您的当前位置：首页colab ssh、frida ,Shark 、opennn 、bash-completion、Dwarf 、tensorflow编译、pytorch编译、bazel安装、git-tui
colab ssh、frida ,Shark 、opennn 、bash-completion、Dwarf 、tensorflow编译、pytorch编译、bazel安装、git-tui
来源：锐游网
centos7 install git2.x
wget https://packages.endpointdev.com/rhel/7/os/x86_64/git-core-doc-2.38.1-1.ep7.noarch.rpm
wget https://packages.endpointdev.com/rhel/7/os/x86_64/git-all-2.38.1-1.ep7.noarch.rpm
wget https://packages.endpointdev.com/rhel/7/os/x86_64/git-2.38.1-1.ep7.x86_64.rpm
wget https://packages.endpointdev.com/rhel/7/os/x86_64/git-core-2.38.1-1.ep7.x86_64.rpm
wget https://packages.endpointdev.com/rhel/7/os/x86_64/perl-Git-2.38.1-1.ep7.noarch.rpm
yum -y install perl-Git-2.38.1-1.ep7.noarch.rpm git-2.38.1-1.ep7.x86_64.rpm git-core-doc-2.38.1-1.ep7.noarch.rpm
kernel header not found for target kernel: vbox guest iso安装
0. 背景
-1. 如果不安装guest iso中的驱动，虚拟机centos7中的网络带宽被限制在2MByte左右，安装完后虚拟机centos7的网速正常
0. vbox中的虚拟机操作系统为 centos 7 x64
mkdir /media/cd/
mount /dev/cdrom /media/cd
cd /media/cd/
./VBoxLinuxAdditions.run #此时报错 kernel header not found for target kernel
1. 报错如下：
./VBoxLinuxAdditions.run
"""Verifying archive integrity... 100% MD5 checksums are OK. All good.
Uncompressing VirtualBox 7.0.6 Guest Additions for Linux 100%
VirtualBox Guest Additions installer
Removing installed version 7.0.6 of VirtualBox Guest Additions...
VirtualBox Guest Additions: Starting.
VirtualBox Guest Additions: Setting up modules
VirtualBox Guest Additions: Building the VirtualBox Guest Additions kernel
modules. This may take a while.
VirtualBox Guest Additions: To build modules for other installed kernels, run
VirtualBox Guest Additions: /sbin/rcvboxadd quicksetup <version>
VirtualBox Guest Additions: or
VirtualBox Guest Additions: /sbin/rcvboxadd quicksetup all
VirtualBox Guest Additions: Kernel headers not found for target kernel
3.10.0-1160.el7.x86_64. Please install them and execute
/sbin/rcvboxadd setup
modprobe vboxguest failed
The log file /var/log/vboxadd-setup.log may contain further information.
"""
2. 解决报错
uname -r
#3.10.0-1160.el7.x86_64
yum -y install gcc make #没有gcc报错也是 : kernel header not found for target kernel, 并不会报没有gcc
yum -y install kernel-devel-`uname -r` kernel-headers-`uname -r`
#/sbin/rcvboxadd quicksetup `uname -r` #如果报错中的版本 和 uname -r 版本不一致， 运行此行
reboot #重启虚拟机后重新安装guest iso
cd /media/cd/
./VBoxLinuxAdditions.run #正常安装
go0gle c0lab ssh
1. nggr0k注册并拿到token
2. coolab安装coolab_ssh
# google coolab有检测coolab_ssh ，所以要劈开
!pip install `echo col``echo ab``echo _ssh`
nggr0kToken='xxx'
paswd='123'
from c0lab_ssh import launch_ssh, init_git
launch_ssh(nggr0kToken,paswd)
#输出类似如下：
"""
Host google_coolab_ssh
HostName 1.tcp.nggr0k.io
User root
Port 11234
"""
3. using online web ssh client
4. style transform ： VToonify
: 也是以下脚本的参考来源
git clone https://github.com/williamyang1991/VToonify.git
cd VToonify
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
source /root/miniconda3/bin/activate
conda env create -f ./environment/vtoonify_env.yaml
注意：claash 实际多写了一个a, prooxychains实际多写了一个o
linux 下prooxy
claash
“WARN[0000] MMDB invalid, remove and download”
修改 1670336651611.yml 修改外部控制端口 为 7777
1670336651611.yml: 从claash for windows中导出
#...
external-controller: 7777
#...
cd /claash-home/
#此即解决claash启动报错:"WARN[0000] MMDB invalid, remove and download"
wget https://github.com/P3TERX/GeoLite.mmdb/raw/download/GeoLite2-Country.mmdb -O Country.mmdb
wget https://github.com/Dreamacro/claash/releases/download/v1.9.0/claash-linux-amd64-v1.9.0.gz
gzip -d claash-linux-amd64-v1.9.0.gz
#1670336651611.yml: 从claash for windows中导出
./claash-linux-amd64-v1.9.0 -d . -f 1670336651611.yml -ext-ctl 10.11.1.107:7777
"""
INFO[0000] Start initial compatible provider 最新域名
INFO[0000] Start initial compatible provider 动画疯
INFO[0000] Start initial compatible provider 故障切换
INFO[0000] Start initial compatible provider 自动选择
INFO[0000] Start initial compatible provider 手动选择
INFO[0000] Start initial compatible provider 节点选择
INFO[0000] Start initial compatible provider 国外网站
INFO[0000] Start initial compatible provider 电报吹水
INFO[0000] HTTP proxy listening at: [::]:7890
INFO[0000] RESTful API listening at: [::]:9090
INFO[0000] SOCKS proxy listening at: [::]:7891
ERRO[0000] Start DNS server error: missing port in address
"""
claash dns配置(对 原配置文件1670336651611.yml 增加dns配置)
dns:
enable: true
ipv6: false
enhanced-mode: fake-ip
fake-ip-range: 192.168.0.1/24
fake-ip-filter:
- "*.lan"
- "*.local"
- dns.msftncsi.com
- www.msftncsi.com
- www.msftconnecttest.com
- stun.*.*.*
- stun.*.*
- miwifi.com
- music.163.com
- "*.music.163.com"
- "*.126.net"
- api-jooxtt.sanook.com
- api.joox.com
- joox.com
- y.qq.com
- "*.y.qq.com"
- streamoc.music.tc.qq.com
- mobileoc.music.tc.qq.com
- isure.stream.qqmusic.qq.com
- dl.stream.qqmusic.qq.com
- aqqmusic.tc.qq.com
- amobile.music.tc.qq.com
- "*.xiami.com"
- "*.music.migu.cn"
- music.migu.cn
- netis.cc
- router.asus.com
- repeater.asus.com
- routerlogin.com
- routerlogin.net
- tendawifi.com
- tendawifi.net
- tplinklogin.net
- tplinkwifi.net
- tplinkrepeater.net
- "*.ntp.org.cn"
- "*.openwrt.pool.ntp.org"
- "*.msftconnecttest.com"
- "*.msftncsi.com"
- localhost.ptlogin2.qq.com
- "*.*.*.srv.nintendo.net"
- "*.*.stun.playstation.net"
- xbox.*.*.microsoft.com
- "*.ipv6.microsoft.com"
- "*.*.xboxlive.com"
- speedtest.cros.wr.pvp.net
default-nameserver:
- 10.11.1.1
- 114.114.114.114
nameserver:
- https://doh.pub/dns-query
- https://dns.alidns.com/dns-query
- https://dns.google/dns-query
#以下是1670336651611.yml的原本内容
port: 7890
socks-port: 7891
即启动了claash的dns解析， 以上报错"ERRO[0000] Start DNS server error: missing port in address"会消失
linux claash 的web client（http://claash.razord.top/#/proxies） （连接本地linux claash的external-controller端口7777)
控制 linux claash , 打开 ,
操作和claash for windows一样的
本地chrome 打开 http://claash.razord.top/#/proxies,
确保 本地chrome 已经安装了 switchyOmega插件 (连接到本地windows claash以科学上网）
注意 要设置 switchyOmega插件 强调： 10.11.1.107 不需要用代理
输入以上 10.11.1.107 和 7777, secret留空 即可 连上 linux中的claash
修改 claash on linux使用某个能正常连接的代理点
proxy (prooxychains)
配置文件 : /etc/prooxychains.conf 内容如下:
#file:/etc/prooxychains.conf
strict_chain
#proxy_dns #这行去掉, dns交给claash
tcp_read_time_out 15000
tcp_connect_time_out 8000
[ProxyList]
socks5 127.0.0.1 7891
#注意看以上claash启动日志, claash 的socks5端口是7891 而不是7890
#使用举例
prooxychains curl https://www.google.com
prooxychains w3m https://www.google.com
docker pull 无法使用 prooxychains4解决
mkdir -p /etc/systemd/system/docker.service.d
vim /etc/systemd/system/docker.service.d/https-proxy.conf
[Service]
Environment="HTTPS_PROXY=socks5://127.0.0.1:1080"
prooxychains4 docker pull fedora:36
frida
frida install
pip install frida==16.0.7 frida-tools==12.0.4
frida script debug没什么用（v8运行时 中有的变量，在真实运行时中根本没有）
frida -f d:\instrmcpp\dork\cmake-build-debug\dork.exe --debug --runtime=v8 -l d:\frida-home\frida-agent-4instrmcpp\attach_operator_new.js --pause
frida 调试 script步骤如下:
但此方法调试, args 其内容 估计和 不加 runtime=v8 不大一样
frida 的args是个数组， 但frida不知道其长度， 只能自己决定其长度（比如从调试信息中拿到其长度、或从源码中拿到其长度）
,
frida script example
frida -f /machine_learning-home/shark/build_debug/bin/Statistics --debug --pause
ls=DebugSymbol.findFunctionsMatching("*C2Ev").map(functionAddressK => DebugSymbol.fromAddress(functionAddressK));
/*ls[0]
{
"address": "0x7f03b53c2230",
"column": 0,
"fileName": "",
"lineNumber": 0,
"moduleName": "libstdc++.so.6.0.28",
"name": "_ZNSt14error_categoryC2Ev"
}
*/
moduleLs = ls.reduce((s, k) => { if (!s.includes(k.moduleName)) return [...s, k.moduleName]; else return s; }, []);
/*
[
"libstdc++.so.6.0.28",
"Statistics"
]
*/
funcLs_Statistics=ls.filter(fK=>fK.moduleName=="Statistics");
frida 运行的应用如何带参数
-
用frida拦截应用程序内的main函数并用frida添加参数， 以给应用程序提供更通用的参数
参考：
“Failed to load script: timeout was reached” 解决办法
参考:
//假设本文件名为 example.js
setTimeout(function () {
//把原有的js代码移到这里, 即不会再报错误："Failed to load script: timeout was reached"
}, 0);
#运行frida, 注入脚本 example.js, 由于 example.js 中指定了不会超时, 因此frida不会再报错误："Failed to load script: timeout was reached"
frida -f xxx_dork.exe --load example.js
dork list (possible)
ml dork
下面的看着就难编译:
dork list
build , Relase
#WSL2 Ubuntu-20.04
apt info libboost-all-dev #Boost version (currently 1.71)
apt install -y libboost-all-dev
cd /machine_learning-home/
git clone git@gitcode.net:machine_learning/shark.git
cd /machine_learning-home/shark/
cmake -S . -B build
cd build; make -j 8
#test:
/machine_learning-home/shark/build/bin/Statistics
build, Debug
#...
cmake -S . -B build_debug -DCMAKE_BUILD_TYPE=Debug
cd build_debug/ ; make -j 8
#...
Relase diff Debug
cd /machine_learning-home/shark/build; tree -L 5 > /tmp/release.tree
cd /machine_learning-home/shark/build_debug; tree -L 5 > /tmp/debug.tree
diff /tmp/release.tree /tmp/debug.tree
"""2839c2839
< │ └── libshark.a
---
> │ └── libshark_debug.a
"""
opennn
cmake -S . -B build_debug -DCMAKE_BUILD_TYPE=Debug -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
cd build_debug ; make -j 8
ln -s /machine_learning-home/opennn/examples/iris_plant/data /machine_learning-home/opennn/build/examples/data
cd /machine_learning-home/opennn/build/examples/iris_plant/
#./iris_plant #正常运行
frida -f ./iris_plant --debug --pause
"""
ls=DebugSymbol.findFunctionsMatching("*").map(functionAddressK => DebugSymbol.fromAddress(functionAddressK));
//试着显示全部函数, 这估计挺慢的.
"""
#dependency:
sudo apt install pytest-benchmark
sudo ln -s /usr/bin/pytest-3 /usr/bin/pytest
sudo apt install python3-pexpect
#build and install:
git clone git@gitcode.net:tmp/bash-completion.git
cd bash-completion
autoreconf -i
./configure
make
sudo make check
sudo make install
#sudo make uninstall#
#run:
echo """
# Use bash-completion, if available
[[ $PS1 && -f /usr/share/bash-completion/bash_completion ]] && \
. /usr/share/bash-completion/bash_completion
""" >> ~/.bashrc
#if upper install occur error: _comp_initialize : command not found,
#fix:
cd bash-completion && sudo make uninstall && cd -
#then install bash-completion by apt-get
sudo apt-get install --reinstall bash-completion
source /etc/bash_completion
调试信息查看
调试信息格式举例: stabs, COFF, PE-COFF, OMF, IEEE-695 , DWARF
: SymView, Pdbripper
Dwarf调试信息查看用
#可用 XELFViewer 查看 以下 elf可执行文件 中的 调试信息 (符号表)
#/machine_learning/shark/build_debug/bin/Statistics
#d:\instrmcpp\dork_simple\User.cpp #将其编译成a.out: g++ -g User.cpp
#/machine_learning/opennn/build_debug/examples/mnist/mnist
#看到的函数名 形如:
# _ZNSt12_Vector_baseIjSaIjEE12_Vector_implD2Ev
# _ZN4UserD2Ev , _ZN4UserC4Ev
#作为对照, 用ida打开这些elf文件, 可以看到 以上 _ZN* 对应的 源码中的函数名
_ZN*D*Ev , _ZN*C*Ev
_ZN*D*Ev , _ZN*C*Ev
tensorflow编译
tensorflow编译v2.11.0
1.构建工具bazel安装 （使用清华镜像)
参考:
sudo apt install apt-transport-https curl gnupg
#curl -fsSL https://bazel.build/bazel-release.pub.gpg | gpg --dearmor >bazel-archive-keyring.gpg
curl -fsSL https://gitee.com/pubz/misc/raw/36806b6978e730016b2cf3a89e73e55094088fd2/bazel-release.pub.gpg | gpg --dearmor >bazel-archive-keyring.gpg
sudo mv bazel-archive-keyring.gpg /usr/share/keyrings/
echo "deb [arch=amd64 signed-by=/usr/share/keyrings/bazel-archive-keyring.gpg] https://mirrors.tuna.tsinghua.edu.cn/bazel-apt stable jdk1.8" | sudo tee /etc/apt/sources.list.d/bazel.list
注意 https://bazel.build/bazel-release.pub.gpg 在gitee有镜像
如果报签名错误，重装证书：
sudo apt-get install --reinstall ca-certificates
#sudo apt update && sudo apt install bazel
sudo apt update && sudo apt install bazel-3.1.0
2. install build-essential and clang
sudo apt install build-essential clang-14 -y
3.
conda create -n build-tensorflow python=3.8
conda activate build-tensorflow
#pip清华镜像
python -m pip install --upgrade pip
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
pip install pip numpy wheel
pip install keras_preprocessing --no-deps
pip install packaging
cd /machine_learning; git clone git@gitcode.net:machine_learning/tensorflow.git
#git clone https://github.com/tensorflow/tensorflow.git
cd /machine_learning/tensorflow; git checkout v2.11.0
cd /machine_learning/tensorflow
./configure
"""
Do you wish to build TensorFlow with OpenCL SYCL support? [y/N]: N
No OpenCL SYCL support will be enabled for TensorFlow.
Do you wish to build TensorFlow with ROCm support? [y/N]: N
No ROCm support will be enabled for TensorFlow.
Do you wish to build TensorFlow with CUDA support? [y/N]: N
No CUDA support will be enabled for TensorFlow.
Do you wish to download a fresh release of clang? (Experimental) [y/N]: N
Clang will be downloaded and used to compile tensorflow.
Please specify optimization flags to use during compilation when bazel option "--config=opt" is specified [Default is -march=native -Wno-sign-compare]: -std=c++14 -Wno-error
Would you like to interactively configure ./WORKSPACE for Android builds? [y/N]: N
"""
bazel build --subcommands //tensorflow/tools/pip_package:build_pip_package
#bazel build //tensorflow/tools/pip_package:build_pip_package
#cat -n /home/z/.cache/bazel/_bazel_z/e167e5f1142e509062dbbcf47c207bce/external/com_google_absl/absl/synchronization/internal/graphcycles.cc
bazel debug build : --compilation_mode dbg
bazel release build : --compilation_mode fastbuild
#这里为还没试过的build命令
#bazel build --subcommands --jobs 4 //tensorflow/tools/pip_package:build_pip_package #"--jobs 4" 没试过 不确定是否对
#bazel build --subcommands --compilation_mode dbg //tensorflow/tools/pip_package:build_pip_package
#bazel --host_jvm_args "-DsocksProxyHost=10.11.1.115 -DsocksProxyPort=7890" build --subcommands --compilation_mode dbg //tensorflow/tools/pip_package:build_pip_package
find /machine_learning/tensorflow/ -type f -not -path '*/\.git/*' | xargs -I% grep -Hn io_bazel_rules_docker %
报错解决
external/ruy/ruy/block_map.cc:375:25: error: ‘numeric_limits’ is not a member of ‘std’
cd ~/.cache/bazel/_bazel_z/e167e5f1142e509062dbbcf47c207bce/execroot/org_tensorflow/
/usr/bin/gcc -U_FORTIFY_SOURCE -fstack-protector -Wall -Wunused-but-set-parameter -Wno-free-nonheap-object -fno-omit-frame-pointer -g0 -O2 '-D_FORTIFY_SOURCE=1' -DNDEBUG -ffunction-sections -fdata-sections -MD -MF bazel-out/host/bin/external/ruy/ruy/_objs/block_map/block_map.d '-frandom-seed=bazel-out/host/bin/external/ruy/ruy/_objs/block_map/block_map.o' -iquote external/ruy -iquote bazel-out/host/bin/external/ruy -g0 -g0 '-std=c++14' -Wall -Wextra -Wundef -O3 -fno-canonical-system-headers -Wno-builtin-macro-redefined '-D__DATE__="redacted"' '-D__TIMESTAMP__="redacted"' '-D__TIME__="redacted"' -c external/ruy/ruy/block_map.cc -o bazel-out/host/bin/external/ruy/ruy/_objs/block_map/block_map.o
external/ruy/ruy/block_map.cc: In function ‘void ruy::MakeBlockMap(int, int, int, int, int, int, int, int, const ruy::CpuCacheParams&, ruy::BlockMap*)’:
external/ruy/ruy/block_map.cc:375:25: error: ‘numeric_limits’ is not a member of ‘std’
375 | int best_score = std::numeric_limits<int>::min();
| ^~~~~~~~~~~~~~
external/ruy/ruy/block_map.cc:375:40: error: expected primary-expression before ‘int’
375 | int best_score = std::numeric_limits<int>::min();
解决 在对应的头文件 ~/.cache/bazel/_bazel_z/e167e5f1142e509062dbbcf47c207bce/execroot/org_tensorflow/external/ruy/ruy/block_map.h 中增加 “#include ” 即可
#include <limits>
ModuleNotFoundError: No module named ‘packaging’
报错：
from packaging import version as packaging_version # pylint: disable=g-bad-import-order
#报错: ModuleNotFoundError: No module named 'packaging'
解决:
conda activate build-tensorflow
pip install packaging
wsl2 ubuntu20.04 @win10 编译 tensorflow v2.11.0 ， 仅写差异部分
git checkout v2.11.0
sudo apt update && sudo apt install bazel-5.3.0
#家里的真机ubuntt22 用的貌似是 bazel-3.1.0 ?
dos2unix /mnt/d/machine_learning-home/tensorflow/configure
dos2unix /mnt/d/machine_learning-home/tensorflow/tensorflow/lite/experimental/acceleration/configuration/BUILD
cd /mnt/d/machine_learning-home/tensorflow/
./configure
bazel build --subcommands //tensorflow/tools/pip_package:build_pip_package
pytorch编译
- 由于 pytorch v0.3.0 没有 torch/torch.h, 因此 c++ example无法直接编译 （也找不到适合v0.3.0的c++例子), 因此放弃v0.3.0,
- 试图转向v0.4.0， 但v0.4.0 中的 子模块NervanaSystems/nervanagpu.git在github中已经不存在了，因此放弃v0.4.0
- 试图转向下一个版本v1.0.0， 但v1.0.0 在python setup.py xxx时 依然报错：tools/shared/init.py中import路径有问题，这个错误在v0.3.0中有类似错误（v0.3.0中的已经修复过一次了，麻烦的很），所以再次转向v1.3.1
pytorch v0.3.0编译
v0.3.0 低版本 好编译
cd /machine_learning-home/
git clone git@gitcode.net:machine_learning/pytorch-group/pytorch.git
#/machine_learning-home/pytorch/.git/
cd /machine_learning-home/pytorch/
#git checkout v0.3.0
git checkout -b branch-v0.3.0
export CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" # [anaconda root directory]
pip install numpy pyyaml mkl setuptools cmake cffi
export NO_CUDA=1
#export CMAKE_BUILD_TYPE=Debug#不奏效
export DEBUG=1 #奏效, DEBUG写在setup.py中
dos2unix pytorch/torch/lib/build_libs.sh
#clone子模块: git submodule update --init
python setup.py install
#编译完后测试一下:
cd /tmp/ & python -c "import torch"
#注意编译完后不要在 目录 /machine_learning-home/pytorch/ 下运行测试语句 "import torch",
#原因是 此时 "import torch" 正好错误的匹配到目录 /machine_learning-home/pytorch/torch/,
#正常情况下,"import torch" 应该是匹配到 目录 /home/z/miniconda3/envs/build-pytorch/lib/python3.8/site-packages/torch
参考:
其他
python setup.py --help-commands | grep install_lib
pytorch编译报错及解决
python setup.py install_lib 中的 py_compile 报 syntax error
错误描述
python setup.py install_lib
#py_compile byte-compiling 时 报错如下:
#syntax error： /home/z/miniconda3/envs/build-pytorch/lib/python3.8/site-packages/torch/autograd/variable.py
#syntax error： /home/z/miniconda3/envs/build-pytorch/lib/python3.8/site-packages/torch/autograd/_functions/tensor.py
#syntax error： /home/z/miniconda3/envs/build-pytorch/lib/python3.8/site-packages/torch/cuda/comm.py
解决办法: 把 async 全部替换为 async_
find /machine_learning-home/pytorch -name "*.py" -a -type f | xargs -I% sh -c "grep -Hn async % && sed -i 's/async/async_/g' % "
解决过程:
python2.7 -c "import py_compile; py_compile.compile('/home/z/miniconda3/envs/build-pytorch/lib/python3.8/site-packages/torch/autograd/_functions/tensor.py')"
#居然需要py2.7 就正常了
#async 不是 py2.x的关键字, 但 async 是 py3.x的关键字
pytorch 编译 放弃版本 v0.4.0，因为子模块NervanaSystems/nervanagpu.git已不存在
pytorch 编译 版本 v1.0.0 (同v1.3.1类似)
v1.0.0 在python setup.py xxx时 依然报错：tools/shared/init.py中import路径有问题，这个错误在v0.3.0中有类似错误（v0.3.0中的已经修复过一次了，麻烦的很），所以再次转向v1.3.1
pytorch 编译 版本 v1.3.1
pytorch v1.3.1 编译过程
cd /machine_learning-home/
git clone git@gitcode.net:machine_learning/pytorch-group/pytorch.git pytorch.v1.3.1
cd /machine_learning-home/pytorch.v1.3.1/
cd pytorch.v1.3.1/
git switch -c branch-v1.3.1 origin/branch-v1.3.1 #或用lazygit切换到分支 branch-v1.3.1
rm -fr .git/modules/third_party/* third_party/* .git/modules/android/libs/* android/libs/*
git submodule update --init --recursive
conda create -n build-pytorch.v1.3.1 python=3.8
conda activate build-pytorch.v1.3.1
#pip清华镜像
python -m pip install --upgrade pip
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
pip install astunparse numpy ninja pyyaml setuptools cmake cffi typing_extensions future six requests dataclasses
pip install mkl mkl-include
export USE_CUDA=0
export USE_ROCM=0
#export CMAKE_BUILD_TYPE=Debug#不奏效
export DEBUG=1 #奏效, DEBUG写在setup.py中
#dos2unix pytorch/torch/lib/build_libs.sh
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
#echo $CMAKE_PREFIX_PATH
#/home/z/miniconda3/envs/build-pytorch.v1.0.0
CMAKE_VERBOSE_MAKEFILE=True python setup.py develop
#直接用conda的base环境、或 直接用的系统自带python时 ，以下安装命令 可能会影响日常使用 不建议使用。 当用的conda专用环境、或在docker中，建议运行以下安装命令
python setup.py install
pytorch v1.3.1 example & frida
pip install frida==16.0.7 frida-tools==12.0.4
cd /machine_learning/pytorch.v1.3.1-py37/example/cpp_frontend_net/
sh build.sh
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/machine_learning/pytorch.v1.3.1-py37/torch/lib/ frida -f net --debug --pause
报错及解决
Device.cpp:218:1: error: cannot convert ‘std::nullptr_t’ to ‘Py_ssize_t’
1. 报错描述
2.1 报错解决办法1: 合入修复提交，
该修复提交 已经在分支origin/branch-v1.3.1-fix上了，因此 切到分支 origin/branch-v1.3.1-fix即可
git pull
git switch -c branch-v1.3.1-fix origin/branch-v1.3.1-fix
担忧的是 理论上应该想办法验证这个修改有没有问题，而解决办法2(退回py3.7)无此担忧.
2.2 报错解决办法2: 将python3.8退回到python3.7
conda create -n build-pytorch.v1.3.1-py37 python=3.7
conda activate build-pytorch.v1.3.1-py37
其他
web ssh client
pip install webssh
wssh --address='0.0.0.0' --port=8888
ip address #10.11.1.107
#browser access: http://10.11.1.107:8888/
git text tui (git tui)
,
,
mangle & demangle, dwarf : search _ZN(尝试找到规则?)
/mnt/d/gcc-home/gcc$ find . -type f -not -path '*/\.git/*' -a -type f | xargs -I% grep -Hn "_ZN" % > ../search_ZN.txt
search_ZN.txt
./libstdc++-v3/testsuite/abi/demangle/regression/cw-06.cc:46: verify_demangle("_ZNKSt17__normal_iteratorIPK6optionSt6vectorIS0_SaIS0_EEEmiERKS6_",
./libstdc++-v3/config/abi/pre/gnu.ver:601: _ZNSt7num_getI[cw]St19istreambuf_iteratorI[cw]St11char_traitsI[cw]EEE[CD][012]*;
demangle & mangle
demangle 即 解码, mangle 即 编码
demangle例子:
#demangle example:
#_Z23this_function_is_a_testi ----> this_function_is_a_test(int)
#_ZNSt22condition_variable_anyD2Ev ----> std::condition_variable_any::~condition_variable_any()
c++filt 也是个demangle工具
demangle : original
demange : python
demangle : javascript
demange : cpp
因篇幅问题不能全部显示，请点此查看更多更全内容