window下源码编译mmcv-full==1.2.1
软件准备：
Git、vs2019_community、Miniconda3-4.6.14、cuda_10.2+cudnn_7.6.5
一、确认NVIDIA GeForce RTX 2080 Ti已安装
cmd 命令：nvidia-smi OR nvidia-smi -L
*************************************
>nvidia-smi -L
GPU 0: NVIDIA GeForce RTX 2080 Ti (UUID: GPU-6a2b3f72-b96d-f2d4-bb59-87e7e869b070)
二、下载Miniconda3安装并设置环境变量
1. Miniconda3-4.6.14-Windows-x86_64.exe
https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
2. 设置环境变量 Set Environment variable:
F:Miniconda3Libraryin
F:Miniconda3Scripts
F:Miniconda3
三、打开Anaconda prompt并创建虚拟环境py37，Launch Anaconda prompt:
conda create -n py37 python=3.7
conda activate py37
# conda remove -n py37 --all# remove 虚拟环境py37
# CUDA 10.2 安装pytorch和torchvision
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch
四、查看torch依赖的cuda和cudnn版本，并下载对应版本的cuda和cudnn，并安装
>python
import torch
print(torch.__version__)
print(torch.version.cuda)
print(torch.backends.cudnn.version())
五、git clone mmcv==1.2.1
git clone https://github.com/open-mmlab/mmcv.git
cd mmcv
git checkout 91a7fee
pip3 install -r requirements.txt
六、设置编译器 Set up MSVC compiler: Set Environment variable, add
C:Program Files (x86)Microsoft Visual Studio2019CommunityVCToolsMSVC14.29.30037inHostx86x64
to PATH, so that cl.exe will be available in prompt
full version with CUDA: Make sure CUDA_PATH or CUDA_HOME is already set, as
CUDA_PATH C:Program FilesNVIDIA GPU Computing ToolkitCUDAv10.2
CUDA_PATH_V10_2 C:Program FilesNVIDIA GPU Computing ToolkitCUDAv10.2
七、设置CUDA的算力Set CUDA target arch
# Suppose you are using RTX 2080 Ti , which is of capability 7.5
TORCH_CUDA_ARCH_LIST=7.5
八、可选，设置CPU运行配置# Optional
MMCV_WITH_OPS=1
MAX_JOBS=16# based on available number of CPU cores and amount of memory
九、build、install、check mmcv
cd mmcv
# build
python setup.py build_ext # if success, cl will be launched to compile ops
# install
python setup.py develop
# check
pip list
其他：查看物理CPU数、CPU核心数、线程数
在cmd命令中输入“wmic”，然后在出现的新窗口中输入“cpu get *”
NumberOfCores：表示CPU核心数
NumberOfLogicalProcessors：表示CPU线程数
cmd->wmic
cmd->CPU GET NumberOfCores,NumberOfLogicalProcessors
NumberOfCores NumberOfLogicalProcessors
16 32
16 32