CS_ALearn/openmlsys-zh
1
0
Fork 0
mirror of https://github.com/openmlsys/openmlsys-zh.git synced 2026-04-24 02:24:05 +08:00
Code Issues Projects Releases Wiki Activity

feat(accelerator): update according to the book (#423)

Browse Source 
* update accelerator_practise.md

* update images in accelerator_practise.md

* update images in accelerator_practise.md

* feat(accelerator): update according to the book

---------

Co-authored-by: Went-Liang <wenteng_liang@163.com>
This commit is contained in:
Hello_World
2023-03-06 15:28:36 +08:00
committed by GitHub
parent a0912ecf41
commit f80e6559b0
20 changed files with 64 additions and 775 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
chapter_accelerator/accelerator_programming.md
View File

@@ -75,6 +75,8 @@ AKG则是MindSpore社区的开源算子编译工具。与上述介绍的算子
- **算子库层级**
:label:`sec-accelerator-use-cublas`
在上述不同层级的编程方式中，直接调用算子加速库使能加速器无疑是最快捷高效的方式。NVIDIA提供了cuBLAS/cuDNN两类算子计算库，cuBLAS提供了使能张量计算核心的接口，用以加速矩阵乘法(GEMM)运算，cuDNN提供了对应接口加速卷积(CONV)运算等。
以 :numref:`accelerator-programable-title`小节的GEMM运算为例，与常规CUDA调用cuBLAS算子库相似，通过cuBLAS加速库使能张量计算核心步骤包括：