You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 
taoxiangdong 55fd4c4c14 decouple cce in graphengine 3 years ago
..
analyzer !107 bugfix:fix core question because of DestroyGraphJson lack of return 4 years ago
client external interface modify: string change to ascendstring 4 years ago
common !499 inference supports dynamic shape 3 years ago
engine_manager ge executor support windows 4 years ago
executor !499 inference supports dynamic shape 3 years ago
ge_local_engine inference supports dynamic shape 3 years ago
ge_runtime security check modify 3 years ago
generator !466 dts:error message add 4 years ago
graph decouple cce in graphengine 3 years ago
host_cpu_engine inference supports dynamic shape 3 years ago
host_kernels modified: ge/host_kernels/strided_slice_kernel.cc 4 years ago
hybrid decouple cce in graphengine 3 years ago
inc update code from yellow zone 20201013 4 years ago
init ge executor support windows 4 years ago
ir_build !466 dts:error message add 4 years ago
model update newest yellow zone code 20201020 4 years ago
offline update atc depend on static_mmpa 4 years ago
omm ge executor support windows 4 years ago
opskernel_manager log rectify 4 years ago
plugin/engine remove soft link and support other scene cmake 4 years ago
proto aipp config add related input name and modify proto 4 years ago
session !453 atc json remove unknow shape check 4 years ago
single_op decouple cce in graphengine 3 years ago
stub remove soft link and support other scene cmake 4 years ago
CMakeLists.txt support mindspore building tests 4 years ago
README.md update cmake 4 years ago
ge_inference.mk update cmakelist find libmmpa.a 4 years ago
ge_runner.mk update cmakelist find libmmpa.a 4 years ago
module.mk update ge folder 4 years ago

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示

C++ Text Makefile Shell CMake other