!182 sig executor 20210805

Merge pull request !182 from baihuawei/sigs

sigs/executor/README.md

@@ -1,26 +1,28 @@
 # MindSpore Executor Special Interest Group (SIG)
 
 This is the working repo for the Executor Special Interest Group (SIG). This repo contains all the artifacts, materials, meeting notes and proposals regarding **graph optimizing** , **graph compiling**, **hardware runtime**, **diagnostics**, and any other backend programs in MindSpore. Feedbacks and contributions are welcomed.
-1. **Graph Optimizing**: Similar to graph "complier", there are many Passes to fuse operators and optimize graphs.
+
+1. **Graph Optimizing**: Similar to graph "compiler", there are many Passes to fuse operators and optimize graphs.
 2. **Graph Compiling**: Allocating memory and other resources, compiling graphs to generate runnable models.
 3. **Hardware Runtime**: Interface for hardware, who supports ascend910, gpu and cpu now.
 4. **Diagnostics**: Provide some tools such as profiling, exporting graph ir and exporting operator data.
 
-# SIG Leads
+## SIG Leads
 
 * Wang Kaisheng (Huawei)
 
-# Logistics
+## Logistics
 
 * SIG leads will drive the meeting.
 * Meeting announcement will be posted on our gitee channel: https://gitee.com/mindspore/community/tree/master/sigs/executor
 * Feedbacks and topic requests are welcomed by all.
 
-# Discussion
+## Discussion
 
 * Slack channel: https://app.slack.com/client/TUKCY4QDR/C0111A9Q24B?cdn_fallback=2
 * Documents and artifacts: https://gitee.com/mindspore/community/tree/master/sigs/executor
 
-# Meeting notes
+## Meeting notes
 
 * [Thursday June 4, 2020](./meetings/001-20200604.md)
+* [Thursday August 5, 2021](./meetings/001-20210805.md)

sigs/executor/meetings/001-20210805.md

@@ -0,0 +1,35 @@
+# Thursday August 5, 2021 at 17:00pm GMT+8
+
+## Agenda
+
+- Introduce the current Ascend memory allocation strategy, that is, allocate memory at compile stage. The input and output memory of the graph cannot be reused, which limits the scale of the model under given devices.
+- Discuss several feasible ways to reuse input and output memory of graph when allocating memory in the graph running stage, and potential performance problems.
+- Overview of the memory swap strategy under runtime memory allocation and the memory benefits it brings.
+
+## Conference links
+
+- Meeting Time:
+
+    2021-08-05 19:00-21:00 ((UTC+08:00)Beijing)
+
+- Meeting Link:
+
+    https://welink.zhumu.com/j/209111783
+
+## Attendees
+
+- Wang Kaisheng (Huawei)
+- Jiang Jianfei (Huawei)
+- Chu Jinjin (Huawei)
+- Lai Yongqiang (Huawei)
+- Bai Huawei (Huawei)
+
+## Notes
+
+- The meeting video can be found:
+
+  [Executor & Serving SIG Meeting 2021](https://www.bilibili.com/video/BV1PX4y1F7i1)
+
+## Action items
+
+- None.