From af75165cb7f22e53140bf2f3ca73acd99dbad29d Mon Sep 17 00:00:00 2001 From: zhoufeng Date: Fri, 13 Nov 2020 11:37:11 +0800 Subject: [PATCH] update sigs leads Signed-off-by: zhoufeng --- sigs/akg/README.md | 8 ++++---- sigs/compiler/README.md | 2 +- sigs/executor/README.md | 2 +- sigs/frontend/README.md | 2 +- sigs/graphengine/README.md | 2 +- sigs/parallel/README.md | 4 ++-- sigs/security/README.md | 2 +- sigs/visualization/README.md | 2 +- 8 files changed, 12 insertions(+), 12 deletions(-) diff --git a/sigs/akg/README.md b/sigs/akg/README.md index 80e36d8..9819105 100644 --- a/sigs/akg/README.md +++ b/sigs/akg/README.md @@ -1,14 +1,14 @@ # MindSpore AKG Special Interest Group (SIG) This is the working repo for the AKG Special Interest Group (SIG). This repo contains all the artifacts, materials, meeting notes and proposals regarding **Normalization** , **Auto Schedule**, **Instruction Emit**, **Backend Optimization**. Feedbacks and contributions are welcomed. -1. **Normalization**: The mainly optimization of normalization includes three address transform, common subexpression elimination, copy propagation and so on. -2. **Auto Schedule**: The auto schedule module mainly have vectorization, loop tiling, mem promotion and loop distribution. -3. **Instruction Emit**: The instruction emitting module has the optimization about loop normalization, auto pragma and emit instruction. +1. **Normalization**: The mainly optimization of normalization includes three address transform, common subexpression elimination, copy propagation and so on. +2. **Auto Schedule**: The auto schedule module mainly have vectorization, loop tiling, mem promotion and loop distribution. +3. **Instruction Emit**: The instruction emitting module has the optimization about loop normalization, auto pragma and emit instruction. 4. **Backend Optimization**: The backend optimization module consists of double buffer optimization, storage rewrite optimization and inject sync optimization. # SIG Leads -* Zhang Kun (Huawei) +* Zhang Renwei (Huawei) # Logistics diff --git a/sigs/compiler/README.md b/sigs/compiler/README.md index 888c2c9..3615396 100644 --- a/sigs/compiler/README.md +++ b/sigs/compiler/README.md @@ -10,7 +10,7 @@ This is the working repo for the Compiler special interest group (SIG). This rep ## SIG Leads -* Wei Luning (Huawei) +* Zhang Qinghua (Huawei) ## Logistics diff --git a/sigs/executor/README.md b/sigs/executor/README.md index 55a0e4f..ad30b96 100644 --- a/sigs/executor/README.md +++ b/sigs/executor/README.md @@ -8,7 +8,7 @@ This is the working repo for the Executor Special Interest Group (SIG). This rep # SIG Leads -* Lu Pengcheng (Huawei) +* Wang Kaisheng (Huawei) # Logistics diff --git a/sigs/frontend/README.md b/sigs/frontend/README.md index 8f42ae5..18d1d5b 100644 --- a/sigs/frontend/README.md +++ b/sigs/frontend/README.md @@ -8,7 +8,7 @@ This is the working repo for the FrontEnd Special Interest Group (SIG). This rep # SIG Leads -* Deng Yiping (Huawei) +* Jin Xiaoxian (Huawei) # Logistics diff --git a/sigs/graphengine/README.md b/sigs/graphengine/README.md index e0773ac..4bccb69 100644 --- a/sigs/graphengine/README.md +++ b/sigs/graphengine/README.md @@ -6,7 +6,7 @@ This is the working repo for the GraphEngine Special Interest Group (SIG). This # SIG Leads -* Bao Chong (Huawei) +* Yang Wei (Huawei) # Logistics diff --git a/sigs/parallel/README.md b/sigs/parallel/README.md index 5e8f5f7..e575bfe 100644 --- a/sigs/parallel/README.md +++ b/sigs/parallel/README.md @@ -2,13 +2,13 @@ This is the working repository for the Parallel Special Interest Group (SIG). This repository contains all the artifacts, materials, meeting notes and proposals regarding **Auto-parallel**, **Model-parallel**, **Pipelined model-parallel**, **Tensor partitioning**, **Cost model**. Feedback and contributions are welcome. 1. **Auto-parallel**: The sizes of popular DNN models are getting larger, thus it is desired to automatically find an efficient way to parallelize the execution (training and inference) of the giant DNNs. This is the ultimate goal of this SIG. -2. **Model-parallel**: Unlike Data-parallel in which each device holds the entire model in training, Model-parallel is to partition the model to available devices, so that each device holds a slice of the entire model. Model-parallel is a more suitable approach for training giant models. +2. **Model-parallel**: Unlike Data-parallel in which each device holds the entire model in training, Model-parallel is to partition the model to available devices, so that each device holds a slice of the entire model. Model-parallel is a more suitable approach for training giant models. 3. **Pipelined model-parallel**: This is a paradigm to implement Model-parallel. This paradigm is to assign operators of a DNN model to different devices, so that different training batches can be pipelined. 4. **Tensor partitioning**: This is another paradigm to implement Model-parallel. This paradigm is to partition tensors of each operator in a DNN model, so that the devices obtain *symmetric* sequences of sliced operators. # SIG Leads -* Cheng Li (University of Science and Technology of China) +* Zhai Zhiqiang (Huawei) # Logistics diff --git a/sigs/security/README.md b/sigs/security/README.md index d7183e8..2075bb7 100644 --- a/sigs/security/README.md +++ b/sigs/security/README.md @@ -7,7 +7,7 @@ This is the working repo for the MindArmour special interest group (SIG). This r # SIG Leads -* Liu Liu (Huawei) +* Wang Ze (Huawei) # Logistics diff --git a/sigs/visualization/README.md b/sigs/visualization/README.md index 81c1e64..4da0f93 100644 --- a/sigs/visualization/README.md +++ b/sigs/visualization/README.md @@ -4,7 +4,7 @@ This is the working repo for the `Visualization` special interest group (SIG). T # SIG Leads -* Gao congli (Huawei) +* Gao Congli (Huawei) # Logistics