fix_bug: analysis error of read-write relationship in multi-filter · Pull Request !778 · MindSpore/akg

<!--  Thanks for sending a pull request!  Here are some tips for you:

If this is your first time, please read our contributor guidelines: https://gitee.com/mindspore/akg/blob/master/CONTRIBUTING.md
-->

**What type of PR is this?**
> Uncomment only one ` /kind <>` line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug

**What does this PR do / why do we need it**:

```
{"composite":true,"composite_graph":"11396.11396","id":0,"input_desc":[[{"data_type":"float32","format":"DefaultFormat","shape":[1],"tensor_name":"input_3"}],[{"data_type":"float32","format":"DefaultFormat","shape":[12288,6288],"tensor_name":"input_12"}],[{"data_type":"float32","format":"DefaultFormat","shape":[12288,6288],"tensor_name":"input_13"}],[{"data_type":"float32","format":"DefaultFormat","shape":[12,1024],"tensor_name":"input_0"}],[{"data_type":"float32","format":"DefaultFormat","shape":[1],"tensor_name":"input_2"}]],"op":"Fused_Reshape_Mul_Add_RealDiv_Reshape_Mul_Reshape_BroadcastTo_Sub_split_9015124442605579809","op_desc":[{"attr":[{"data_type":"listInt","name":"shape","value":[-1]}],"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[12,1024],"tensor_name":"input_0"}]],"name":"Reshape","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[12288],"tensor_name":"output_0_0"}]},{"attr":null,"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"input_1","value":1.0}],[{"data_type":"float32","format":"DefaultFormat","name":"input_1","shape":[1],"tensor_name":"input_2"}]],"name":"Mul","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_1"}]},{"attr":null,"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"input_3"}],[{"data_type":"float32","format":"DefaultFormat","name":"input_1","shape":[1],"tensor_name":"input_4","value":9.999999747378752e-06}]],"name":"Add","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_2"}]},{"attr":null,"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"output_0_1"}],[{"data_type":"float32","format":"DefaultFormat","name":"input_1","shape":[1],"tensor_name":"output_0_2"}]],"name":"RealDiv","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_3"}]},{"attr":[{"data_type":"listInt","name":"shape","value":[]}],"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"output_0_3"}]],"name":"Reshape","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_4"}]},{"attr":null,"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"output_0_4"}],[{"data_type":"float32","format":"DefaultFormat","name":"input_1","shape":[1],"tensor_name":"input_9","value":0.125}]],"name":"Mul","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_5"}]},{"attr":[{"data_type":"listInt","name":"shape","value":[1]}],"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"output_0_5"}]],"name":"Reshape","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[1],"tensor_name":"output_0_6"}]},{"attr":[{"data_type":"listInt","name":"shape","value":[12288]}],"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[1],"tensor_name":"output_0_6"}]],"name":"BroadcastTo","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[12288],"tensor_name":"output_0_7"}]},{"attr":null,"impl_path":"","input_desc":[[{"data_type":"float32","format":"DefaultFormat","name":"input_0","shape":[12288,6288],"tensor_name":"input_12"}],[{"data_type":"float32","format":"DefaultFormat","name":"input_1","shape":[12288,6288],"tensor_name":"input_13"}]],"name":"Sub","output_desc":[{"data_type":"float32","format":"DefaultFormat","name":"output_0","shape":[12288,6288],"tensor_name":"output_0_8"}]}],"output_desc":[{"data_type":"float32","format":"DefaultFormat","shape":[12288],"tensor_name":"output_0_0"},{"data_type":"float32","format":"DefaultFormat","shape":[12288],"tensor_name":"output_0_7"},{"data_type":"float32","format":"DefaultFormat","shape":[12288,6288],"tensor_name":"output_0_8"},{"data_type":"float32","format":"DefaultFormat","shape":[1],"tensor_name":"output_0_2"}],"platform":"AKG","process":"cuda","target_info":{"compute_capability":"8.0","sm_count":108},"version":1}
```
![输入图片说明](https://images.gitee.com/uploads/images/2022/0729/094900_55057cb8_7432416.png "屏幕截图.png")
![输入图片说明](https://images.gitee.com/uploads/images/2022/0729/095142_09bcde50_7432416.png "屏幕截图.png")

在整图tunning过程中，该类算子经过poly调度之后，最外层为set节点，并且第一个filter节点中不存在可以做向量化的轴，但是分析向量化轴时是在全局的读写关系中（而不是在当前语句的读写关系中）分析的，导致该filter节点被强制赋予一个向量化轴，导致结果错误

**Which issue(s) this PR fixes**:

Fixes #
【AR】实现基于图算融合的推理整图Tuning
https://e.gitee.com/mind_spore/projects/69994/milestones/170997/issues/table?issue=I5I8ON

**Special notes for your reviewers**:

MindSpore/akg关闭 .gitee-modal { width: 500px !important; }

搜索帮助

MindSpore/akg
关闭