1 | P1-4 | JSON format unsuitable for large data | Performance optimization | Collection-side msgpack/flatbuffers, frontend WASM parsing |
411-| P2 | P2-1 | Describe() can serve as MVP source | Implementation strategy | Phase 1: only Describe() + delta |
412-| P2 | P2-2 | Notify and other non-CCU tasks missing | Information gap | Add simplified records |
413-| P2 | P2-3 | JSON Schema model inconsistency | Model consistency | Align with 3.3.1-3.3.2 |
414- 
415----
416- 
417-## V. Fix Priority Recommendations
418- 
419-**Step 1 (Required)**: Fix P0-1, P0-2, P0-3; otherwise the approach cannot be implemented
420- 
421-**Step 2 (Recommended)**: Fix P1-1, P1-2, P1-3; improve approach quality and practicality
422- 
423-**Step 3 (Optional)**: Fix P1-4, P2-1, P2-2, P2-3; refine details and user experience
Mtest/hccl_vm/include/runnerdb/db_sim_communicator.h+18-9
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef DB_SIM_COMMUNICATOR_H18#ifndef DB_SIM_COMMUNICATOR_H
@@ -22,10 +29,12 @@ struct CommunicatorMemberInfo {
22 uint32_t rankId;29 uint32_t rankId;
23};30};
24 31 
25-bool GetCommunicatorMembers(uint64_t commId, std::vector<CommunicatorMemberInfo>& members);32+bool GetCommunicatorMembers(uint64_t commId,
33+ std::vector<CommunicatorMemberInfo> &members);
26// 根据通信域成员 ID 查询其所属通信域名称。34// 根据通信域成员 ID 查询其所属通信域名称。
27-bool GetCommunicatorName(uint64_t commId, std::string& commName);35+bool GetCommunicatorName(uint64_t commId, std::string &commName);
28-bool GetCommunicatorIdentity(uint64_t commId, std::string& commName, uint64_t& commHash);36+bool GetCommunicatorIdentity(uint64_t commId, std::string &commName,
37+ uint64_t &commHash);
29} // namespace sim38} // namespace sim
30 39 
31#endif // DB_SIM_COMMUNICATOR_H40#endif // DB_SIM_COMMUNICATOR_H
Mtest/hccl_vm/include/runnerdb/db_sim_op_db_ops.h+72-41
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_OP_DB_OPS_H18#ifndef SIM_OP_DB_OPS_H
@@ -20,52 +27,76 @@
20 27 
21namespace sim {28namespace sim {
22extern uint32_t g_currOpDetailId;29extern uint32_t g_currOpDetailId;
30+extern uint32_t g_currOpMemId;
31+ 
32+/**
33+ * @brief 当前进程正在处理的算子迭代（= 最近一次 InsertOpDetail 分配的
34+ * opIter）。
35+ *
36+ * 用途：CCU 交换表（CcuSyncResTab）需要跨 rank 一致的 op
37+ * 身份来隔离不同算子的交换值。 opIter 由 InsertOpDetail 按 (commName, commHash)
38+ * 自增，两侧进程按同一算子序列 推进时天然对齐，可直接当 op 身份使用。
39+ *
40+ * @return 最新分配的 opIter；尚无算子插入时为 0。
41+ */
42+uint32_t GetCurOpIter();
23 43 
24int InitOpDataDb();44int InitOpDataDb();
25-int SetDbConfig(DBConfig& config);45+int SetDbConfig(DBConfig &config);
26 46 
27void AddSyncStreamIter(uint64_t streamId);47void AddSyncStreamIter(uint64_t streamId);
28void RemoveSyncStreamIter(uint64_t streamId);48void RemoveSyncStreamIter(uint64_t streamId);
29-bool NextStreamSyncIdx(uint64_t streamId, uint64_t& syncIdx);49+bool NextStreamSyncIdx(uint64_t streamId, uint64_t &syncIdx);
30 50 
31-int InsertOpDetail(OpDetailTab& rec);51+int InsertOpDetail(OpDetailTab &rec);
32-int InsertOpMem(OpMemInfoTab& rec);52+int InsertOpMem(OpMemInfoTab &rec);
33-int InsertOpDetailAndMem(OpDetailTab& detail, OpMemInfoTab& mem);53+int InsertOpDetailAndMem(OpDetailTab &detail, OpMemInfoTab &mem);
34-int InsertCcuChannel(CcuChannelTab& rec);54+int InsertCcuChannel(CcuChannelTab &rec);
35-int InsertJettyMap(JettyMapTab& rec);55+int InsertJettyMap(JettyMapTab &rec);
36-int InsertOpTask(OpTaskTab& rec, bool isDevice = false);56+int InsertOpTask(OpTaskTab &rec, bool isDevice = false);
37-int InsertSyncRecord(SyncRecordTab& rec);57+int InsertSyncRecord(SyncRecordTab &rec);
38-int InsertCcuInstrRes(CcuInstrResTab& rec);58+int InsertCcuInstrRes(CcuInstrResTab &rec);
39-int InsertCcuInstr(CcuInstrTab& rec);59+int InsertCcuInstr(CcuInstrTab &rec);
40-int InsertHalfRTT(HalfRTTTab& rec);60+int InsertHalfRTT(HalfRTTTab &rec);
41 61 
42-int UpdateAndInsertByCcuId(62+int UpdateAndInsertByCcuId(uint64_t &ccuId, uint32_t deviceId, uint32_t dieId,
43- uint64_t& ccuId, uint32_t deviceId, uint32_t dieId, uint32_t startId, uint32_t instrCount, uint32_t instrOffset,63+ uint32_t startId, uint32_t instrCount,
44- uint32_t instrInfoSize, const void* instrInfo, std::vector<uint8_t>* mergedInstrSpace = nullptr,64+ uint32_t instrOffset, uint32_t instrInfoSize,
45- uint32_t* totalInstrCount = nullptr);65+ const void *instrInfo,
46-int UpdateSyncRecordStatus(std::vector<SyncRecordTab>& syncRecord);66+ std::vector<uint8_t> *mergedInstrSpace = nullptr,
67+ uint32_t *totalInstrCount = nullptr);
68+int UpdateSyncRecordStatus(std::vector<SyncRecordTab> &syncRecord);
47int UpdateOpMemCclBuffer(uint64_t cclAddr, uint64_t cclSize);69int UpdateOpMemCclBuffer(uint64_t cclAddr, uint64_t cclSize);
48int UpdateOpExpansionMode(uint8_t mode);70int UpdateOpExpansionMode(uint8_t mode);
71+int UpdateOpDetailMainStream(uint32_t opDetailId, uint64_t streamId);
49 72 
50int QueryLatestOpExpansionMode();73int QueryLatestOpExpansionMode();
51-int QueryCcuChannelAll(std::vector<CcuChannelTab>& out);74+int QueryCcuChannelAll(std::vector<CcuChannelTab> &out);
52-int QueryJettyMapAll(std::vector<JettyMapTab>& out);75+int QueryJettyMapAll(std::vector<JettyMapTab> &out);
53-int QuerySyncRecordAll(std::vector<SyncRecordTab>& out);76+int QuerySyncRecordAll(std::vector<SyncRecordTab> &out);
54-int QueryCcuInstrResAll(std::vector<CcuInstrResTab>& out);77+int QueryCcuInstrResAll(std::vector<CcuInstrResTab> &out);
55-int QueryCcuResourceMetaCount(uint32_t& instrLoadCnt, uint32_t& channelCnt);78+int QueryCcuResourceMetaCount(uint32_t &instrLoadCnt, uint32_t &channelCnt);
56-int QueryOpDetailIdentity(uint32_t opDetailId, uint64_t& commId, uint32_t& rankId, uint32_t& deviceId);79+int QueryOpDetailIdentity(uint32_t opDetailId, uint64_t &commId,
57-int QueryCurrentOpMemInfo(uint64_t commId, uint32_t deviceId, OpMemInfoTab& out);80+ uint32_t &rankId, uint32_t &deviceId);
58-int QuerySyncRecordByStatus(uint8_t status, std::vector<SyncRecordTab>& out);81+int QueryOpDetailById(uint32_t opDetailId, OpDetailTab &out);
59-int QueryOpExecutionIndexEntries(std::vector<OpExecutionIndexEntry>& out);82+int QueryLatestOpDetailByDeviceId(uint32_t deviceId, OpDetailTab &out);
60-int QueryCompositeOpDetailBySyncIter(uint32_t syncIter, std::map<uint32_t, std::vector<CompositeOpDetail>>& detail);83+int QueryCurrentOpMemInfo(uint64_t commId, uint32_t deviceId,
61-int QueryHalfRTTAll(std::vector<HalfRTTTab>& out);84+ OpMemInfoTab &out);
62-int QueryOpTaskTabNames(std::vector<std::string>& names);85+int QueryOpMemInfoByOpDetailId(uint32_t opDetailId, OpMemInfoTab &out);
63-int QueryCompositeOpDetailByOpIter(86+int QuerySyncRecordByStatus(uint8_t status, std::vector<SyncRecordTab> &out);
64- const std::string& commName, uint64_t commHash, uint32_t opIter, std::vector<CompositeOpDetail>& details);87+int QueryOpExecutionIndexEntries(std::vector<OpExecutionIndexEntry> &out);
65-int QueryAllOpTasks(std::vector<OpTaskTab>& tasks, bool filterDone = false);88+int QueryCompositeOpDetailBySyncIter(
66-int QueryOpTasksByStreamId(uint64_t streamId, std::vector<OpTaskTab>& tasks);89+ uint32_t syncIter,
67-int FinishOpTask(const OpTaskTab& task);90+ std::map<uint32_t, std::vector<CompositeOpDetail>> &detail);
68-int QueryHalfRTTAll(std::vector<HalfRTTTab>& out);91+int QueryHalfRTTAll(std::vector<HalfRTTTab> &out);
92+int QueryOpTaskTabNames(std::vector<std::string> &names);
93+int QueryCompositeOpDetailByOpIter(const std::string &commName,
94+ uint64_t commHash, uint32_t opIter,
95+ std::vector<CompositeOpDetail> &details);
96+int QueryAllOpTasks(std::vector<OpTaskTab> &tasks, bool filterDone = false);
97+int QueryOpTasksByStreamId(uint64_t streamId, std::vector<OpTaskTab> &tasks);
98+int FinishOpTask(const OpTaskTab &task);
99+int QueryHalfRTTAll(std::vector<HalfRTTTab> &out);
69} // namespace sim100} // namespace sim
70 101 
71#endif102#endif
Mtest/hccl_vm/include/runnerdb/db_sim_runner_common.h+51-33
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef _SIM_RUNNER_COMMOM_H_18#ifndef _SIM_RUNNER_COMMOM_H_
@@ -30,18 +37,20 @@ namespace sim {
30// 通信域server信息: 由TopoMetaConfig表项指向的topo_meta配置解析生成,37// 通信域server信息: 由TopoMetaConfig表项指向的topo_meta配置解析生成,
31// 按rankId分配顺序排列38// 按rankId分配顺序排列
32struct CommConfigServer {39struct CommConfigServer {
33- uint64_t serverKey{0}; // Server表主键40+ uint64_t serverKey{0}; // Server表主键
34- uint32_t podId{0}; // 超级节点id41+ uint32_t podId{0}; // 超级节点id
35- uint32_t serverId{0}; // server id42+ uint32_t serverId{0}; // server id
36- uint32_t devNum{0}; // 通信域内该server的device数43+ uint32_t devNum{0}; // 通信域内该server的device数
37- uint32_t rankOffset{0}; // 该server首个rankId(前面server的device累计数)44+ uint32_t rankOffset{0}; // 该server首个rankId(前面server的device累计数)
38 std::vector<PhyDeviceId> phyIds; // server内按rank顺序的物理device id45 std::vector<PhyDeviceId> phyIds; // server内按rank顺序的物理device id
39};46};
40 47 
41// 进程级通信域配置缓存(TopoMetaConfig指向的配置解析结果)48// 进程级通信域配置缓存(TopoMetaConfig指向的配置解析结果)
42struct CommConfigData {49struct CommConfigData {
43- std::string fileName; // topo_meta yaml文件名; "ranktable"表示ranktable.json模式50+ std::string
44- std::vector<CommConfigServer> servers; // 按(podId, serverId)升序, 与rankId分配顺序一致51+ fileName; // topo_meta yaml文件名; "ranktable"表示ranktable.json模式
52+ std::vector<CommConfigServer>
53+ servers; // 按(podId, serverId)升序, 与rankId分配顺序一致
45};54};
46} // namespace sim55} // namespace sim
47 56 
@@ -52,48 +61,57 @@ extern sim::CommConfigData g_comm_config_data;
52namespace sim {61namespace sim {
53// 仅供仍需"查找或复用"通信域成员的旧调用方使用; 主/子通信域创建路径均直接插入本62// 仅供仍需"查找或复用"通信域成员的旧调用方使用; 主/子通信域创建路径均直接插入本
54// rank 的表行.63// rank 的表行.
55-bool GetOrInsertCommunicator(64+bool GetOrInsertCommunicator(const char *commName, uint32_t rankSize,
56- const char* commName, uint32_t rankSize, uint32_t rankId, uint64_t deviceId, uint64_t commHash, uint64_t& commId);65+ uint32_t rankId, uint64_t deviceId,
57-bool WaitCommunicatorReady(const char* commName, uint64_t commHash, uint32_t rankSize);66+ uint64_t commHash, uint64_t &commId);
67+bool WaitCommunicatorReady(const char *commName, uint64_t commHash,
68+ uint32_t rankSize);
58// level 1 专用69// level 1 专用
59bool WaitCommunicatorDestroyReady(uint64_t commId);70bool WaitCommunicatorDestroyReady(uint64_t commId);
60// level 0 专用: 避免快卡提前销毁, 标记本 rank71// level 0 专用: 避免快卡提前销毁, 标记本 rank
61// 已调用销毁接口并等待同通信域全部成员到达后再返回72// 已调用销毁接口并等待同通信域全部成员到达后再返回
62bool MarkAndSyncCommunicatorDestroy(uint64_t commId);73bool MarkAndSyncCommunicatorDestroy(uint64_t commId);
63 74 
64-aclError GetDeviceByLogicId(uint32_t deviceId, sim::Device& device);75+aclError GetDeviceByLogicId(uint32_t deviceId, sim::Device &device);
65-aclError GetDeviceByCommRank(uint64_t commId, uint32_t rankId, sim::Device& device);76+aclError GetDeviceByCommRank(uint64_t commId, uint32_t rankId,
66-aclError GetDeviceByPhysicalId(uint32_t deviceId, sim::Device& device);77+ sim::Device &device);
67-aclError UpdateDeviceLogicId(uint64_t serverKey, uint32_t phyDevId, uint32_t logicDevId, uint32_t userId);78+aclError GetDeviceByPhysicalId(uint32_t deviceId, sim::Device &device);
79+aclError UpdateDeviceLogicId(uint64_t serverKey, uint32_t phyDevId,
80+ uint32_t logicDevId, uint32_t userId);
68aclError UpdateSuperDeviceId(uint32_t logicDevId, uint32_t superDeviceId);81aclError UpdateSuperDeviceId(uint32_t logicDevId, uint32_t superDeviceId);
69-aclError GetCcuFromDeviceByDieId(uint64_t deviceKey, uint8_t dieId, sim::Ccu& ccu);82+aclError GetCcuFromDeviceByDieId(uint64_t deviceKey, uint8_t dieId,
70-aclError GetCcuResourceByCcu(uint64_t ccuKey, sim::CcuResource& ccuRes);83+ sim::Ccu &ccu);
71-aclError GetContextByDevId(uint32_t deviceId, sim::Context& context);84+aclError GetContextByDevId(uint32_t deviceId, sim::Context &context);
72-aclError GetPortByName(uint64_t serverKey, uint32_t phyDevId, const std::string& name, sim::Port& port);85+aclError GetPortByName(uint64_t serverKey, uint32_t phyDevId,
73-aclError GetEndPointByIpAddr(const std::string& ip, sim::EndPoint& endPoint);86+ const std::string &name, sim::Port &port);
74-aclError GetEndPointByEid(const IpAddress& addr, sim::EndPoint& endPoint);87+aclError GetEndPointByIpAddr(const std::string &ip, sim::EndPoint &endPoint);
75-aclError GetPortById(uint64_t portId, sim::Port& port);88+aclError GetEndPointByEid(const IpAddress &addr, sim::EndPoint &endPoint);
89+aclError GetPortById(uint64_t portId, sim::Port &port);
76 90 
77uint32_t GetAICpuCount(uint64_t deviceId);91uint32_t GetAICpuCount(uint64_t deviceId);
78uint32_t GetAICoreCount(uint64_t deviceId);92uint32_t GetAICoreCount(uint64_t deviceId);
79uint32_t GetVectorCoreCount(uint64_t deviceId);93uint32_t GetVectorCoreCount(uint64_t deviceId);
80std::string GetHardwareTypeByDevice(uint64_t deviceId);94std::string GetHardwareTypeByDevice(uint64_t deviceId);
81-bool GetCommRankByDeviceId(uint64_t commId, uint32_t deviceId, uint32_t& rankId);95+bool GetCommRankByDeviceId(uint64_t commId, uint32_t deviceId,
96+ uint32_t &rankId);
82bool ResetAllDeviceLogicId();97bool ResetAllDeviceLogicId();
83-bool GetRankIdByMPI(uint32_t& rankId, uint64_t& serverId);98+bool GetRankIdByMPI(uint32_t &rankId, uint64_t &serverId);
84uint64_t GetCurServerId();99uint64_t GetCurServerId();
85 100 
86// 公共解析函数: 解析topo_meta101// 公共解析函数: 解析topo_meta
87// yaml(或ranktable模式下的data/ranktable.json)生成通信域server有序表.102// yaml(或ranktable模式下的data/ranktable.json)生成通信域server有序表.
88// topoMeta非空时直接使用调用方已解析的结果, 避免重复读文件103// topoMeta非空时直接使用调用方已解析的结果, 避免重复读文件
89-bool ParseCommConfigData(const std::string& fileName, const TopoMeta* topoMeta, CommConfigData& data);104+bool ParseCommConfigData(const std::string &fileName, const TopoMeta *topoMeta,
105+ CommConfigData &data);
90// mock-comm每次执行时调用: 覆盖TopoMetaConfig表项并刷新本进程g_comm_config_data106// mock-comm每次执行时调用: 覆盖TopoMetaConfig表项并刷新本进程g_comm_config_data
91-bool RefreshTopoMetaConfig(const std::string& fileName, const TopoMeta* topoMeta = nullptr);107+bool RefreshTopoMetaConfig(const std::string &fileName,
108+ const TopoMeta *topoMeta = nullptr);
92// 通信域重置时清空进程缓存, 下次使用时按TopoMetaConfig表项重新加载109// 通信域重置时清空进程缓存, 下次使用时按TopoMetaConfig表项重新加载
93void ResetCommConfigData();110void ResetCommConfigData();
94// rankId → 所属server的serverKey111// rankId → 所属server的serverKey
95-bool GetServerKeyByRankId(uint32_t rankId, uint64_t& serverKey);112+bool GetServerKeyByRankId(uint32_t rankId, uint64_t &serverKey);
96// serverKey + 物理device id → 通信域内device序号(与Device.logic_id一致)113// serverKey + 物理device id → 通信域内device序号(与Device.logic_id一致)
97-aclError GetDevIndexByPhyId(uint64_t serverKey, uint32_t phyId, uint32_t& devIndex);114+aclError GetDevIndexByPhyId(uint64_t serverKey, uint32_t phyId,
115+ uint32_t &devIndex);
98} // namespace sim116} // namespace sim
99#endif // _SIM_RUNNER_COMMOM_H_117#endif // _SIM_RUNNER_COMMOM_H_
Mtest/hccl_vm/include/runnerdb/db_sim_runner_db.h+23-18
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_RUNNER_DB_H18#ifndef SIM_RUNNER_DB_H
@@ -18,26 +25,24 @@
18#include "sim_models.h"25#include "sim_models.h"
19#include "sim_runner_impl.inl"26#include "sim_runner_impl.inl"
20namespace RunnerDB {27namespace RunnerDB {
21-template <typename T>28+template <typename T> uint64_t Add(T &rec);
22-uint64_t Add(T& rec);29+ 
30+template <typename T> std::vector<uint64_t> AddBatch(std::vector<T> &recs);
31+ 
32+template <typename T> std::optional<T> GetById(uint64_t id);
23 33 
24template <typename T>34template <typename T>
25-std::optional<T> GetById(uint64_t id);35+std::vector<T> GetByPred(std::function<bool(const T &)> pred);
26 36 
27template <typename T>37template <typename T>
28-std::vector<T> GetByPred(std::function<bool(const T&)> pred);38+std::pair<T, bool> GetOneByPred(std::function<bool(const T &)> pred);
29 39 
30template <typename T>40template <typename T>
31-std::pair<T, bool> GetOneByPred(std::function<bool(const T&)> pred);41+bool Update(uint64_t id, std::function<void(T &)> updater);
32 42 
33-template <typename T>43+template <typename T> bool Delete(uint64_t id);
34-bool Update(uint64_t id, std::function<void(T&)> updater);
35 44 
36-template <typename T>45+template <typename T> bool DeleteAll();
37-bool Delete(uint64_t id);
38- 
39-template <typename T>
40-bool DeleteAll();
41 46 
42std::vector<std::string> GetAllTableName();47std::vector<std::string> GetAllTableName();
43} // namespace RunnerDB48} // namespace RunnerDB
Mtest/hccl_vm/include/runnerdb/db_sim_runner_ops.h+14-7
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef _SIM_RUNNER_OPS_H_18#ifndef _SIM_RUNNER_OPS_H_
@@ -17,7 +24,7 @@
17#include "sim_models.h"24#include "sim_models.h"
18 25 
19namespace sim {26namespace sim {
20-bool GetCurrRunnerTls(uint64_t serverKey, Runner& runner);27+bool GetCurrRunnerTls(uint64_t serverKey, Runner &runner);
21bool SetCurrCtxTls(uint64_t ctx);28bool SetCurrCtxTls(uint64_t ctx);
22uint64_t GetCurrDeviceId();29uint64_t GetCurrDeviceId();
23uint64_t GetCurrDeviceKey();30uint64_t GetCurrDeviceKey();
Mtest/hccl_vm/include/runnerdb/db_sim_sqlite_db.h+64-78
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_SQLITE_DB_H18#ifndef SIM_SQLITE_DB_H
@@ -35,7 +42,7 @@ template <typename T>
35using TableValue = typename sim::SqliteTable<T>::ValueType;42using TableValue = typename sim::SqliteTable<T>::ValueType;
36 43 
37struct SimRunnerSqliteDB {44struct SimRunnerSqliteDB {
38-private:45+ private:
39 sim::SqliteDatabase m_db;46 sim::SqliteDatabase m_db;
40 47 
41 sim::SqliteTable<sim::Server> m_serverTbl;48 sim::SqliteTable<sim::Server> m_serverTbl;
@@ -59,7 +66,6 @@ private:
59 sim::SqliteTable<sim::EventSyncTask> m_eventSyncTaskTbl;66 sim::SqliteTable<sim::EventSyncTask> m_eventSyncTaskTbl;
60 sim::SqliteTable<sim::Port> m_portTbl;67 sim::SqliteTable<sim::Port> m_portTbl;
61 sim::SqliteTable<sim::Ccu> m_ccuTbl;68 sim::SqliteTable<sim::Ccu> m_ccuTbl;
62- sim::SqliteTable<sim::CcuResource> m_ccuResTbl;
63 sim::SqliteTable<sim::DeviceConnection> m_deviceConnTbl;69 sim::SqliteTable<sim::DeviceConnection> m_deviceConnTbl;
64 sim::SqliteTable<sim::EndPointPair> m_endPointPairTbl;70 sim::SqliteTable<sim::EndPointPair> m_endPointPairTbl;
65 sim::SqliteTable<sim::EndPointPortMapping> m_endPointPortMappingTbl;71 sim::SqliteTable<sim::EndPointPortMapping> m_endPointPortMappingTbl;
@@ -103,23 +109,22 @@ private:
103 sim::SqliteTable<sim::Plugin> m_pluginTbl;109 sim::SqliteTable<sim::Plugin> m_pluginTbl;
104 sim::SqliteTable<sim::DpuPendingNotify> m_dpuPendingNotifyTbl;110 sim::SqliteTable<sim::DpuPendingNotify> m_dpuPendingNotifyTbl;
105 sim::SqliteTable<sim::DpuDeviceInfo> m_dpuDeviceInfoTbl;111 sim::SqliteTable<sim::DpuDeviceInfo> m_dpuDeviceInfoTbl;
112+ sim::SqliteTable<sim::CcuSyncResTab> m_ccuSyncResTbl;
106 113 
107 std::mutex m_lazyMutex;114 std::mutex m_lazyMutex;
108- std::unordered_map<std::type_index, sim::TableBase*> m_tableMap;115+ std::unordered_map<std::type_index, sim::TableBase *> m_tableMap;
109 std::vector<std::string> m_tableNames;116 std::vector<std::string> m_tableNames;
110 std::vector<std::unique_ptr<sim::TableBase>> m_lazyTables;117 std::vector<std::unique_ptr<sim::TableBase>> m_lazyTables;
111 118 
112 template <typename T>119 template <typename T>
113- void RegisterTable(sim::SqliteTable<T>& table, const std::string& name)120+ void RegisterTable(sim::SqliteTable<T> &table, const std::string &name) {
114- {
115 m_tableMap[std::type_index(typeid(T))] = &table;121 m_tableMap[std::type_index(typeid(T))] = &table;
116 m_tableNames.push_back(name);122 m_tableNames.push_back(name);
117 }123 }
118 124 
119-public:125+ public:
120 SimRunnerSqliteDB()126 SimRunnerSqliteDB()
121- : m_serverTbl(m_db.GetDb(), "Server"),127+ : m_serverTbl(m_db.GetDb(), "Server"), m_hostTbl(m_db.GetDb(), "Host"),
122- m_hostTbl(m_db.GetDb(), "Host"),
123 m_runnerTbl(m_db.GetDb(), "Runner"),128 m_runnerTbl(m_db.GetDb(), "Runner"),
124 m_communicatorTbl(m_db.GetDb(), "Communicator"),129 m_communicatorTbl(m_db.GetDb(), "Communicator"),
125 m_communicatorDestroySyncTbl(m_db.GetDb(), "CommunicatorDestroySync"),130 m_communicatorDestroySyncTbl(m_db.GetDb(), "CommunicatorDestroySync"),
@@ -137,9 +142,7 @@ public:
137 m_fdMemWhiteListTbl(m_db.GetDb(), "FdMemWhiteList"),142 m_fdMemWhiteListTbl(m_db.GetDb(), "FdMemWhiteList"),
138 m_taskTbl(m_db.GetDb(), "Task"),143 m_taskTbl(m_db.GetDb(), "Task"),
139 m_eventSyncTaskTbl(m_db.GetDb(), "EventSyncTask"),144 m_eventSyncTaskTbl(m_db.GetDb(), "EventSyncTask"),
140- m_portTbl(m_db.GetDb(), "Port"),145+ m_portTbl(m_db.GetDb(), "Port"), m_ccuTbl(m_db.GetDb(), "Ccu"),
141- m_ccuTbl(m_db.GetDb(), "Ccu"),
142- m_ccuResTbl(m_db.GetDb(), "CcuResource"),
143 m_deviceConnTbl(m_db.GetDb(), "DeviceConnection"),146 m_deviceConnTbl(m_db.GetDb(), "DeviceConnection"),
144 m_endPointTbl(m_db.GetDb(), "EndPoint"),147 m_endPointTbl(m_db.GetDb(), "EndPoint"),
145 m_endPointPairTbl(m_db.GetDb(), "EndPointPair"),148 m_endPointPairTbl(m_db.GetDb(), "EndPointPair"),
@@ -151,10 +154,8 @@ public:
151 m_raSocketTbl(m_db.GetDb(), "RaSocket"),154 m_raSocketTbl(m_db.GetDb(), "RaSocket"),
152 m_raSocketPairTbl(m_db.GetDb(), "RaSocketPair"),155 m_raSocketPairTbl(m_db.GetDb(), "RaSocketPair"),
153 m_raDeviceTbl(m_db.GetDb(), "RaDevice"),156 m_raDeviceTbl(m_db.GetDb(), "RaDevice"),
154- m_raQPTbl(m_db.GetDb(), "RaQP"),157+ m_raQPTbl(m_db.GetDb(), "RaQP"), m_raCQTbl(m_db.GetDb(), "RaCQ"),
155- m_raCQTbl(m_db.GetDb(), "RaCQ"),158+ m_raCQETbl(m_db.GetDb(), "RaCQE"), m_raMRTbl(m_db.GetDb(), "RaMR"),
156- m_raCQETbl(m_db.GetDb(), "RaCQE"),
157- m_raMRTbl(m_db.GetDb(), "RaMR"),
158 m_memoryLayoutTbl(m_db.GetDb(), "MemoryLayout"),159 m_memoryLayoutTbl(m_db.GetDb(), "MemoryLayout"),
159 m_simModelDataTbl(m_db.GetDb(), "SimModelData"),160 m_simModelDataTbl(m_db.GetDb(), "SimModelData"),
160 m_ipcNotifyTbl(m_db.GetDb(), "IpcNotify"),161 m_ipcNotifyTbl(m_db.GetDb(), "IpcNotify"),
@@ -164,12 +165,10 @@ public:
164 m_raContextTbl(m_db.GetDb(), "RaContext"),165 m_raContextTbl(m_db.GetDb(), "RaContext"),
165 m_raChanTbl(m_db.GetDb(), "RaChan"),166 m_raChanTbl(m_db.GetDb(), "RaChan"),
166 m_raTokenIdTbl(m_db.GetDb(), "RaTokenId"),167 m_raTokenIdTbl(m_db.GetDb(), "RaTokenId"),
167- m_raTpTbl(m_db.GetDb(), "RaTp"),168+ m_raTpTbl(m_db.GetDb(), "RaTp"), m_raRmemTbl(m_db.GetDb(), "RaRmem"),
168- m_raRmemTbl(m_db.GetDb(), "RaRmem"),
169 m_raLmemTbl(m_db.GetDb(), "RaLmem"),169 m_raLmemTbl(m_db.GetDb(), "RaLmem"),
170 m_raJettyTbl(m_db.GetDb(), "RaJetty"),170 m_raJettyTbl(m_db.GetDb(), "RaJetty"),
171- m_raJfcTbl(m_db.GetDb(), "RaJfc"),171+ m_raJfcTbl(m_db.GetDb(), "RaJfc"), m_raCrTbl(m_db.GetDb(), "RaCr"),
172- m_raCrTbl(m_db.GetDb(), "RaCr"),
173 m_raTlvTbl(m_db.GetDb(), "RaTlv"),172 m_raTlvTbl(m_db.GetDb(), "RaTlv"),
174 m_runModeConfigTbl(m_db.GetDb(), "RunModeConfig"),173 m_runModeConfigTbl(m_db.GetDb(), "RunModeConfig"),
175 m_hcclBufferTbl(m_db.GetDb(), "HcclBuffer"),174 m_hcclBufferTbl(m_db.GetDb(), "HcclBuffer"),
@@ -182,8 +181,8 @@ public:
182 m_topoMetaConfigTbl(m_db.GetDb(), "TopoMetaConfig"),181 m_topoMetaConfigTbl(m_db.GetDb(), "TopoMetaConfig"),
183 m_pluginTbl(m_db.GetDb(), "Plugin"),182 m_pluginTbl(m_db.GetDb(), "Plugin"),
184 m_dpuPendingNotifyTbl(m_db.GetDb(), "DpuPendingNotify"),183 m_dpuPendingNotifyTbl(m_db.GetDb(), "DpuPendingNotify"),
185- m_dpuDeviceInfoTbl(m_db.GetDb(), "DpuDeviceInfo")184+ m_dpuDeviceInfoTbl(m_db.GetDb(), "DpuDeviceInfo"),
186- {185+ m_ccuSyncResTbl(m_db.GetDb(), "CcuSyncRes") {
187 RegisterTable(m_serverTbl, "Server");186 RegisterTable(m_serverTbl, "Server");
188 RegisterTable(m_hostTbl, "Host");187 RegisterTable(m_hostTbl, "Host");
189 RegisterTable(m_runnerTbl, "Runner");188 RegisterTable(m_runnerTbl, "Runner");
@@ -205,7 +204,6 @@ public:
205 RegisterTable(m_eventSyncTaskTbl, "EventSyncTask");204 RegisterTable(m_eventSyncTaskTbl, "EventSyncTask");
206 RegisterTable(m_portTbl, "Port");205 RegisterTable(m_portTbl, "Port");
207 RegisterTable(m_ccuTbl, "Ccu");206 RegisterTable(m_ccuTbl, "Ccu");
208- RegisterTable(m_ccuResTbl, "CcuResource");
209 RegisterTable(m_deviceConnTbl, "DeviceConnection");207 RegisterTable(m_deviceConnTbl, "DeviceConnection");
210 RegisterTable(m_endPointPairTbl, "EndPointPair");208 RegisterTable(m_endPointPairTbl, "EndPointPair");
211 RegisterTable(m_endPointPortMappingTbl, "EndPointPortMapping");209 RegisterTable(m_endPointPortMappingTbl, "EndPointPortMapping");
@@ -249,17 +247,17 @@ public:
249 RegisterTable(m_pluginTbl, "Plugin");247 RegisterTable(m_pluginTbl, "Plugin");
250 RegisterTable(m_dpuPendingNotifyTbl, "DpuPendingNotify");248 RegisterTable(m_dpuPendingNotifyTbl, "DpuPendingNotify");
251 RegisterTable(m_dpuDeviceInfoTbl, "DpuDeviceInfo");249 RegisterTable(m_dpuDeviceInfoTbl, "DpuDeviceInfo");
250+ RegisterTable(m_ccuSyncResTbl, "CcuSyncRes");
252 }251 }
253 252 
254- SimRunnerSqliteDB(const SimRunnerSqliteDB&) = delete;253+ SimRunnerSqliteDB(const SimRunnerSqliteDB &) = delete;
255- SimRunnerSqliteDB& operator=(const SimRunnerSqliteDB&) = delete;254+ SimRunnerSqliteDB &operator=(const SimRunnerSqliteDB &) = delete;
256- SimRunnerSqliteDB(SimRunnerSqliteDB&&) = delete;255+ SimRunnerSqliteDB(SimRunnerSqliteDB &&) = delete;
257- SimRunnerSqliteDB& operator=(SimRunnerSqliteDB&&) = delete;256+ SimRunnerSqliteDB &operator=(SimRunnerSqliteDB &&) = delete;
258 257 
259- static SimRunnerSqliteDB& Instance()258+ static SimRunnerSqliteDB &Instance() {
260- {
261 static std::mutex s_mtx;259 static std::mutex s_mtx;
262- static SimRunnerSqliteDB* s_instance = nullptr;260+ static SimRunnerSqliteDB *s_instance = nullptr;
263 261 
264 std::lock_guard<std::mutex> lock(s_mtx);262 std::lock_guard<std::mutex> lock(s_mtx);
265 if (s_instance == nullptr) {263 if (s_instance == nullptr) {
@@ -268,73 +266,62 @@ public:
268 return *s_instance;266 return *s_instance;
269 }267 }
270 268 
271- template <typename T>269+ template <typename T> sim::SqliteTable<T> &GetTable() {
272- sim::SqliteTable<T>& GetTable()
273- {
274 std::lock_guard<std::mutex> lock(m_lazyMutex);270 std::lock_guard<std::mutex> lock(m_lazyMutex);
275 auto it = m_tableMap.find(std::type_index(typeid(T)));271 auto it = m_tableMap.find(std::type_index(typeid(T)));
276 if (it != m_tableMap.end()) {272 if (it != m_tableMap.end()) {
277- return *static_cast<sim::SqliteTable<T>*>(it->second);273+ return *static_cast<sim::SqliteTable<T> *>(it->second);
278 }274 }
279 std::string name = std::string("Unregistered_") + typeid(T).name();275 std::string name = std::string("Unregistered_") + typeid(T).name();
280- HCCL_VM_WARN(276+ HCCL_VM_WARN("[SimRunnerSqliteDB::GetTable] Lazily registering "
281- "[SimRunnerSqliteDB::GetTable] Lazily registering "277+ "unregistered type: {}",
282- "unregistered type: {}",278+ name.c_str());
283- name.c_str());
284 auto table = std::make_unique<sim::SqliteTable<T>>(m_db.GetDb(), name);279 auto table = std::make_unique<sim::SqliteTable<T>>(m_db.GetDb(), name);
285- auto* rawPtr = table.get();280+ auto *rawPtr = table.get();
286 m_lazyTables.push_back(std::move(table));281 m_lazyTables.push_back(std::move(table));
287 m_tableMap[std::type_index(typeid(T))] = rawPtr;282 m_tableMap[std::type_index(typeid(T))] = rawPtr;
288 m_tableNames.push_back(name);283 m_tableNames.push_back(name);
289 return *rawPtr;284 return *rawPtr;
290 }285 }
291 286 
292- template <typename T>287+ template <typename T> std::optional<T> Find(TableKeyType<T> id) {
293- std::optional<T> Find(TableKeyType<T> id)
294- {
295 return GetTable<T>().Find(id);288 return GetTable<T>().Find(id);
296 }289 }
297 290 
298 template <typename T>291 template <typename T>
299- std::pair<TableValue<T>, bool> Query(std::function<bool(const T&)> pred)292+ std::pair<TableValue<T>, bool> Query(std::function<bool(const T &)> pred) {
300- {
301 return GetTable<T>().Query(pred);293 return GetTable<T>().Query(pred);
302 }294 }
303 295 
304 template <typename T>296 template <typename T>
305- std::vector<TableValue<T>> QueryList(std::function<bool(const T&)> pred)297+ std::vector<TableValue<T>> QueryList(std::function<bool(const T &)> pred) {
306- {
307 return GetTable<T>().QueryList(pred);298 return GetTable<T>().QueryList(pred);
308 }299 }
309 300 
310- template <typename T>301+ template <typename T> TableKeyType<T> Add(T rec) {
311- TableKeyType<T> Add(T rec)
312- {
313 return GetTable<T>().Add(rec);302 return GetTable<T>().Add(rec);
314 }303 }
315 304 
316 template <typename T>305 template <typename T>
317- bool Update(TableKeyType<T> id, std::function<void(T&)> updater)306+ std::vector<TableKeyType<T>> AddBatch(std::vector<T> &recs) {
318- {307+ return GetTable<T>().AddBatch(recs);
308+ }
309+ 
310+ template <typename T>
311+ bool Update(TableKeyType<T> id, std::function<void(T &)> updater) {
319 return GetTable<T>().Update(id, updater);312 return GetTable<T>().Update(id, updater);
320 }313 }
321 314 
322- template <typename T>315+ template <typename T> bool Delete(TableKeyType<T> id) {
323- bool Delete(TableKeyType<T> id)
324- {
325 return GetTable<T>().Delete(id);316 return GetTable<T>().Delete(id);
326 }317 }
327 318 
328- template <typename T>319+ template <typename T> bool DeleteAll() { return GetTable<T>().DeleteAll(); }
329- bool DeleteAll()
330- {
331- return GetTable<T>().DeleteAll();
332- }
333 320 
334 std::vector<std::string> GetAllTableName() const { return m_tableNames; }321 std::vector<std::string> GetAllTableName() const { return m_tableNames; }
335 322 
336 // 获取底层数据库连接对象(供批量写事务SqliteWriteTransaction使用)323 // 获取底层数据库连接对象(供批量写事务SqliteWriteTransaction使用)
337- sim::SqliteDatabase& GetDatabase() { return m_db; }324+ sim::SqliteDatabase &GetDatabase() { return m_db; }
338 325 
339 void ClearAll() { m_db.ClearAllTables(); }326 void ClearAll() { m_db.ClearAllTables(); }
340};327};
@@ -345,26 +332,25 @@ namespace sim {
345// 显式调用Commit()提交; 未提交即离开作用域(异常/提前返回)时自动回滚。332// 显式调用Commit()提交; 未提交即离开作用域(异常/提前返回)时自动回滚。
346// 注意: 事务期间应保证无其他线程并发写库。333// 注意: 事务期间应保证无其他线程并发写库。
347class SqliteWriteTransaction {334class SqliteWriteTransaction {
348-public:335+ public:
349- SqliteWriteTransaction() : db_(SimRunnerSqliteDB::Instance().GetDatabase()), began_(false), committed_(false)336+ SqliteWriteTransaction()
350- {337+ : db_(SimRunnerSqliteDB::Instance().GetDatabase()), began_(false),
338+ committed_(false) {
351 began_ = db_.BeginTransaction();339 began_ = db_.BeginTransaction();
352 }340 }
353 341 
354- ~SqliteWriteTransaction()342+ ~SqliteWriteTransaction() {
355- {
356 if (began_ && !committed_) {343 if (began_ && !committed_) {
357 db_.RollbackTransaction();344 db_.RollbackTransaction();
358 }345 }
359 }346 }
360 347 
361- SqliteWriteTransaction(const SqliteWriteTransaction&) = delete;348+ SqliteWriteTransaction(const SqliteWriteTransaction &) = delete;
362- SqliteWriteTransaction& operator=(const SqliteWriteTransaction&) = delete;349+ SqliteWriteTransaction &operator=(const SqliteWriteTransaction &) = delete;
363- SqliteWriteTransaction(SqliteWriteTransaction&&) = delete;350+ SqliteWriteTransaction(SqliteWriteTransaction &&) = delete;
364- SqliteWriteTransaction& operator=(SqliteWriteTransaction&&) = delete;351+ SqliteWriteTransaction &operator=(SqliteWriteTransaction &&) = delete;
365 352 
366- bool Commit()353+ bool Commit() {
367- {
368 if (!began_ || committed_) {354 if (!began_ || committed_) {
369 return began_ && committed_;355 return began_ && committed_;
370 }356 }
@@ -372,8 +358,8 @@ public:
372 return committed_;358 return committed_;
373 }359 }
374 360 
375-private:361+ private:
376- SqliteDatabase& db_;362+ SqliteDatabase &db_;
377 bool began_;363 bool began_;
378 bool committed_;364 bool committed_;
379};365};
Mtest/hccl_vm/include/runnerdb/sim_models.h+97-56
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_MODEL_DEFS_H18#ifndef SIM_MODEL_DEFS_H
@@ -68,9 +75,9 @@ enum CommStatus : uint8_t {
68};75};
69 76 
70typedef struct {77typedef struct {
71- uint64_t id; // PK（通信域锚点成员：同一 (comm_id, comm_hash) 下 id78+ uint64_t id; // PK（通信域锚点成员：同一 (comm_id, comm_hash) 下 id
72- // 最小者作为通信域标识）79+ // 最小者作为通信域标识）
73- char comm_id[128]; // 通信域名称，与 comm_hash 共同唯一标识通信域80+ char comm_id[128]; // 通信域名称，与 comm_hash 共同唯一标识通信域
74 uint32_t rank_size; // 通信域总 Rank 数81 uint32_t rank_size; // 通信域总 Rank 数
75 uint32_t rank_id; // 当前线程在通信域中的 Rank82 uint32_t rank_id; // 当前线程在通信域中的 Rank
76 uint64_t device_id; // FK -> Device.id83 uint64_t device_id; // FK -> Device.id
@@ -103,11 +110,16 @@ typedef struct {
103 uint32_t physical_id;110 uint32_t physical_id;
104 uint32_t super_device_id;111 uint32_t super_device_id;
105 uint32_t overflow_mode;112 uint32_t overflow_mode;
106- char soc_version[16];113+ // 完整SoC型号名(如"Ascend950DT_95A1"，16字符)需16+1字节，32对齐hal侧MAX_CHIP_NAME
114+ char soc_version[32];
107 uint32_t status; // 0 可用， 1不可用115 uint32_t status; // 0 可用， 1不可用
108} Device;116} Device;
109 117 
110-enum TsDevType { TS_DEV_TYPE_SCALAR = 0, TS_DEV_TYPE_CPU = 1, TS_DEV_TYPE_CCU = 2 };118+enum TsDevType {
119+ TS_DEV_TYPE_SCALAR = 0,
120+ TS_DEV_TYPE_CPU = 1,
121+ TS_DEV_TYPE_CCU = 2
122+};
111 123 
112typedef struct {124typedef struct {
113 uint64_t id; // PK125 uint64_t id; // PK
@@ -115,7 +127,11 @@ typedef struct {
115 uint8_t type; // 0:Scalar, 1:CCU, 2:CPU127 uint8_t type; // 0:Scalar, 1:CCU, 2:CPU
116} TaskSchedulerDevice;128} TaskSchedulerDevice;
117 129 
118-enum ComputeDieType { COMPUTE_TYPE_CUBE = 0, COMPUTE_TYPE_VECTOR = 1, COMPUTE_TYPE_HYBRID = 2 };130+enum ComputeDieType {
131+ COMPUTE_TYPE_CUBE = 0,
132+ COMPUTE_TYPE_VECTOR = 1,
133+ COMPUTE_TYPE_HYBRID = 2
134+};
119 135 
120typedef struct {136typedef struct {
121 uint64_t id; // PK137 uint64_t id; // PK
@@ -152,18 +168,6 @@ typedef struct {
152 uint8_t status;168 uint8_t status;
153} Ccu;169} Ccu;
154 170 
155-typedef struct {
156- uint64_t id; // PK
157- uint64_t ccu_id; // FK
158- uint8_t instr_space[CCU_INSTRUCTION_NUM][CCU_INSTRUCTION_SIZE]; // 指令空间1M: 32K个指令
159- uint16_t instr_cnt{0};
160- uint16_t state{0};
161- // uint64_t xn[CCU_RESOURCE_XN_MAX]; // xn寄存器
162- // uint64_t gsa[CCU_RESOURCE_GSA_MAX]; // gsa寄存器
163- // uint16_t cke[CCU_RESOURCE_CKE_MAX]; // cke寄存器
164- // char ms[CCU_RESOURCE_MS_SIZE]; // ms寄存器
165-} CcuResource; // todo: 后续可能改为关系表，数据保存在共享内存
166- 
167typedef struct {171typedef struct {
168 uint64_t id; // PK172 uint64_t id; // PK
169 uint64_t src_dev_id; // FK173 uint64_t src_dev_id; // FK
@@ -181,6 +185,7 @@ typedef struct {
181 uint8_t eid[16];185 uint8_t eid[16];
182 char ip_addr[64];186 char ip_addr[64];
183 uint8_t status{0};187 uint8_t status{0};
188+ bool is_uboe{false};
184} EndPoint;189} EndPoint;
185 190 
186typedef struct {191typedef struct {
@@ -312,8 +317,8 @@ typedef struct {
312 uint64_t id;317 uint64_t id;
313 uint64_t start_ptr; // 按照卡分配的虚拟编址的地址318 uint64_t start_ptr; // 按照卡分配的虚拟编址的地址
314 uint64_t dev_mapped_ptr; // device进程打开共享内存后的地址319 uint64_t dev_mapped_ptr; // device进程打开共享内存后的地址
315- uint8_t is_dev_access; // 0: dev不可直接访问, 1: dev可以直接访问320+ uint8_t is_dev_access; // 0: dev不可直接访问, 1: dev可以直接访问
316- uint64_t host_ptr{0}; // host进程打开共享内存后的地址321+ uint64_t host_ptr{0}; // host进程打开共享内存后的地址
317 uint64_t size;322 uint64_t size;
318 uint64_t ctx_id;323 uint64_t ctx_id;
319 uint64_t device_id; // Device表主键324 uint64_t device_id; // Device表主键
@@ -385,10 +390,12 @@ typedef struct {
385 390 
386// ReduceScatterV AllGatherV使用391// ReduceScatterV AllGatherV使用
387struct VDataDesTagInner {392struct VDataDesTagInner {
388- uint16_t dataType; // 数据类型393+ uint16_t dataType; // 数据类型
389- uint32_t count{0}; // rank size394+ uint32_t count{0}; // rank size
390 uint64_t displs[64]; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）395 uint64_t displs[64]; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）
391- uint64_t counts[64];396+ uint64_t
397+ counts[64]; // 每个rank在sendBuf中的数据size，第i个元素表示需要向rank
398+ // i发送/接受的数据量
392};399};
393 400 
394struct All2AllDataDesTagInner {401struct All2AllDataDesTagInner {
@@ -539,6 +546,8 @@ typedef struct {
539 uint8_t state; // 0: RESET, 1: INIT, 2: RTR, 3:RTS546 uint8_t state; // 0: RESET, 1: INIT, 2: RTR, 3:RTS
540 uint32_t mode; // jetty mode 0: URMA, 2: CCU, 3: Normal547 uint32_t mode; // jetty mode 0: URMA, 2: CCU, 3: Normal
541 uint64_t peer_jetty_handle;548 uint64_t peer_jetty_handle;
549+ uint64_t
550+ peer_endpoint_id; // 本jetty所在channel的对端EndPoint id(FK EndPoint.id)
542 uint64_t pid;551 uint64_t pid;
543} RaJetty;552} RaJetty;
544 553 
@@ -585,9 +594,9 @@ typedef struct {
585 594 
586// 北向打桩接口——HCCL_Channel表595// 北向打桩接口——HCCL_Channel表
587typedef struct {596typedef struct {
588- uint64_t id; // channel的ID，PK597+ uint64_t id; // channel的ID，PK
589- uint64_t commId; // 所属的通信域FK598+ uint64_t commId; // 所属的通信域FK
590- uint64_t endpoint_id; // 所属的北向 HcommEndpoint 逻辑外键，0 表示未绑定599+ uint64_t endpoint_id; // 所属的北向 HcommEndpoint 逻辑外键，0 表示未绑定
591 uint8_t engine; // 通信引起engine600 uint8_t engine; // 通信引起engine
592 uint64_t remoteRankId; // 对端rankId601 uint64_t remoteRankId; // 对端rankId
593 uint16_t notifyNum; // Notify的数量602 uint16_t notifyNum; // Notify的数量
@@ -597,44 +606,46 @@ typedef struct {
597 606 
598// 北向打桩接口——HComm Endpoint 与南向拓扑 Endpoint 的绑定状态607// 北向打桩接口——HComm Endpoint 与南向拓扑 Endpoint 的绑定状态
599enum HcommEndpointBindState : uint8_t {608enum HcommEndpointBindState : uint8_t {
600- HCOMM_ENDPOINT_UNRESOLVED = 0, // 尚未找到唯一的南向 Endpoint609+ HCOMM_ENDPOINT_UNRESOLVED = 0, // 尚未找到唯一的南向 Endpoint
601- HCOMM_ENDPOINT_BOUND = 1, // 已绑定唯一的南向 Endpoint610+ HCOMM_ENDPOINT_BOUND = 1, // 已绑定唯一的南向 Endpoint
602 HCOMM_ENDPOINT_NOT_APPLICABLE = 2, // 当前位置或协议不需要南向 Endpoint611 HCOMM_ENDPOINT_NOT_APPLICABLE = 2, // 当前位置或协议不需要南向 Endpoint
603};612};
604 613 
605// 北向打桩接口——HComm Endpoint 表614// 北向打桩接口——HComm Endpoint 表
606typedef struct {615typedef struct {
607- uint64_t id; // EndpointHandle 对应的主键616+ uint64_t id; // EndpointHandle 对应的主键
608- uint64_t runner_id; // 创建该 Endpoint 的 Runner 逻辑外键617+ uint64_t runner_id; // 创建该 Endpoint 的 Runner 逻辑外键
609- uint64_t south_endpoint_id; // 南向 sim::EndPoint 逻辑外键，0 表示未绑定或不适用618+ uint64_t
610- int32_t protocol; // CommProtocol619+ south_endpoint_id; // 南向 sim::EndPoint 逻辑外键，0 表示未绑定或不适用
611- int32_t addr_type; // CommAddrType620+ int32_t protocol; // CommProtocol
612- uint8_t addr[36]; // CommAddr 中的原始地址内容621+ int32_t addr_type; // CommAddrType
613- int32_t loc_type; // EndpointLocType622+ uint8_t addr[36]; // CommAddr 中的原始地址内容
614- uint8_t loc[60]; // EndpointLoc 中的原始位置内容623+ int32_t loc_type; // EndpointLocType
615- uint8_t extension[52]; // EndpointDesc 预留扩展内容624+ uint8_t loc[60]; // EndpointLoc 中的原始位置内容
616- uint8_t bind_state; // HcommEndpointBindState625+ uint8_t extension[52]; // EndpointDesc 预留扩展内容
626+ uint8_t bind_state; // HcommEndpointBindState
617} HcommEndpoint;627} HcommEndpoint;
618 628 
619// 北向打桩接口——HComm 内存记录类型629// 北向打桩接口——HComm 内存记录类型
620enum HcommMemRecordKind : uint8_t {630enum HcommMemRecordKind : uint8_t {
621 HCOMM_MEM_LOCAL_REGISTERED = 0, // HcommMemReg 创建的本地注册内存631 HCOMM_MEM_LOCAL_REGISTERED = 0, // HcommMemReg 创建的本地注册内存
622- HCOMM_MEM_REMOTE_IMPORTED = 1, // HcommMemImport 创建的远端导入内存632+ HCOMM_MEM_REMOTE_IMPORTED = 1, // HcommMemImport 创建的远端导入内存
623};633};
624 634 
625// 北向打桩接口——HComm 内存资源表635// 北向打桩接口——HComm 内存资源表
626typedef struct {636typedef struct {
627- uint64_t id; // HcommMemHandle 对应的主键637+ uint64_t id; // HcommMemHandle 对应的主键
628- uint64_t endpoint_id; // 所属的北向 HcommEndpoint 逻辑外键638+ uint64_t endpoint_id; // 所属的北向 HcommEndpoint 逻辑外键
629- uint64_t runner_id; // 创建该记录的 Runner 逻辑外键639+ uint64_t runner_id; // 创建该记录的 Runner 逻辑外键
630- uint64_t source_endpoint_id; // 导入记录的源端 HcommEndpoint 主键，本地记录为 0640+ uint64_t
631- uint64_t source_mem_id; // 导入记录的源端本地内存主键，本地记录为 0641+ source_endpoint_id; // 导入记录的源端 HcommEndpoint 主键，本地记录为 0
632- uint64_t descriptor_key; // 导出描述符的稳定校验键642+ uint64_t source_mem_id; // 导入记录的源端本地内存主键，本地记录为 0
633- uint64_t addr; // CommMem.addr 的整数形式643+ uint64_t descriptor_key; // 导出描述符的稳定校验键
634- uint64_t size; // CommMem.size644+ uint64_t addr; // CommMem.addr 的整数形式
635- int32_t mem_type; // CommMemType645+ uint64_t size; // CommMem.size
636- uint8_t record_kind; // HcommMemRecordKind646+ int32_t mem_type; // CommMemType
637- char mem_tag[255]; // 本地注册内存标签，导入记录为空647+ uint8_t record_kind; // HcommMemRecordKind
648+ char mem_tag[255]; // 本地注册内存标签，导入记录为空
638} HcommMemReg;649} HcommMemReg;
639 650 
640// 北向打桩接口——HCCL_EngineCtx表651// 北向打桩接口——HCCL_EngineCtx表
@@ -673,5 +684,35 @@ typedef struct {
673 uint8_t consumed{0}; // 0: 未被消费, 1: 已被消费684 uint8_t consumed{0}; // 0: 未被消费, 1: 已被消费
674} DpuPendingNotify;685} DpuPendingNotify;
675 686 
687+// 北向打桩接口——CCU Level1 交换资源表（跨 rank 同步值，per-channel 交换区，**按
688+// op+轮次隔离**） 语义：hcomm CcuTransport::INIT_XN_NUM/INIT_CKE_NUM(4) —— 每
689+// channel 固定预留
690+// XN/CKE 0..3；PreSync 时写方把值写到对端坐标，读方（NotifyWait /
691+// 首次使用处）查询。
692+// 逻辑键：(opIter, round, srcRank, targetRank, connOrdinal, dieId, resType,
693+// resId)；
694+// opIter = 算子迭代（OpDetailTab.opIter，两侧进程按同一序列自增，天然跨
695+// rank 对齐）， round = 写方 CCU launch 序号。**一轮一行、不覆盖** ——
696+// 读方按自己所在 op+轮次精确取，
697+// 可免疫两进程异步漂移（写方领先时不会被"未来轮次/未来 op 的值"污染）。
698+// op 维度不可省：round 是进程内全局 launch 计数，表跨 op 累积，若不带 op
699+// 则后一 op 的 行会覆盖/裁掉前一 op 尚未被读走的行，落后 rank
700+// 只能取到别的 op 的地址（现象： checker 报 MEM_CPY 目标地址落不到本 op
701+// 内存布局）。
702+typedef struct {
703+ uint64_t id{0}; // PK
704+ uint32_t opIter{0}; // 算子迭代（跨 rank 一致的 op 身份）
705+ uint32_t round{0}; // 轮次（= 写方 CCU launch 序号）
706+ uint32_t targetRank{0}; // 被写入方 rank（= 读方 rank）
707+ uint32_t connOrdinal{
708+ 0}; // 同 peer 多 channel 连接序号（multi-jetty；单连接=0）
709+ uint32_t dieId{0}; // 资源所在 die
710+ uint32_t resType{0}; // 0=XN(Var), 1=CKE
711+ uint32_t resId{0}; // 0..3（交换区槽位）
712+ uint64_t value{0}; // 交换值（地址/token/CKE 位置位）
713+ uint32_t srcRank{0}; // 写方 rank（排障）
714+ uint64_t ownerPid{0}; // 写方进程（排障）
715+} CcuSyncResTab;
716+ 
676} // namespace sim717} // namespace sim
677#endif718#endif
Mtest/hccl_vm/include/runnerdb/sim_op_db_types.h+23-17
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_OP_DB_TYPES_H18#ifndef SIM_OP_DB_TYPES_H
@@ -29,21 +36,21 @@ struct DBConfig {
29};36};
30 37 
31struct OpDetailTab {38struct OpDetailTab {
32- uint32_t id; // PK (自增)39+ uint32_t id; // PK (自增)
33- uint32_t pid; // proxy 进程 ID40+ uint32_t pid; // proxy 进程 ID
34 uint32_t deviceId; // 物理设备 ID（全局唯一，内部逻辑统一使用）41 uint32_t deviceId; // 物理设备 ID（全局唯一，内部逻辑统一使用）
35 uint32_t rankId; // 通信域内 rank ID42 uint32_t rankId; // 通信域内 rank ID
36 uint64_t commId; // 通信域成员 ID，FK -> Communicator.id43 uint64_t commId; // 通信域成员 ID，FK -> Communicator.id
37 uint32_t opIter; // 算子迭代次数 (从 0 开始)44 uint32_t opIter; // 算子迭代次数 (从 0 开始)
38 uint32_t syncIter; // 所属 sync 周期 (从 0 开始) // todo 可删45 uint32_t syncIter; // 所属 sync 周期 (从 0 开始) // todo 可删
39 46 
40- uint64_t streamId; // streamId47+ uint64_t streamId; // streamId
41- uint32_t root; // root48+ uint32_t root; // root
42- uint32_t opExpansionMode; // 算子执行模式 0: ccu, 1: aicpu, 2: aiv49+ uint32_t opExpansionMode; // 算子执行模式 0: ccu, 1: aicpu, 2: aiv
43- uint32_t devType; // 设备类型50+ uint32_t devType; // 设备类型
44- uint32_t rankSize; // ransize51+ uint32_t rankSize; // ransize
45- uint32_t srcRank; // srcRank52+ uint32_t srcRank; // srcRank
46- uint32_t dstRank; // dstRank53+ uint32_t dstRank; // dstRank
47 std::vector<uint8_t> opDetail; // BLOB54 std::vector<uint8_t> opDetail; // BLOB
48 std::vector<uint8_t> opExtInfo; // BLOB55 std::vector<uint8_t> opExtInfo; // BLOB
49};56};
@@ -138,8 +145,7 @@ struct OpExecutionKey {
138 uint64_t commHash{0};145 uint64_t commHash{0};
139 uint32_t opIter{0};146 uint32_t opIter{0};
140 147 
141- bool operator<(const OpExecutionKey& other) const148+ bool operator<(const OpExecutionKey &other) const {
142- {
143 if (commName != other.commName) {149 if (commName != other.commName) {
144 return commName < other.commName;150 return commName < other.commName;
145 }151 }
Mtest/hccl_vm/include/runnerdb/sim_runner_impl.inl+20-25
@@ -1,11 +1,13 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
9 */11 */
10 12 
11#ifndef SIM_RUNNER_IMPL_INL13#ifndef SIM_RUNNER_IMPL_INL
@@ -17,47 +19,40 @@
17 19 
18namespace RunnerDB {20namespace RunnerDB {
19 21 
20-template <typename T>22+template <typename T> uint64_t Add(T &rec) {
21-uint64_t Add(T& rec)
22-{
23 return SimRunnerSqliteDB::Instance().Add<T>(rec);23 return SimRunnerSqliteDB::Instance().Add<T>(rec);
24}24}
25 25 
26-template <typename T>26+template <typename T> std::vector<uint64_t> AddBatch(std::vector<T> &recs) {
27-std::optional<T> GetById(uint64_t id)27+ return SimRunnerSqliteDB::Instance().AddBatch<T>(recs);
28-{28+}
29+ 
30+template <typename T> std::optional<T> GetById(uint64_t id) {
29 return SimRunnerSqliteDB::Instance().Find<T>(id);31 return SimRunnerSqliteDB::Instance().Find<T>(id);
30}32}
31 33 
32template <typename T>34template <typename T>
33-std::vector<T> GetByPred(std::function<bool(const T&)> pred)35+std::vector<T> GetByPred(std::function<bool(const T &)> pred) {
34-{
35 return SimRunnerSqliteDB::Instance().QueryList<T>(pred);36 return SimRunnerSqliteDB::Instance().QueryList<T>(pred);
36}37}
37 38 
38template <typename T>39template <typename T>
39-std::pair<T, bool> GetOneByPred(std::function<bool(const T&)> pred)40+std::pair<T, bool> GetOneByPred(std::function<bool(const T &)> pred) {
40-{
41 return SimRunnerSqliteDB::Instance().Query<T>(pred);41 return SimRunnerSqliteDB::Instance().Query<T>(pred);
42}42}
43 43 
44template <typename T>44template <typename T>
45-bool Update(uint64_t id, std::function<void(T&)> updater)45+bool Update(uint64_t id, std::function<void(T &)> updater) {
46-{
47 return SimRunnerSqliteDB::Instance().Update<T>(id, updater);46 return SimRunnerSqliteDB::Instance().Update<T>(id, updater);
48}47}
49 48 
50-template <typename T>49+template <typename T> bool Delete(uint64_t id) {
51-bool Delete(uint64_t id)
52-{
53 return SimRunnerSqliteDB::Instance().Delete<T>(id);50 return SimRunnerSqliteDB::Instance().Delete<T>(id);
54}51}
55 52 
56-template <typename T>53+template <typename T> bool DeleteAll() {
57-bool DeleteAll()
58-{
59 return SimRunnerSqliteDB::Instance().DeleteAll<T>();54 return SimRunnerSqliteDB::Instance().DeleteAll<T>();
60}55}
61} // namespace RunnerDB56} // namespace RunnerDB
62 57 
63-#endif58+#endif
Mtest/hccl_vm/include/sim_aicpu_pipe_msg.h+15-7
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_AICPU_PIPE_MSG_H18#ifndef SIM_AICPU_PIPE_MSG_H
@@ -38,7 +45,8 @@ constexpr uint8_t PIPE_RSP_FREE_WQE_PTR = 0x86;
38constexpr uint8_t PIPE_RSP_READY = 0xA5;45constexpr uint8_t PIPE_RSP_READY = 0xA5;
39 46 
40constexpr uint8_t PIPE_RSP_ERROR = 0xFF;47constexpr uint8_t PIPE_RSP_ERROR = 0xFF;
41-constexpr uint32_t IPC_MSG_LEN_MAX = 128 * 1024; // 变长消息 payload 上限（容纳 64KB args）48+constexpr uint32_t IPC_MSG_LEN_MAX =
49+ 128 * 1024; // 变长消息 payload 上限（容纳 64KB args）
42 50 
43#pragma pack(push, 1)51#pragma pack(push, 1)
44 52 
Mtest/hccl_vm/include/sim_binary_data_type_pub.h+36-26
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,  *
15+the full text of the License.
9 */16 */
10 17 
11#ifndef BINRAY_DATA_TYPE_PUB_H18#ifndef BINRAY_DATA_TYPE_PUB_H
@@ -16,7 +23,6 @@
16 23 
17#include "ccu_microcode_v1.h"24#include "ccu_microcode_v1.h"
18#include "hccl_types.h"25#include "hccl_types.h"
19-#include "sim_capacity_limits.h"
20#include "sim_common_defs.h"26#include "sim_common_defs.h"
21 27 
22namespace HcclSim {28namespace HcclSim {
@@ -41,9 +47,9 @@ struct FileHeader {
41struct VDataDesTag {47struct VDataDesTag {
42 uint16_t dataType; // 数据类型48 uint16_t dataType; // 数据类型
43 uint32_t count; // rank size49 uint32_t count; // rank size
44- uint64_t* displs; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）50+ uint64_t *displs; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）
45- uint64_t* counts; // 每个rank在sendBuf中的数据size，第i个元素表示需要向rank51+ uint64_t *counts; // 每个rank在sendBuf中的数据size，第i个元素表示需要向rank
46- // i发送/接受的数据量52+ // i发送/接受的数据量
47};53};
48 54 
49struct All2AllDataDesTag {55struct All2AllDataDesTag {
@@ -52,7 +58,7 @@ struct All2AllDataDesTag {
52 uint64_t sendCount; // 发送数据量 (All2All)58 uint64_t sendCount; // 发送数据量 (All2All)
53 uint64_t recvCount; // 接收数据量 (All2All)59 uint64_t recvCount; // 接收数据量 (All2All)
54 uint32_t count; // count = rankSize * rankSize60 uint32_t count; // count = rankSize * rankSize
55- uint64_t* sendCountMatrix; // (All2AllVC) sendCountMatrix[i * ranksize + j]61+ uint64_t *sendCountMatrix; // (All2AllVC) sendCountMatrix[i * ranksize + j]
56 // 代表rank i发送到rank j的count参数62 // 代表rank i发送到rank j的count参数
57};63};
58 64 
@@ -92,12 +98,12 @@ struct ChannelData {
92 uint8_t reid[16];98 uint8_t reid[16];
93 uint16_t protocol;99 uint16_t protocol;
94 uint16_t jettyNum;100 uint16_t jettyNum;
95- uint32_t jettyId[HCCL_VM_MAX_CHANNEL_JETTIES];101+ uint32_t jettyId[32];
96};102};
97 103 
98struct ChannelInfo {104struct ChannelInfo {
99 uint32_t count;105 uint32_t count;
100- ChannelData* data;106+ ChannelData *data;
101};107};
102 108 
103struct MemLayoutData {109struct MemLayoutData {
@@ -111,7 +117,7 @@ struct MemLayoutData {
111 117 
112struct MemLayoutInfo {118struct MemLayoutInfo {
113 uint32_t count;119 uint32_t count;
114- MemLayoutData* data;120+ MemLayoutData *data;
115};121};
116 122 
117struct MicrocodeInstrDesc {123struct MicrocodeInstrDesc {
@@ -123,7 +129,7 @@ struct MicrocodeInstrDesc {
123 129 
124struct MicrocodeInstr {130struct MicrocodeInstr {
125 MicrocodeInstrDesc desc;131 MicrocodeInstrDesc desc;
126- hcomm::CcuRep::CcuInstr* data;132+ hcomm::CcuRep::CcuInstr *data;
127};133};
128 134 
129struct ModelInfoCommInner {135struct ModelInfoCommInner {
@@ -160,20 +166,24 @@ struct MemLayoutInfoInner {
160 166 
161// ReduceScatterV AllGatherV使用167// ReduceScatterV AllGatherV使用
162struct VDataDesTagInner {168struct VDataDesTagInner {
163- uint16_t dataType; // 数据类型169+ uint16_t dataType; // 数据类型
164- uint32_t count{0}; // rank size170+ uint32_t count{0}; // rank size
165- std::vector<uint64_t> displs; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）171+ std::vector<uint64_t>
166- std::vector<uint64_t> counts;172+ displs; // 每个rank的数据在sendBuf中的偏移量（单位为dataType）
173+ std::vector<uint64_t>
174+ counts; // 每个rank在sendBuf中的数据size，第i个元素表示需要向rank
175+ // i发送/接受的数据量
167};176};
168 177 
169struct All2AllDataDesTagInner {178struct All2AllDataDesTagInner {
170- uint16_t sendType; // 发送数据的数据类型179+ uint16_t sendType; // 发送数据的数据类型
171- uint16_t recvType; // 接收数据的数据类型180+ uint16_t recvType; // 接收数据的数据类型
172- uint64_t sendCount; // 发送数据量 (All2All)181+ uint64_t sendCount; // 发送数据量 (All2All)
173- uint64_t recvCount; // 接收数据量 (All2All)182+ uint64_t recvCount; // 接收数据量 (All2All)
174- uint32_t count{0}; // count = rankSize * rankSize183+ uint32_t count{0}; // count = rankSize * rankSize
175- std::vector<uint64_t> sendCountMatrix; // (All2AllVC) sendCountMatrix[i * ranksize + j]184+ std::vector<uint64_t>
176- // 代表rank i发送到rank j的count参数185+ sendCountMatrix; // (All2AllVC) sendCountMatrix[i * ranksize + j]
186+ // 代表rank i发送到rank j的count参数
177};187};
178 188 
179struct ModelInfoInner {189struct ModelInfoInner {
Mtest/hccl_vm/include/sim_ccu_channel_ctx.h+17-22
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,  *
15+the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_CCU_CHANNEL_CTX_H18#ifndef SIM_CCU_CHANNEL_CTX_H
@@ -94,22 +101,10 @@ struct ChannelCtxDataV1 {
94 /********64 Bytes**********/101 /********64 Bytes**********/
95 102 
96 ChannelCtxDataV1()103 ChannelCtxDataV1()
97- : vtpHigh(0),104+ : vtpHigh(0), srcPfeId{0}, startJettyIdLow{0}, startJettyIdHigh{0},
98- srcPfeId{0},105+ jettyNumLow{0}, jettyNumHigh{0}, ioDieId{0}, dstTokenIdLow(0),
99- startJettyIdLow{0},106+ dstTokenIdHigh{0}, dstTokenValueLow{0}, dstTokenValueHigh{0},
100- startJettyIdHigh{0},107+ dstVaLow{0}, dstVaHigher{0}, dstTokenValueValid{0}, rsv14Bits{0} {}
101- jettyNumLow{0},
102- jettyNumHigh{0},
103- ioDieId{0},
104- dstTokenIdLow(0),
105- dstTokenIdHigh{0},
106- dstTokenValueLow{0},
107- dstTokenValueHigh{0},
108- dstVaLow{0},
109- dstVaHigher{0},
110- dstTokenValueValid{0},
111- rsv14Bits{0}
112- {}
113};108};
114 109 
115struct ChannelDataV2 {110struct ChannelDataV2 {
Mtest/hccl_vm/include/sim_ccu_jetty_ctx.h+21-24
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,  *
15+the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_CCU_JETTY_CTX_H18#ifndef SIM_CCU_JETTY_CTX_H
@@ -36,8 +43,9 @@ struct LocalJettyCtxData {
36 uint16_t doorbellAddr[4] = {0}; // jetty doorbell addr43 uint16_t doorbellAddr[4] = {0}; // jetty doorbell addr
37 /********8 Bytes**********/44 /********8 Bytes**********/
38 45 
39- uint16_t pfeIdx : 4; // jetty relegation use PFE num.46+ uint16_t pfeIdx : 4; // jetty relegation use PFE num.
40- uint16_t ioDieId : 1; // 0: locall jetty use IODIE0, 1: locall jetty use IODIE1.47+ uint16_t
48+ ioDieId : 1; // 0: locall jetty use IODIE0, 1: locall jetty use IODIE1.
41 uint16_t doorbellAddrType : 1; // doorbell addr type: 0:PA, 1:VA.49 uint16_t doorbellAddrType : 1; // doorbell addr type: 0:PA, 1:VA.
42 uint16_t tokenValueIsValid : 1; // doorbell addr releate token value valid50 uint16_t tokenValueIsValid : 1; // doorbell addr releate token value valid
43 // type: invalid(0), vailid(1).51 // type: invalid(0), vailid(1).
@@ -76,22 +84,11 @@ struct LocalJettyCtxData {
76 /********32 Bytes**********/84 /********32 Bytes**********/
77 85 
78 LocalJettyCtxData()86 LocalJettyCtxData()
79- : pfeIdx(0),87+ : pfeIdx(0), ioDieId{0}, doorbellAddrType{0}, tokenValueIsValid{0},
80- ioDieId{0},88+ cqeErrValue{0}, tokenIdLow{0}, tokenIdHigh{0}, tokenValueLow(0),
81- doorbellAddrType{0},89+ tokenValueHigh{0}, sqeBasicBlockLeftShifts{0}, oooCqeCnt{0},
82- tokenValueIsValid{0},90+ startWqeBasicBlockIdxLow{0}, startWqeBasicBlockIdxHigh{0},
83- cqeErrValue{0},91+ doorbellSendState{0}, rsvSixBits{0} {}
84- tokenIdLow{0},
85- tokenIdHigh{0},
86- tokenValueLow(0),
87- tokenValueHigh{0},
88- sqeBasicBlockLeftShifts{0},
89- oooCqeCnt{0},
90- startWqeBasicBlockIdxLow{0},
91- startWqeBasicBlockIdxHigh{0},
92- doorbellSendState{0},
93- rsvSixBits{0}
94- {}
95};92};
96#pragma pack(pop)93#pragma pack(pop)
97} // namespace HcclSim94} // namespace HcclSim
Mtest/hccl_vm/include/sim_common_api.h+17-10
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_COMMON_API_H18#ifndef SIM_COMMON_API_H
@@ -19,16 +26,16 @@
19#include "sim_common_defs.h"26#include "sim_common_defs.h"
20 27 
21class InstallPath {28class InstallPath {
22-public:29+ public:
23 // 获取 hccl-vm-install 根目录的绝对路径30 // 获取 hccl-vm-install 根目录的绝对路径
24- static const std::string& GetHcclVmInstallAbsPath();31+ static const std::string &GetHcclVmInstallAbsPath();
25 32 
26 // 把相对路径解析为基于 install root 的绝对路径33 // 把相对路径解析为基于 install root 的绝对路径
27 // @param relPath 相对路径(如 "data/hccl_vm_data.db")34 // @param relPath 相对路径(如 "data/hccl_vm_data.db")
28 // @note - 已是绝对路径(以 '/' 开头)则原样返回35 // @note - 已是绝对路径(以 '/' 开头)则原样返回
29 // - 以 "./" 开头也原样返回36 // - 以 "./" 开头也原样返回
30 // - 空字符串原样返回37 // - 空字符串原样返回
31- static std::string ResolveToInstallRoot(const std::string& relPath);38+ static std::string ResolveToInstallRoot(const std::string &relPath);
32};39};
33 40 
34std::string GetDataTypeStr(HcclDataType type);41std::string GetDataTypeStr(HcclDataType type);
@@ -36,7 +43,7 @@ std::string GetDataTypeStr(HcclDataType type);
36std::string GetReduceOpStr(HcclReduceOp op);43std::string GetReduceOpStr(HcclReduceOp op);
37 44 
38#if !defined(NO_YAML_CONFIG) && defined(HAVE_YAML_CPP)45#if !defined(NO_YAML_CONFIG) && defined(HAVE_YAML_CPP)
39-bool ParseTopoMetaYaml(const std::string& fileName, TopoMeta& topo);46+bool ParseTopoMetaYaml(const std::string &fileName, TopoMeta &topo);
40#endif47#endif
41 48 
42#endif // SIM_COMMON_API_H49#endif // SIM_COMMON_API_H
Mtest/hccl_vm/include/sim_common_defs.h+120-76
@@ -1,15 +1,22 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
9 */11 */
10 12 
11-#ifndef HCOMM_HCCL_VM_SIM_COMMON_DEFS_H13+/**
12-#define HCOMM_HCCL_VM_SIM_COMMON_DEFS_H14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
16+ */
17+ 
18+#ifndef HCCL_COMMON_DEFS_H
19+#define HCCL_COMMON_DEFS_H
13 20 
14#include <array>21#include <array>
15#include <cstdint>22#include <cstdint>
@@ -50,6 +57,8 @@ using SimEid = std::array<uint8_t, URMA_EID_LEN>;
50constexpr uint8_t DEVICE_WAIT = 0;57constexpr uint8_t DEVICE_WAIT = 0;
51constexpr uint8_t DEVICE_RUN = 1;58constexpr uint8_t DEVICE_RUN = 1;
52 59 
60+constexpr uint32_t AICPU_JETTY_NUM_MAX = 40960; // 暂定扩大至40960个
61+ 
53typedef struct {62typedef struct {
54 uint8_t dieId;63 uint8_t dieId;
55 uint8_t missionId;64 uint8_t missionId;
@@ -98,9 +107,9 @@ enum HcclVmResult {
98 HCCL_SIM_E_REMOTE = 21, /**< error cqe */107 HCCL_SIM_E_REMOTE = 21, /**< error cqe */
99 HCCL_SIM_E_SUSPENDING = 22, /**< error communicator suspending */108 HCCL_SIM_E_SUSPENDING = 22, /**< error communicator suspending */
100 HCCL_SIM_E_OPRETRY_FAIL = 23, /**< retry constraint */109 HCCL_SIM_E_OPRETRY_FAIL = 23, /**< retry constraint */
101- HCCL_SIM_E_IN_STATUS = 1041, /**< The error information is in the status. */110+ HCCL_SIM_E_IN_STATUS = 1041, /**< The error information is in the status. */
102- HCCL_SIM_E_RESERVED, /**< reserved */111+ HCCL_SIM_E_RESERVED, /**< reserved */
103- HCCL_SIM_E_SKIP, /**< skip */112+ HCCL_SIM_E_SKIP, /**< skip */
104 113 
105 // 4096 ~ 8191: HCCL_SIM错误码114 // 4096 ~ 8191: HCCL_SIM错误码
106 // 4096 ~ 5119: HCCL_SIM Host错误码115 // 4096 ~ 5119: HCCL_SIM Host错误码
@@ -160,14 +169,13 @@ using ServerMeta = std::vector<PhyDeviceId>;
160using SuperPodMeta = std::map<uint32_t, ServerMeta>;169using SuperPodMeta = std::map<uint32_t, ServerMeta>;
161using TopoMeta = std::map<uint32_t, SuperPodMeta>;170using TopoMeta = std::map<uint32_t, SuperPodMeta>;
162 171 
163-static inline uint32_t ShmGetPhyDeviceTotalCount(const TopoMeta& topo_meta)172+static inline uint32_t ShmGetPhyDeviceTotalCount(const TopoMeta &topo_meta) {
164-{
165 uint32_t total_count = 0;173 uint32_t total_count = 0;
166 174 
167 // 第一层：遍历所有 SuperPod（超级节点）175 // 第一层：遍历所有 SuperPod（超级节点）
168- for (const auto& [pod_id, super_pod] : topo_meta) {176+ for (const auto &[pod_id, super_pod] : topo_meta) {
169 // 第二层：遍历每个 SuperPod 中的所有 Server（服务器）177 // 第二层：遍历每个 SuperPod 中的所有 Server（服务器）
170- for (const auto& [server_id, server] : super_pod) {178+ for (const auto &[server_id, server] : super_pod) {
171 // 第三层：累加每个 Server 中的 PhyDeviceId 数量179 // 第三层：累加每个 Server 中的 PhyDeviceId 数量
172 total_count += static_cast<uint32_t>(server.size());180 total_count += static_cast<uint32_t>(server.size());
173 }181 }
@@ -183,6 +191,7 @@ enum class HccLTaskMetaType : char {
183 CCU_GRAPH,191 CCU_GRAPH,
184 AIV_GRAPH,192 AIV_GRAPH,
185 SYNC_STREAM,193 SYNC_STREAM,
194+ MODEL_EXEC,
186 INVALID195 INVALID
187};196};
188 197 
@@ -235,11 +244,11 @@ typedef struct {
235 uint64_t len;244 uint64_t len;
236 uint8_t protocol;245 uint8_t protocol;
237 246 
238- std::string Describe() const247+ std::string Describe() const {
239- {
240 std::ostringstream oss;248 std::ostringstream oss;
241- oss << "srcDeviceId=" << srcDeviceId << " srcOffset=" << srcOffset << " dstDeviceId=" << dstDeviceId249+ oss << "srcDeviceId=" << srcDeviceId << " srcOffset=" << srcOffset
242- << " dstOffset=" << dstOffset << " len=" << len;250+ << " dstDeviceId=" << dstDeviceId << " dstOffset=" << dstOffset
251+ << " len=" << len;
243 return oss.str();252 return oss.str();
244 }253 }
245} TransMemTask;254} TransMemTask;
@@ -247,14 +256,32 @@ typedef struct {
247typedef struct {256typedef struct {
248 uint64_t syncIdx;257 uint64_t syncIdx;
249 258 
250- std::string Describe() const259+ std::string Describe() const {
251- {
252 std::ostringstream oss;260 std::ostringstream oss;
253 oss << "syncIdx=" << syncIdx;261 oss << "syncIdx=" << syncIdx;
254 return oss.str();262 return oss.str();
255 }263 }
256} SyncStreamTask;264} SyncStreamTask;
257 265 
266+// 模型重放触发任务角色：同一次 rtModelExecute 的 4 个任务共享同一个
267+// execSeq，checker 据此配对建边
268+typedef struct {
269+ uint32_t
270+ execSeq; // 同一次 rtModelExecute 的 4 个任务共享，用于 checker 配对
271+ uint8_t role; // 0=MODEL_EXEC_START, 1=MODEL_EXEC_START_SUB,
272+ // 2=MODEL_EXEC_END_SUB, 3=MODEL_EXEC_END
273+ uint64_t modelId; // 所属 model 句柄（观察用）
274+ uint64_t
275+ peerStreamId; // 对端流：START 填 sub 流，START_SUB 填 main 流，反向同理
276+ 
277+ std::string Describe() const {
278+ std::ostringstream oss;
279+ oss << "execSeq=" << execSeq << " role=" << static_cast<uint32_t>(role)
280+ << " modelId=" << modelId << " peerStreamId=" << peerStreamId;
281+ return oss.str();
282+ }
283+} ModelExecTask;
284+ 
258typedef struct {285typedef struct {
259 uint32_t srcDeviceId;286 uint32_t srcDeviceId;
260 uint64_t srcOffset;287 uint64_t srcOffset;
@@ -266,12 +293,13 @@ typedef struct {
266 uint8_t protocol;293 uint8_t protocol;
267 uint8_t resv;294 uint8_t resv;
268 295 
269- std::string Describe() const296+ std::string Describe() const {
270- {
271 std::ostringstream oss;297 std::ostringstream oss;
272- oss << "srcDeviceId=" << srcDeviceId << " srcOffset=" << srcOffset << " dstDeviceId=" << dstDeviceId298+ oss << "srcDeviceId=" << srcDeviceId << " srcOffset=" << srcOffset
273- << " dstOffset=" << dstOffset << " dataCount=" << dataCount299+ << " dstDeviceId=" << dstDeviceId << " dstOffset=" << dstOffset
274- << " dataType=" << static_cast<uint32_t>(dataType) << " reduceOp=" << static_cast<uint32_t>(reduceOp);300+ << " dataCount=" << dataCount
301+ << " dataType=" << static_cast<uint32_t>(dataType)
302+ << " reduceOp=" << static_cast<uint32_t>(reduceOp);
275 return oss.str();303 return oss.str();
276 }304 }
277} ReduceTask;305} ReduceTask;
@@ -283,10 +311,10 @@ typedef struct {
283 uint8_t notifyCount;311 uint8_t notifyCount;
284 uint8_t protocol;312 uint8_t protocol;
285 313 
286- std::string Describe() const314+ std::string Describe() const {
287- {
288 std::ostringstream oss;315 std::ostringstream oss;
289- oss << "srcDeviceId=" << srcDeviceId << " dstDeviceId=" << dstDeviceId << " notifyId=" << notifyId;316+ oss << "srcDeviceId=" << srcDeviceId << " dstDeviceId=" << dstDeviceId
317+ << " notifyId=" << notifyId;
290 return oss.str();318 return oss.str();
291 }319 }
292} NotifyTask;320} NotifyTask;
@@ -301,11 +329,12 @@ typedef struct {
301 uint32_t argSize;329 uint32_t argSize;
302 uint64_t args[RT_CCU_SQE_ARGS_LEN];330 uint64_t args[RT_CCU_SQE_ARGS_LEN];
303 331 
304- std::string Describe() const332+ std::string Describe() const {
305- {
306 std::ostringstream oss;333 std::ostringstream oss;
307- oss << "dieId=" << static_cast<uint32_t>(dieId) << " missionId=" << static_cast<uint32_t>(missionId)334+ oss << "dieId=" << static_cast<uint32_t>(dieId)
308- << " timeout=" << timeout << " instStart=" << instStartId << " instCnt=" << instCnt << " key=" << key335+ << " missionId=" << static_cast<uint32_t>(missionId)
336+ << " timeout=" << timeout << " instStart=" << instStartId
337+ << " instCnt=" << instCnt << " key=" << key
309 << " argSize=" << argSize;338 << " argSize=" << argSize;
310 return oss.str();339 return oss.str();
311 }340 }
@@ -314,8 +343,7 @@ typedef struct {
314typedef struct {343typedef struct {
315 uint64_t launchIdx;344 uint64_t launchIdx;
316 345 
317- std::string Describe() const346+ std::string Describe() const {
318- {
319 std::ostringstream oss;347 std::ostringstream oss;
320 oss << "launchIdx=" << launchIdx;348 oss << "launchIdx=" << launchIdx;
321 return oss.str();349 return oss.str();
@@ -337,10 +365,10 @@ typedef struct HcclTaskMetaData {
337 CcuTask ccu;365 CcuTask ccu;
338 AivTask aiv;366 AivTask aiv;
339 SyncStreamTask syncStreamTask;367 SyncStreamTask syncStreamTask;
368+ ModelExecTask modelExec;
340 } taskData;369 } taskData;
341 370 
342- HcclTaskMetaData()371+ HcclTaskMetaData() {
343- {
344 taskType = HccLTaskMetaType::INVALID;372 taskType = HccLTaskMetaType::INVALID;
345 commId = 0;373 commId = 0;
346 rankId = UINT32_MAX;374 rankId = UINT32_MAX;
@@ -351,45 +379,51 @@ typedef struct HcclTaskMetaData {
351 memset(&taskData, 0, sizeof(taskData));379 memset(&taskData, 0, sizeof(taskData));
352 }380 }
353 381 
354- std::string TaskTypeName() const382+ std::string TaskTypeName() const {
355- {
356 static const std::map<HccLTaskMetaType, std::string> TASK_NAMES{383 static const std::map<HccLTaskMetaType, std::string> TASK_NAMES{
357- {HccLTaskMetaType::NOTIFY_WAIT, "NOTIFY_WAIT"}, {HccLTaskMetaType::NOTIFY_RECORD, "NOTIFY_RECORD"},384+ {HccLTaskMetaType::NOTIFY_WAIT, "NOTIFY_WAIT"},
358- {HccLTaskMetaType::REDUCE, "REDUCE"}, {HccLTaskMetaType::MEM_CPY, "MEM_CPY"},385+ {HccLTaskMetaType::NOTIFY_RECORD, "NOTIFY_RECORD"},
359- {HccLTaskMetaType::CCU_GRAPH, "CCU_GRAPH"}, {HccLTaskMetaType::AIV_GRAPH, "AIV_GRAPH"},386+ {HccLTaskMetaType::REDUCE, "REDUCE"},
387+ {HccLTaskMetaType::MEM_CPY, "MEM_CPY"},
388+ {HccLTaskMetaType::CCU_GRAPH, "CCU_GRAPH"},
389+ {HccLTaskMetaType::AIV_GRAPH, "AIV_GRAPH"},
360 {HccLTaskMetaType::SYNC_STREAM, "SYNC_STREAM"},390 {HccLTaskMetaType::SYNC_STREAM, "SYNC_STREAM"},
391+ {HccLTaskMetaType::MODEL_EXEC, "MODEL_EXEC"},
361 };392 };
362 auto it = TASK_NAMES.find(taskType);393 auto it = TASK_NAMES.find(taskType);
363 return it != TASK_NAMES.end() ? it->second : "INVALID";394 return it != TASK_NAMES.end() ? it->second : "INVALID";
364 }395 }
365 396 
366- std::string Describe() const397+ std::string Describe() const {
367- {
368 std::ostringstream oss;398 std::ostringstream oss;
369- oss << "taskType=" << TaskTypeName() << " commId=" << commId << " rankId=" << rankId << " deviceId=" << deviceId399+ oss << "taskType=" << TaskTypeName() << " commId=" << commId
400+ << " rankId=" << rankId << " deviceId=" << deviceId
370 << " streamId=" << streamId << " jettyId=" << jettyId;401 << " streamId=" << streamId << " jettyId=" << jettyId;
371 switch (taskType) {402 switch (taskType) {
372- case HccLTaskMetaType::NOTIFY_WAIT:403+ case HccLTaskMetaType::NOTIFY_WAIT:
373- case HccLTaskMetaType::NOTIFY_RECORD:404+ case HccLTaskMetaType::NOTIFY_RECORD:
374- oss << " taskData={" << taskData.notify.Describe() << "}";405+ oss << " taskData={" << taskData.notify.Describe() << "}";
375- break;406+ break;
376- case HccLTaskMetaType::REDUCE:407+ case HccLTaskMetaType::REDUCE:
377- oss << " taskData={" << taskData.reduce.Describe() << "}";408+ oss << " taskData={" << taskData.reduce.Describe() << "}";
378- break;409+ break;
379- case HccLTaskMetaType::MEM_CPY:410+ case HccLTaskMetaType::MEM_CPY:
380- oss << " taskData={" << taskData.transMem.Describe() << "}";411+ oss << " taskData={" << taskData.transMem.Describe() << "}";
381- break;412+ break;
382- case HccLTaskMetaType::CCU_GRAPH:413+ case HccLTaskMetaType::CCU_GRAPH:
383- oss << " taskData={" << taskData.ccu.Describe() << "}";414+ oss << " taskData={" << taskData.ccu.Describe() << "}";
384- break;415+ break;
385- case HccLTaskMetaType::AIV_GRAPH:416+ case HccLTaskMetaType::AIV_GRAPH:
386- oss << " taskData={" << taskData.aiv.Describe() << "}";417+ oss << " taskData={" << taskData.aiv.Describe() << "}";
387- break;418+ break;
388- case HccLTaskMetaType::SYNC_STREAM:419+ case HccLTaskMetaType::SYNC_STREAM:
389- oss << " taskData={" << taskData.syncStreamTask.Describe() << "}";420+ oss << " taskData={" << taskData.syncStreamTask.Describe() << "}";
390- break;421+ break;
391- default:422+ case HccLTaskMetaType::MODEL_EXEC:
392- break;423+ oss << " taskData={" << taskData.modelExec.Describe() << "}";
424+ break;
425+ default:
426+ break;
393 }427 }
394 return oss.str();428 return oss.str();
395 }429 }
@@ -410,13 +444,14 @@ typedef struct {
410enum BufferType { INPUT = 0, OUTPUT, CCL, RESERVED };444enum BufferType { INPUT = 0, OUTPUT, CCL, RESERVED };
411 445 
412class DataSlice {446class DataSlice {
413-public:447+ public:
414 DataSlice() : type_(BufferType::INPUT), offset_(0), size_(0) {}448 DataSlice() : type_(BufferType::INPUT), offset_(0), size_(0) {}
415- DataSlice(BufferType type, uint64_t offset, uint64_t size) : type_(type), offset_(offset), size_(size) {}449+ DataSlice(BufferType type, uint64_t offset, uint64_t size)
416- std::string Describe() const450+ : type_(type), offset_(offset), size_(size) {}
417- {451+ std::string Describe() const {
418- return "BufferType: " + std::to_string(static_cast<int>(type_)) + ", Offset: " + std::to_string(offset_)452+ return "BufferType: " + std::to_string(static_cast<int>(type_)) +
419- + ", Size: " + std::to_string(size_);453+ ", Offset: " + std::to_string(offset_) +
454+ ", Size: " + std::to_string(size_);
420 }455 }
421 inline BufferType GetType() const { return type_; }456 inline BufferType GetType() const { return type_; }
422 inline uint64_t GetOffset() const { return offset_; }457 inline uint64_t GetOffset() const { return offset_; }
@@ -425,7 +460,7 @@ public:
425 void SetOffset(uint64_t offset) { offset_ = offset; }460 void SetOffset(uint64_t offset) { offset_ = offset; }
426 void SetSize(uint64_t size) { size_ = size; }461 void SetSize(uint64_t size) { size_ = size; }
427 462 
428-private:463+ private:
429 BufferType type_;464 BufferType type_;
430 uint64_t offset_;465 uint64_t offset_;
431 uint64_t size_;466 uint64_t size_;
@@ -441,8 +476,7 @@ private:
441 *476 *
442 * @return "x86_64" 或 "aarch64"477 * @return "x86_64" 或 "aarch64"
443 */478 */
444-inline std::string GetArchStr()479+inline std::string GetArchStr() {
445-{
446#if defined(__aarch64__)480#if defined(__aarch64__)
447 return "aarch64";481 return "aarch64";
448#else482#else
@@ -450,4 +484,14 @@ inline std::string GetArchStr()
450#endif484#endif
451}485}
452 486 
453-#endif // HCOMM_HCCL_VM_SIM_COMMON_DEFS_H487+struct ErrorInfo {
488+ unsigned int errIndex;
489+ int errCode;
490+};
491+ 
492+struct RmaConnErrorInfo {
493+ struct ErrorInfo errInfo;
494+ int resType;
495+};
496+ 
497+#endif
Mtest/hccl_vm/include/sim_common_macro.h+29-21
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_COMMON_MACRO_H18#ifndef SIM_COMMON_MACRO_H
@@ -17,24 +24,25 @@
17#define UNLIKELY(x) __builtin_expect(!!(x), 0)24#define UNLIKELY(x) __builtin_expect(!!(x), 0)
18 25 
19/* 检查函数返回值, 并返回指定错误码 */26/* 检查函数返回值, 并返回指定错误码 */
20-#define HCCLVM_CHK_RET(call) \27+#define HCCLVM_CHK_RET(call) \
21- do { \28+ do { \
22- auto hcclRet = call; \29+ auto hcclRet = call; \
23- if (UNLIKELY(hcclRet != 0)) { \30+ if (UNLIKELY(hcclRet != 0)) { \
24- HCCL_VM_INFO("call trace: hcclRet -> {}", static_cast<int>(hcclRet)); \31+ HCCL_VM_INFO("call trace: hcclRet -> {}", \
25- return hcclRet; \32+ static_cast<int>(hcclRet)); \
26- } \33+ return hcclRet; \
34+ } \
27 } while (0)35 } while (0)
28 36 
29/* 检查指针, 若指针为NULL, 则记录日志, 并返回错误 */37/* 检查指针, 若指针为NULL, 则记录日志, 并返回错误 */
30-#define HCCLVM_CHK_PTR(ptr) \38+#define HCCLVM_CHK_PTR(ptr) \
31- do { \39+ do { \
32- if (UNLIKELY((ptr) == nullptr)) { \40+ if (UNLIKELY((ptr) == nullptr)) { \
33- HCCL_VM_INFO( \41+ HCCL_VM_INFO( \
34- "errNo[0x{:x}], ptr [{}] is nullptr, return HCCL_E_PTR", static_cast<unsigned int>(HCCL_SIM_E_PTR), \42+ "errNo[0x{:x}], ptr [{}] is nullptr, return HCCL_E_PTR", \
35- #ptr); \43+ static_cast<unsigned int>(HCCL_SIM_E_PTR), #ptr); \
36- return HCCL_SIM_E_PTR; \44+ return HCCL_SIM_E_PTR; \
37- } \45+ } \
38 } while (0)46 } while (0)
39 47 
40#endif // SIM_COMMON_MACRO_H48#endif // SIM_COMMON_MACRO_H
Mtest/hccl_vm/include/sim_ip_address.h+40-33
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_IP_ADDRESS_H18#ifndef SIM_IP_ADDRESS_H
@@ -51,27 +58,26 @@ union BinaryAddr {
51};58};
52 59 
53class IpAddress {60class IpAddress {
54-public:61+ public:
55- IpAddress()62+ IpAddress() {
56- {
57 scopeID_ = 0;63 scopeID_ = 0;
58 family_ = AF_INET;64 family_ = AF_INET;
59 binaryAddr_.addr.s_addr = 0;65 binaryAddr_.addr.s_addr = 0;
60 }66 }
61 67 
62- explicit IpAddress(const union BinaryAddr& ip, s32 family, s32 scopeID = 0) : family_(family), scopeID_(scopeID)68+ explicit IpAddress(const union BinaryAddr &ip, s32 family, s32 scopeID = 0)
63- {69+ : family_(family), scopeID_(scopeID) {
64 binaryAddr_ = ip;70 binaryAddr_ = ip;
65 // 区分ipv4和ipv6转eid71 // 区分ipv4和ipv6转eid
66 if (family_ == AF_INET6) {72 if (family_ == AF_INET6) {
67- (void)memcpy(eid_.raw, binaryAddr_.addr6.s6_addr, sizeof(binaryAddr_.addr6.s6_addr));73+ (void)memcpy(eid_.raw, binaryAddr_.addr6.s6_addr,
74+ sizeof(binaryAddr_.addr6.s6_addr));
68 } else {75 } else {
69 ipv4AddrToEid(binaryAddr_.addr.s_addr);76 ipv4AddrToEid(binaryAddr_.addr.s_addr);
70 }77 }
71 }78 }
72 79 
73- explicit IpAddress(u32 ipAddr)80+ explicit IpAddress(u32 ipAddr) {
74- {
75 struct in_addr addr {81 struct in_addr addr {
76 ipAddr82 ipAddr
77 };83 };
@@ -80,8 +86,7 @@ public:
80 ipv4AddrToEid(ipAddr);86 ipv4AddrToEid(ipAddr);
81 }87 }
82 88 
83- explicit IpAddress(const Eid& eidInput)89+ explicit IpAddress(const Eid &eidInput) {
84- {
85 for (uint32_t i = 0; i < URMA_EID_LEN; i++) {90 for (uint32_t i = 0; i < URMA_EID_LEN; i++) {
86 eid_.raw[i] = eidInput.raw[i];91 eid_.raw[i] = eidInput.raw[i];
87 }92 }
@@ -90,11 +95,12 @@ public:
90 (void)memcpy(binaryAddr_.addr6.s6_addr, eid_.raw, sizeof(eid_.raw));95 (void)memcpy(binaryAddr_.addr6.s6_addr, eid_.raw, sizeof(eid_.raw));
91 }96 }
92 97 
93- explicit IpAddress(const std::string& ip, s32 family = AF_INET) : family_(family)98+ explicit IpAddress(const std::string &ip, s32 family = AF_INET)
94- {99+ : family_(family) {
95- void* dst;100+ void *dst;
96 if (family != AF_INET && family != AF_INET6) {101 if (family != AF_INET && family != AF_INET6) {
97- throw std::invalid_argument("[IpAddress] Unsupported Address Family");102+ throw std::invalid_argument(
103+ "[IpAddress] Unsupported Address Family");
98 }104 }
99 if (family == AF_INET6) {105 if (family == AF_INET6) {
100 dst = &binaryAddr_.addr6;106 dst = &binaryAddr_.addr6;
@@ -103,12 +109,14 @@ public:
103 }109 }
104 int res = inet_pton(family, ip.c_str(), dst);110 int res = inet_pton(family, ip.c_str(), dst);
105 if (res == -1) {111 if (res == -1) {
106- throw std::invalid_argument("[IpAddress] Unsupported Address Family");112+ throw std::invalid_argument(
113+ "[IpAddress] Unsupported Address Family");
107 } else if (res == 0) {114 } else if (res == 0) {
108 throw std::invalid_argument("[IpAddress] Invalid Network Address");115 throw std::invalid_argument("[IpAddress] Invalid Network Address");
109 }116 }
110 if (family_ == AF_INET6) {117 if (family_ == AF_INET6) {
111- (void)memcpy(eid_.raw, binaryAddr_.addr6.s6_addr, sizeof(binaryAddr_.addr6.s6_addr));118+ (void)memcpy(eid_.raw, binaryAddr_.addr6.s6_addr,
119+ sizeof(binaryAddr_.addr6.s6_addr));
112 } else {120 } else {
113 ipv4AddrToEid(binaryAddr_.addr.s_addr);121 ipv4AddrToEid(binaryAddr_.addr.s_addr);
114 }122 }
@@ -118,8 +126,7 @@ public:
118 126 
119 Eid GetEid() const { return eid_; }127 Eid GetEid() const { return eid_; }
120 128 
121- std::string EidToHexString() const129+ std::string EidToHexString() const {
122- {
123 std::ostringstream oss;130 std::ostringstream oss;
124 oss << std::hex << std::setfill('0');131 oss << std::hex << std::setfill('0');
125 for (uint32_t i = 0; i < URMA_EID_LEN; i++) {132 for (uint32_t i = 0; i < URMA_EID_LEN; i++) {
@@ -128,34 +135,34 @@ public:
128 return oss.str();135 return oss.str();
129 }136 }
130 137 
131- std::string GetIpStr() const138+ std::string GetIpStr() const {
132- {139+ const void *src = nullptr;
133- const void* src = nullptr;
134 if (family_ == AF_INET) {140 if (family_ == AF_INET) {
135 src = &binaryAddr_.addr;141 src = &binaryAddr_.addr;
136 } else if (family_ == AF_INET6) {142 } else if (family_ == AF_INET6) {
137 src = &binaryAddr_.addr6;143 src = &binaryAddr_.addr6;
138 } else {144 } else {
139- throw std::invalid_argument("[IpAddress] Unsupported Address Family");145+ throw std::invalid_argument(
146+ "[IpAddress] Unsupported Address Family");
140 }147 }
141 char dst[INET6_ADDRSTRLEN];148 char dst[INET6_ADDRSTRLEN];
142- const char* res = inet_ntop(family_, src, dst, INET6_ADDRSTRLEN);149+ const char *res = inet_ntop(family_, src, dst, INET6_ADDRSTRLEN);
143 if (res == nullptr) {150 if (res == nullptr) {
144- throw std::invalid_argument("[IpAddress] Invalid Binary Network Address");151+ throw std::invalid_argument(
152+ "[IpAddress] Invalid Binary Network Address");
145 }153 }
146 return dst;154 return dst;
147 }155 }
148 156 
149 auto GetFamily() const { return family_; }157 auto GetFamily() const { return family_; }
150 158 
151-private:159+ private:
152 union BinaryAddr binaryAddr_ {}; // 二进制IP地址160 union BinaryAddr binaryAddr_ {}; // 二进制IP地址
153 s32 family_{AF_INET};161 s32 family_{AF_INET};
154 s32 scopeID_{0};162 s32 scopeID_{0};
155 Eid eid_{};163 Eid eid_{};
156 164 
157- void ipv4AddrToEid(const uint32_t& inAddr)165+ void ipv4AddrToEid(const uint32_t &inAddr) {
158- {
159 eid_.in4.reserved = 0;166 eid_.in4.reserved = 0;
160 eid_.in4.prefix = URMA_EID_IPV4_PREFIX;167 eid_.in4.prefix = URMA_EID_IPV4_PREFIX;
161 eid_.in4.addr = inAddr;168 eid_.in4.addr = inAddr;
Mtest/hccl_vm/include/sim_loader.h+47-32
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_LOADER_H18#ifndef SIM_LOADER_H
@@ -18,59 +25,67 @@
18#include <tuple>25#include <tuple>
19#include <vector>26#include <vector>
20 27 
21-#include "operation_data/operation_data_types.h"28+#include "sim_op_db_types.h"
22 29 
23namespace loader {30namespace loader {
24using ProxyId = uint32_t;31using ProxyId = uint32_t;
25 32 
26-using OpStage = std::vector<sim::operation::CompositeOpDetail>; // 某一次算子任务33+using OpStage = std::vector<sim::CompositeOpDetail>; // 某一次算子任务
27-using OpPipeline = std::map<uint32_t, OpStage>; // 所有算子任务34+using OpPipeline = std::map<uint32_t, OpStage>; // 所有算子任务
28 35 
29class Loader {36class Loader {
30-public:37+ public:
31 Loader();38 Loader();
32 ~Loader();39 ~Loader();
33 40 
34 // dbPath: 数据库路径41 // dbPath: 数据库路径
35 HcclResult LoadOpTaskFile(const std::string dbPath = "");42 HcclResult LoadOpTaskFile(const std::string dbPath = "");
36 43 
37- HcclResult GetSyncInfo(std::vector<sim::operation::SyncRecordTab>& syncRecords);44+ HcclResult GetSyncInfo(std::vector<sim::SyncRecordTab> &syncRecords);
38 45 
39- HcclResult GetSyncRecordsByStatus(uint8_t status, std::vector<sim::operation::SyncRecordTab>& syncRecords);
40- 
41- HcclResult LoadOpExecutionKeys(std::vector<sim::operation::OpExecutionKey>& keys);
42 HcclResult46 HcclResult
43- LoadOpExecutionByKey(const sim::operation::OpExecutionKey& key, sim::operation::OpExecution& opExecution);47+ GetSyncRecordsByStatus(uint8_t status,
44- HcclResult LoadAllOpExecutions(std::vector<sim::operation::OpExecution>& opExecutions);48+ std::vector<sim::SyncRecordTab> &syncRecords);
49+ 
50+ HcclResult LoadOpExecutionKeys(std::vector<sim::OpExecutionKey> &keys);
51+ HcclResult LoadOpExecutionByKey(const sim::OpExecutionKey &key,
52+ sim::OpExecution &opExecution);
53+ HcclResult LoadAllOpExecutions(std::vector<sim::OpExecution> &opExecutions);
45 54 
46 // Runner: 每 sync 一次调用一次，outSyncIter: 输出本次加载的 syncIter55 // Runner: 每 sync 一次调用一次，outSyncIter: 输出本次加载的 syncIter
47- HcclResult LoadRunnerSingleSync(56+ HcclResult
48- const uint32_t& outSyncIter,57+ LoadRunnerSingleSync(const uint32_t &outSyncIter,
49- std::map<uint32_t, std::vector<sim::operation::CompositeOpDetail>>& compositeDataMap);58+ std::map<uint32_t, std::vector<sim::CompositeOpDetail>>
59+ &compositeDataMap);
50 60 
51 // 根据 syncIter 查询复合算子详情61 // 根据 syncIter 查询复合算子详情
52 HcclResult LoadCompositeOpDetailBySyncIter(62 HcclResult LoadCompositeOpDetailBySyncIter(
53- uint32_t syncIter, std::map<uint32_t, std::vector<sim::operation::CompositeOpDetail>>& compositeDataMap);63+ uint32_t syncIter,
64+ std::map<uint32_t, std::vector<sim::CompositeOpDetail>>
65+ &compositeDataMap);
54 66 
55- HcclResult GetCcuChannelInfo(std::vector<sim::operation::CcuChannelTab>& channels);67+ HcclResult GetCcuChannelInfo(std::vector<sim::CcuChannelTab> &channels);
56- HcclResult GetHalfRTTInfo(std::vector<sim::operation::HalfRTTTab>& halfRTT);68+ HcclResult GetHalfRTTInfo(std::vector<sim::HalfRTTTab> &halfRTT);
57- HcclResult GetJettyMapInfo(std::vector<sim::operation::JettyMapTab>& jettyMaps);69+ HcclResult GetJettyMapInfo(std::vector<sim::JettyMapTab> &jettyMaps);
58- HcclResult GetInstrResInfo(std::vector<sim::operation::CcuInstrResTab>& instrRes);70+ HcclResult GetInstrResInfo(std::vector<sim::CcuInstrResTab> &instrRes);
59 71 
60 // 返回内部已组装好的完整 Pipeline 缓存72 // 返回内部已组装好的完整 Pipeline 缓存
61- const OpPipeline& GetOpTasks() const;73+ const OpPipeline &GetOpTasks() const;
62 74 
63- HcclResult LoadAllOpTasks(std::vector<sim::operation::OpTaskTab>& tasks, bool filterDone = false);75+ HcclResult LoadAllOpTasks(std::vector<sim::OpTaskTab> &tasks,
64- HcclResult LoadCompositeOpDetailByOpIter(76+ bool filterDone = false);
65- const std::string& commName, uint64_t commHash, uint32_t opIter,77+ HcclResult
66- std::vector<sim::operation::CompositeOpDetail>& details);78+ LoadCompositeOpDetailByOpIter(const std::string &commName,
67- HcclResult FinishOpTask(const sim::operation::OpTaskTab& task);79+ uint64_t commHash, uint32_t opIter,
80+ std::vector<sim::CompositeOpDetail> &details);
81+ HcclResult FinishOpTask(const sim::OpTaskTab &task);
68 82 
69-private:83+ private:
70 // 成员变量84 // 成员变量
71 OpPipeline opTaskCache_;85 OpPipeline opTaskCache_;
72 using OpExecutionIdentity = std::tuple<std::string, uint64_t, uint32_t>;86 using OpExecutionIdentity = std::tuple<std::string, uint64_t, uint32_t>;
73- std::map<OpExecutionIdentity, std::vector<sim::operation::OpExecutionIndexEntry>> opExecutionEntriesByKey_;87+ std::map<OpExecutionIdentity, std::vector<sim::OpExecutionIndexEntry>>
88+ opExecutionEntriesByKey_;
74};89};
75} // namespace loader90} // namespace loader
76 91 
Mtest/hccl_vm/include/sim_log.h+48-40
@@ -1,17 +1,25 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_LOG_H18#ifndef SIM_LOG_H
12#define SIM_LOG_H19#define SIM_LOG_H
13 20 
14-#define SPDLOG_ACTIVE_LEVEL SPDLOG_LEVEL_TRACE // global级别设置为最低, 只通过sink的级别控制日志输出21+#define SPDLOG_ACTIVE_LEVEL \
22+ SPDLOG_LEVEL_TRACE // global级别设置为最低, 只通过sink的级别控制日志输出
15 23 
16#include <iostream>24#include <iostream>
17#include <memory>25#include <memory>
@@ -20,7 +28,7 @@
20 28 
21#include "spdlog/spdlog.h"29#include "spdlog/spdlog.h"
22 30 
23-extern spdlog::logger* g_logger;31+extern spdlog::logger *g_logger;
24 32 
25#ifdef HCCL_VM_MODULE33#ifdef HCCL_VM_MODULE
26#define HCCL_VM_TAG "[HCCL-VM-" HCCL_VM_MODULE "]"34#define HCCL_VM_TAG "[HCCL-VM-" HCCL_VM_MODULE "]"
@@ -28,46 +36,46 @@ extern spdlog::logger* g_logger;
28#define HCCL_VM_TAG "[HCCL-VM]"36#define HCCL_VM_TAG "[HCCL-VM]"
29#endif37#endif
30 38 
31-#define HCCL_VM_TRACE(...) \39+#define HCCL_VM_TRACE(...) \
32- do { \40+ do { \
33- if (g_logger != nullptr) { \41+ if (g_logger != nullptr) { \
34- SPDLOG_LOGGER_TRACE(g_logger, HCCL_VM_TAG __VA_ARGS__); \42+ SPDLOG_LOGGER_TRACE(g_logger, HCCL_VM_TAG __VA_ARGS__); \
35- } \43+ } \
36 } while (0)44 } while (0)
37 45 
38-#define HCCL_VM_DEBUG(...) \46+#define HCCL_VM_DEBUG(...) \
39- do { \47+ do { \
40- if (g_logger != nullptr) { \48+ if (g_logger != nullptr) { \
41- SPDLOG_LOGGER_DEBUG(g_logger, HCCL_VM_TAG __VA_ARGS__); \49+ SPDLOG_LOGGER_DEBUG(g_logger, HCCL_VM_TAG __VA_ARGS__); \
42- } \50+ } \
43 } while (0)51 } while (0)
44 52 
45-#define HCCL_VM_INFO(...) \53+#define HCCL_VM_INFO(...) \
46- do { \54+ do { \
47- if (g_logger != nullptr) { \55+ if (g_logger != nullptr) { \
48- SPDLOG_LOGGER_INFO(g_logger, HCCL_VM_TAG __VA_ARGS__); \56+ SPDLOG_LOGGER_INFO(g_logger, HCCL_VM_TAG __VA_ARGS__); \
49- } \57+ } \
50 } while (0)58 } while (0)
51 59 
52-#define HCCL_VM_WARN(...) \60+#define HCCL_VM_WARN(...) \
53- do { \61+ do { \
54- if (g_logger != nullptr) { \62+ if (g_logger != nullptr) { \
55- SPDLOG_LOGGER_WARN(g_logger, HCCL_VM_TAG __VA_ARGS__); \63+ SPDLOG_LOGGER_WARN(g_logger, HCCL_VM_TAG __VA_ARGS__); \
56- } \64+ } \
57 } while (0)65 } while (0)
58 66 
59-#define HCCL_VM_ERROR(...) \67+#define HCCL_VM_ERROR(...) \
60- do { \68+ do { \
61- if (g_logger != nullptr) { \69+ if (g_logger != nullptr) { \
62- SPDLOG_LOGGER_ERROR(g_logger, HCCL_VM_TAG __VA_ARGS__); \70+ SPDLOG_LOGGER_ERROR(g_logger, HCCL_VM_TAG __VA_ARGS__); \
63- } \71+ } \
64 } while (0)72 } while (0)
65 73 
66-#define HCCL_VM_CRITICAL(...) \74+#define HCCL_VM_CRITICAL(...) \
67- do { \75+ do { \
68- if (g_logger != nullptr) { \76+ if (g_logger != nullptr) { \
69- SPDLOG_LOGGER_CRITICAL(g_logger, HCCL_VM_TAG __VA_ARGS__); \77+ SPDLOG_LOGGER_CRITICAL(g_logger, HCCL_VM_TAG __VA_ARGS__); \
70- } \78+ } \
71 } while (0)79 } while (0)
72 80 
73struct LogConfig {81struct LogConfig {
@@ -81,8 +89,8 @@ struct LogConfig {
81 bool enableCompress{false};89 bool enableCompress{false};
82};90};
83 91 
84-LogConfig LoadLogConfig(const std::string& process_name);92+LogConfig LoadLogConfig(const std::string &process_name);
85-void InitLogger(const LogConfig& config);93+void InitLogger(const LogConfig &config);
86void FlushLog();94void FlushLog();
87void DeInitLogger();95void DeInitLogger();
88 96 
Mtest/hccl_vm/include/sim_thread_safe_map.h+26-31
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_THREAD_SAFE_MAP_H18#ifndef SIM_THREAD_SAFE_MAP_H
@@ -14,23 +21,19 @@
14#include <map>21#include <map>
15#include <mutex>22#include <mutex>
16 23 
17-template <typename K, typename V>24+template <typename K, typename V> class ThreadSafeMap {
18-class ThreadSafeMap {25+ public:
19-public:26+ V &operator[](const K &key) {
20- V& operator[](const K& key)
21- {
22 std::lock_guard<std::mutex> lock(mutex_);27 std::lock_guard<std::mutex> lock(mutex_);
23 return data_[key];28 return data_[key];
24 }29 }
25 30 
26- void insert(const K& key, const V& value)31+ void insert(const K &key, const V &value) {
27- {
28 std::lock_guard<std::mutex> lock(mutex_);32 std::lock_guard<std::mutex> lock(mutex_);
29 data_[key] = value;33 data_[key] = value;
30 }34 }
31 35 
32- bool try_get(const K& key, V& output) const36+ bool try_get(const K &key, V &output) const {
33- {
34 std::lock_guard<std::mutex> lock(mutex_);37 std::lock_guard<std::mutex> lock(mutex_);
35 auto it = data_.find(key);38 auto it = data_.find(key);
36 if (it != data_.end()) {39 if (it != data_.end()) {
@@ -40,44 +43,36 @@ public:
40 return false;43 return false;
41 }44 }
42 45 
43- bool erase(const K& key)46+ bool erase(const K &key) {
44- {
45 std::lock_guard<std::mutex> lock(mutex_);47 std::lock_guard<std::mutex> lock(mutex_);
46 return data_.erase(key) > 0;48 return data_.erase(key) > 0;
47 }49 }
48 50 
49- bool contains(const K& key) const51+ bool contains(const K &key) const {
50- {
51 std::lock_guard<std::mutex> lock(mutex_);52 std::lock_guard<std::mutex> lock(mutex_);
52 return data_.find(key) != data_.end();53 return data_.find(key) != data_.end();
53 }54 }
54 55 
55- size_t size() const56+ size_t size() const {
56- {
57 std::lock_guard<std::mutex> lock(mutex_);57 std::lock_guard<std::mutex> lock(mutex_);
58 return data_.size();58 return data_.size();
59 }59 }
60 60 
61- void clear()61+ void clear() {
62- {
63 std::lock_guard<std::mutex> lock(mutex_);62 std::lock_guard<std::mutex> lock(mutex_);
64 data_.clear();63 data_.clear();
65 }64 }
66 65 
67 // 遍历操作66 // 遍历操作
68- template <typename F>67+ template <typename F> void for_each(F func) {
69- void for_each(F func)
70- {
71 std::lock_guard<std::mutex> lock(mutex_);68 std::lock_guard<std::mutex> lock(mutex_);
72- for (auto& pair : data_) {69+ for (auto &pair : data_) {
73 func(pair.first, pair.second);70 func(pair.first, pair.second);
74 }71 }
75 }72 }
76 73 
77 // 条件删除74 // 条件删除
78- template <typename Predicate>75+ template <typename Predicate> size_t erase_if(Predicate pred) {
79- size_t erase_if(Predicate pred)
80- {
81 std::lock_guard<std::mutex> lock(mutex_);76 std::lock_guard<std::mutex> lock(mutex_);
82 size_t count = 0;77 size_t count = 0;
83 for (auto it = data_.begin(); it != data_.end();) {78 for (auto it = data_.begin(); it != data_.end();) {
@@ -91,7 +86,7 @@ public:
91 return count;86 return count;
92 }87 }
93 88 
94-private:89+ private:
95 std::map<K, V> data_;90 std::map<K, V> data_;
96 mutable std::mutex mutex_;91 mutable std::mutex mutex_;
97};92};
Mtest/hccl_vm/include/sim_yaml_config.h+15-8
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_YAML_CONFIG_H18#ifndef SIM_YAML_CONFIG_H
@@ -15,7 +22,7 @@
15#include <string>22#include <string>
16 23 
17// 解析yaml文件24// 解析yaml文件
18-bool LoadYamlStringMap(25+bool LoadYamlStringMap(const std::string &yamlPath, const std::string &nodeName,
19- const std::string& yamlPath, const std::string& nodeName, std::map<std::string, std::string>& out);26+ std::map<std::string, std::string> &out);
20 27 
21#endif // SIM_YAML_CONFIG_H28#endif // SIM_YAML_CONFIG_H
Mtest/hccl_vm/include/store/store_sim_comm_memory_manager.h+34-25
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef STORE_SIM_COMM_MEMORY_MANAGER_H18#ifndef STORE_SIM_COMM_MEMORY_MANAGER_H
@@ -25,7 +32,7 @@ constexpr uint32_t kSlotsPerPage = 127099;
25constexpr uint64_t kCommPageSize = 1ULL << 30;32constexpr uint64_t kCommPageSize = 1ULL << 30;
26constexpr uint32_t kPageMagic = 0x5241504F;33constexpr uint32_t kPageMagic = 0x5241504F;
27constexpr uint32_t kMaxPages = 4;34constexpr uint32_t kMaxPages = 4;
28-constexpr uint32_t kAllocBatchSize = 4;35+constexpr uint32_t kAllocBatchSize = 1024;
29constexpr uint32_t kSlotRefCountOff = 136;36constexpr uint32_t kSlotRefCountOff = 136;
30constexpr uint32_t kSlotHeaderSize = 256;37constexpr uint32_t kSlotHeaderSize = 256;
31constexpr uint32_t kC2sLockOff = 0;38constexpr uint32_t kC2sLockOff = 0;
@@ -36,7 +43,8 @@ constexpr int32_t kSpinCountLimit = 20;
36 43 
37constexpr uint32_t kRsFdClient = 0x01U;44constexpr uint32_t kRsFdClient = 0x01U;
38constexpr uint32_t kRsFdServer = 0x00U;45constexpr uint32_t kRsFdServer = 0x00U;
39-#define MAKE_FD(slot_idx, role) (((uint64_t)(role) << 63) | (uint64_t)(slot_idx))46+#define MAKE_FD(slot_idx, role) \
47+ (((uint64_t)(role) << 63) | (uint64_t)(slot_idx))
40#define FD_SLOT_IDX(fd) ((fd) & 0x7FFFFFFFFFFFFFFFULL)48#define FD_SLOT_IDX(fd) ((fd) & 0x7FFFFFFFFFFFFFFFULL)
41#define FD_ROLE(fd) ((fd) >> 63)49#define FD_ROLE(fd) ((fd) >> 63)
42 50 
@@ -58,21 +66,22 @@ struct PageHeader {
58};66};
59 67 
60struct SlotHandle {68struct SlotHandle {
61- char* base;69+ char *base;
62- char* sendBuf;70+ char *sendBuf;
63- volatile uint32_t* sendLock;71+ volatile uint32_t *sendLock;
64- volatile uint32_t* sendSize;72+ volatile uint32_t *sendSize;
65- char* recvBuf;73+ char *recvBuf;
66- volatile uint32_t* recvLock;74+ volatile uint32_t *recvLock;
67- volatile uint32_t* recvSize;75+ volatile uint32_t *recvSize;
68};76};
69 77 
70class CommunicationMemoryManager {78class CommunicationMemoryManager {
71-public:79+ public:
72- static CommunicationMemoryManager& GetInstance();80+ static CommunicationMemoryManager &GetInstance();
73 81 
74- CommunicationMemoryManager(const CommunicationMemoryManager&) = delete;82+ CommunicationMemoryManager(const CommunicationMemoryManager &) = delete;
75- CommunicationMemoryManager& operator=(const CommunicationMemoryManager&) = delete;83+ CommunicationMemoryManager &
84+ operator=(const CommunicationMemoryManager &) = delete;
76 85 
77 bool InitPool();86 bool InitPool();
78 87 
@@ -80,9 +89,9 @@ public:
80 89 
81 SlotHandle GetSlotHandle(uint32_t slotIdx, uint32_t role);90 SlotHandle GetSlotHandle(uint32_t slotIdx, uint32_t role);
82 91 
83- int64_t Send(const SlotHandle& handle, const void* data, size_t size);92+ int64_t Send(const SlotHandle &handle, const void *data, size_t size);
84 93 
85- int64_t Recv(const SlotHandle& handle, void* data, size_t size);94+ int64_t Recv(const SlotHandle &handle, void *data, size_t size);
86 95 
87 void AddRef(uint32_t slotIdx);96 void AddRef(uint32_t slotIdx);
88 97 
@@ -90,7 +99,7 @@ public:
90 99 
91 void Shutdown();100 void Shutdown();
92 101 
93-private:102+ private:
94 struct BatchCache {103 struct BatchCache {
95 uint32_t start = 0;104 uint32_t start = 0;
96 uint32_t current = 0;105 uint32_t current = 0;
@@ -100,13 +109,13 @@ private:
100 CommunicationMemoryManager();109 CommunicationMemoryManager();
101 ~CommunicationMemoryManager();110 ~CommunicationMemoryManager();
102 111 
103- void* GetOrCreatePage(uint32_t pageId);112+ void *GetOrCreatePage(uint32_t pageId);
104 113 
105 std::mutex m_mutex;114 std::mutex m_mutex;
106- std::vector<void*> m_pages;115+ std::vector<void *> m_pages;
107 BatchCache m_batchCache;116 BatchCache m_batchCache;
108};117};
109 118 
110} // namespace sim119} // namespace sim
111 120 
112-#endif // STORE_SIM_COMM_MEMORY_MANAGER_H121+#endif // STORE_SIM_COMM_MEMORY_MANAGER_H
Mtest/hccl_vm/include/store/store_sim_comm_pool_policy.h+21-13
@@ -1,11 +1,18 @@
1/**1/**
2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.2 * Copyright (c) 2026 Huawei Technologies Co., Ltd.
3- * This program is free software, you can redistribute it and/or modify it under the terms and conditions of3+ * This program is free software, you can redistribute it and/or modify it under
4- * CANN Open Software License Agreement Version 2.0 (the "License").4+ * the terms and conditions of CANN Open Software License Agreement Version 2.0
5- * Please refer to the License for details. You may not use this file except in compliance with the License.5+ * (the "License"). Please refer to the License for details. You may not use
6- * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,6+ * this file except in compliance with the License. THIS SOFTWARE IS PROVIDED ON
7- * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.7+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
8- * See LICENSE in the root of the software repository for the full text of the License.8+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS
9+ * FOR A PARTICULAR PURPOSE. See LICENSE in the root of the software repository
10+ * for the full text of the License.
11+ */
12+ 
13+/**
14+ * AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
15+ * for the full text of the License.
9 */16 */
10 17 
11#ifndef SIM_COMM_POOL_POLICY_H18#ifndef SIM_COMM_POOL_POLICY_H
@@ -17,25 +24,26 @@ namespace sim {
17// HcclCommPool24// HcclCommPool
18// 复用区的准入策略，主机侧与设备侧共用的纯判定。阈值与上界集中在此定义。25// 复用区的准入策略，主机侧与设备侧共用的纯判定。阈值与上界集中在此定义。
19struct CommPoolPolicy {26struct CommPoolPolicy {
20- static constexpr size_t kBigBlockThreshold = 200ULL * 1024 * 1024; // 200MB，对齐 CCL buffer 默认27+ static constexpr size_t kBigBlockThreshold =
28+ 200ULL * 1024 * 1024; // 200MB，对齐 CCL buffer 默认
21 // 复用区规格 4GB，大块可申请和可寻址的上界（含等于）。超过 4GB 由29 // 复用区规格 4GB，大块可申请和可寻址的上界（含等于）。超过 4GB 由
22 // ExceedsCeiling 报错。30 // ExceedsCeiling 报错。
23 static constexpr size_t kPoolSize = 4ULL * 1024 * 1024 * 1024; // 4GB31 static constexpr size_t kPoolSize = 4ULL * 1024 * 1024 * 1024; // 4GB
24 32 
25 // 复用区共享内存名。33 // 复用区共享内存名。
26- static constexpr const char* kPoolName = "HcclCommPool";34+ static constexpr const char *kPoolName = "HcclCommPool";
27 35 
28 // 本次申请是否应引流到复用区：仅校验模式开且 threshold <= size <=36 // 本次申请是否应引流到复用区：仅校验模式开且 threshold <= size <=
29 // poolSize。37 // poolSize。
30- static bool38+ static bool ShouldRedirect(size_t size, bool checkOnlyMode,
31- ShouldRedirect(size_t size, bool checkOnlyMode, size_t threshold = kBigBlockThreshold, size_t poolSize = kPoolSize)39+ size_t threshold = kBigBlockThreshold,
32- {40+ size_t poolSize = kPoolSize) {
33 return checkOnlyMode && size >= threshold && size <= poolSize;41 return checkOnlyMode && size >= threshold && size <= poolSize;
34 }42 }
35 43 
36 // 本次申请是否超出复用区上界、应报错：仅仅校验模式下拦 size > poolSize。44 // 本次申请是否超出复用区上界、应报错：仅仅校验模式下拦 size > poolSize。
37- static bool ExceedsCeiling(size_t size, bool checkOnlyMode, size_t poolSize = kPoolSize)45+ static bool ExceedsCeiling(size_t size, bool checkOnlyMode,
38- {46+ size_t poolSize = kPoolSize) {
39 return checkOnlyMode && size > poolSize;47 return checkOnlyMode && size > poolSize;
40 }48 }
41};49};