Consider the situation when a C++ project is built and shipped within a Centos 7 virtual machine or container. Default gcc for Centos 7 is 4.8. In order to allow developers to use modern C++, the more recent version of gcc (for example, 6.3) is installed into Centos 7 which runs as a CI server. This provides -std=c++14 support.
[builder#f7279ae9f33f build (master %)]$ /usr/bin/c++ -v 2>&1 | grep version
gcc version 4.8.5 20150623 (Red Hat 4.8.5-28) (GCC)
[builder#f7279ae9f33f build (master %)]$ /opt/rh/devtoolset-6/root/usr/bin/c++ -v 2>&1 | grep version
gcc version 6.3.1 20170216 (Red Hat 6.3.1-3) (GCC)
export CXX=/opt/rh/devtoolset-6/root/usr/bin/c++
make all -j4
...
This is short example of compilation and linkage command:
[ 78%] Building CXX object CMakeFiles/ucsdos.dir/src/merge_operator_string.cpp.o
/opt/rh/devtoolset-6/root/usr/bin/c++ -Ducsdos_EXPORTS -I/home/builder/src/dos/libucsdos/./src -I/home/builder/src/dos/libucsdos/./include -I/home/builder/src/dos/libucsdos/build/schema/cpp -I/home/builder/src/dos/libucsdos/build/schema -isystem /usr/local/include -O2 -g -DNDEBUG -fPIC -frtti -pedantic -Wall -Wextra -Wcast-align -Wcast-qual -Wctor-dtor-privacy -Wdisabled-optimization -Wformat=2 -Winit-self -Wlogical-op -Wmissing-declarations -Wmissing-include-dirs -Wnoexcept -Wold-style-cast -Woverloaded-virtual -Wredundant-decls -Wshadow -Wsign-conversion -Wsign-promo -Wstrict-null-sentinel -Wstrict-overflow=5 -Wswitch-default -Wundef -Werror -Wno-unused -std=gnu++14 -o CMakeFiles/ucsdos.dir/src/merge_operator_string.cpp.o -c /home/builder/src/dos/libucsdos/src/merge_operator_string.cpp
[ 80%] Linking CXX shared library libucsdos.so
/usr/bin/cmake3 -E cmake_link_script CMakeFiles/ucsdos.dir/link.txt --verbose=1
/opt/rh/devtoolset-6/root/usr/bin/c++ -fPIC -O2 -g -DNDEBUG -shared -Wl,-soname,libucsdos.so.0 -o libucsdos.so.0.3.23 CMakeFiles/ucsdos.dir/src/c.cpp.o CMakeFiles/ucsdos.dir/src/crdt_2p_set.cpp.o CMakeFiles/ucsdos.dir/src/crdt_pn_counter.cpp.o CMakeFiles/ucsdos.dir/src/errors.cpp.o CMakeFiles/ucsdos.dir/src/merge_index_document.cpp.o CMakeFiles/ucsdos.dir/src/merge_index_segment.cpp.o CMakeFiles/ucsdos.dir/src/merge_operator_string.cpp.o -Wl,-rpath,/usr/local/lib: schema/libschema.a /usr/lib64/librocksdb.so /usr/lib64/libjemalloc.so /usr/local/lib/libgrpc++_reflection.so /usr/local/lib/libgrpc++.so /usr/local/lib/libgrpc.so -ldl -lgrpc++ /usr/lib64/libprotobuf.so -lpthread /usr/lib64/libprotobuf-lite.so
Anyway, the resulting artifacts appear to be linked with system default version of libstdc++:
[builder#f7279ae9f33f build (master %)]$ ldd libucsdos.so | grep libstdc++.so.6
libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f2a4a054000)
It's easy to find out that /lib64/libstdc++.so.6 version is 4.8.5:
[builder#f7279ae9f33f build (master %)]$ yum whatprovides "/lib64/libstdc++.so.6"
libstdc++-4.8.5-28.el7_5.1.x86_64 : GNU Standard C++ Library
Repo : #Updates
Matched from:
Filename : /lib64/libstdc++.so.6
Is this build environment configuration valid?
Anyway, the resulting artifacts appear to be linked with system default version of libstdc++:
Yes. The devtoolset-6-gcc-c++ package provides a custom version of GCC that uses a special linker script instead of a dynamic library for libstdc++.so. That means the binaries it produces do not depend on the newer libstdc++.so.6 and can be run on other CentOS machines that don't have devtoolset installed (i.e. they only have the older libstdc++ library from GCC 4.8).
Is this build environment configuration valid?
Yes. What you're seeing is completely normal, and how it's supposed to work.
The pieces of the newer C++ runtime from GCC 6.4.0 get statically linked into your binary, and at runtime it only depends on the old libstdc++.so which every CentOS system has installed.
That's the whole point of the devtoolset version of GCC.
Related
I'm compiling an old version of OGRE ON CMake 3.6 to see if I can compile or revive an old piece of software, and Boost is one of the dependencies for OGRE. It's essentially for multi-threading and without it, OGRE 1.9.0 builds completely fine on the macOS 12.0 SDK but fails macOS 10.8 SDK (specified in ZSH as -DCMAKE_OSX_DEPLOYMENT_TARGET but that's another issue for later).
The CMake command: cmake -DOGRE_CONFIG_DOUBLE=ON -DOGRE_BUILD_SAMPLES=OFF -DCMAKE_BUILD_TYPE=Release -DCMAKE_OSX_ARCHITECTURES="x86_64" -DCMAKE_PREFIX_PATH=/usr/local (yes, I'm compiling x86_64 on a arm64 machine running macOS 12.1 & XCode 13.1 if it matters)
When I build OGRE on the 12.0 SDK with Boost support (all is found here), it successfully checks for all libraries. As a note, Boost also has the atomic library but it's only needed as .hpp files and not an actual .a or whatever-lib file format CPP uses. Compiling OGRE goes all well until I see this dreaded message:
Copying OS X content lib/macosx/Ogre.framework/Versions/1.9.0/Resources/ogrelogo.png
28 warnings generated.
21 warnings generated.
[ 44%] Linking CXX shared library ../lib/macosx/Ogre.framework/Ogre
> ld: library not found for -latomic <
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Key questions: What is this atomic library? Is it only on an older SDK? Is it the hpp files? Here is the linker command shortened to where I think is the most important parts
cd /Users/meh/Downloads/project/ogre-1-9-0/OgreMain && /Applications/CMake.app/Contents/bin/cmake -E cmake_link_script CMakeFiles/OgreMain.dir/link.txt --verbose=1
/usr/bin/clang++ -msse -stdlib=libc++ -std=c++11 -Wall -Winit-self -Wno-overloaded-virtual -Wcast-qual -Wwrite-strings -Wextra -Wno-unused-parameter -Wshadow -Wno-missing-field-initializers -Wno-long-long -O3 -DNDEBUG -arch x86_64 -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk -dynamiclib -Wl,-headerpad_max_install_names -framework AGL -framework IOKit -framework Cocoa -framework Carbon -framework OpenGL -framework CoreVideo -compatibility_version 1.9.0 -current_version 1.9.0 -o ../lib/macosx/Ogre.framework/Versions/1.9.0/Ogre -install_name #executable_path/../Frameworks/Ogre.framework/Versions/1.9.0/Ogre (ALL THOSE OGRE .O FILES) -L/usr/local/lib /usr/local/lib/libboost_thread.a /usr/local/lib/libboost_date_time.a /usr/local/lib/libboost_system.a /usr/local/lib/libboost_chrono.a /usr/local/lib/Release/libFreeImage.a /usr/local/lib/Release/libzziplib.a /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX12.0.sdk/usr/lib/libz.tbd -latomic
I had to play around with the CMake file to not error listed as:
framework AGL
-stdlib=libc++ -std=c++11
I've done research and atomic shows up in C++11 so that shouldn't be a problem? There's some include files under the SDK but definitely none under /usr/local/lib... (I will have to add -L/usr/local/include as a prefix to -latomic later)
I'm trying to compile C code on a Jetson Nano and I get this error during compiling. I tried removing any occurrence of 'm -64' but it seems like its added automatically. This is the cmd where it fails: /usr/bin/gcc-7 -Wall -Wextra -Wconversion -pedantic -Wshadow -m64 -Wfatal-errors -O0 -g -o CMakeFiles/dir/testCCompiler.c.o -c /home/user/dir/CMakeFiles/CMakeTmp/testCCompiler.c
uname -a: Linux jetson-nano 4.9.140-tegra aarch64 aarch64 aarch64 GNU/Linux
gcc-7 -v: Using built-in specs.
COLLECT_GCC=gcc-7
COLLECT_LTO_WRAPPER=/usr/lib/gcc/aarch64-linux-gnu/7/lto-wrapper
Target: aarch64-linux-gnu
gcc version 7.4.0 (Ubuntu/Linaro 7.4.0-1ubuntu1~18.04.1)
CMAKE_C_COMPILER = gcc-7
CMAKE_CXX_COMPILER = g++-7
CXX_COMPILE_FLAGS = "-Wall -Werror -Wextra -Wnon-virtual-dtor -Wconversion -Wold-style-cast -pedantic -Wshadow"
C_COMPILE_FLAGS = "-Wall -Wextra -Wconversion -pedantic -Wshadow"
gcc-7: error: unrecognized command line option ‘-m64’
error: unrecognized command line option ‘-m64’
I believe you are looking for -march=armv8-a (and friends), and not -m64. The GCC arm64 options are available at 3.18.1 AArch64 Options in the manual.
Aarch64 includes ASIMD in the base specification, so there are no extra gyrations needed for it. ASIMD is "Advanced SIMD instructions", and it is what ARM calls NEON on the Aarch32 and Aarch64 architectures.
If you want to enable extensions, like CRC or Crypto, then the option would look like -march=armv8.1-a+crc or -march=armv8.1-a+crypto or -march=armv8.1-a+crc+crypto.
The equivalent x86 options would be the following. Obviously, the ARM port of GCC does not use the same model as x86. It is confusing for new users (or it was confusing for me).
-march=armv8-a → -msse2
-march=armv8.1-a+crc → -msse2 -msse4.1
-march=armv8.1-a+crypto → -msse2 -mpclmul -maes
-march=armv8.1-a+crc+crypto → -msse2 -msse4.1 -mpclmul -maes
ARM instruction set includes SHA in crypto, so the x86 options should probably include -msha. The problem is, x86 SHA did not arrive until about 8 years after carryless multiplies and AES.
Also, GCC ARM compilers usually don't understand -march=native. On older GCC compilers, the compiler will just crash. On mid-ranged GCC it is simply ignored. I believe the latest GCC compilers honor it.
This error often happens when cross-compiling with Rust/Cargo, because Cargo isn't smart enough to find cross-build tools by itself.
You need to set appropriate env vars. In the example replace x86_64_unknown_linux_gnu with your target, and paths to your cross-build paths (the example is for Debian). Watch out the env vars are case-sensitive and inconsistently named!
# for the cc crate
export HOST_CC=gcc
export CC_x86_64_unknown_linux_gnu=/usr/bin/x86_64-linux-gnu-gcc
# for Cargo
export CARGO_TARGET_X86_64-UNKNOWN-LINUX-GNU_LINKER=/usr/bin/x86_64-linux-gnu-gcc
I'm writing a lot of cross platform C++, and am trying to unify my build process between platforms (primarily targeting Windows and Mac at the moment, potentially mobile in the future.) In addition to our normal CI builds, we are built from source as part of an extremely large C++ project's build and consequently have quite a complex toolchain.
Right now, my code compiles cleanly on Windows using CMake to generate a Visual Studio project and then using clang-cl to compile. Similarly, we are using CMake to generate a ninja project and then clang to compile.
We are a Windows shop, and I would like to leverage WSL to set up cross compilation, which should allow us to use the same ninja project to target both supported platforms and modify in the future.
I am able to successfully compile a simple hello world exe program using clang from my wsl-hosted bash terminal. In order to do this, I had to provide a large amount of options to clang as well as move a significant amount of Win10 SDK libraries to my build environment for clang to build and link against.
Here are the two commands I run to successfully produce helloworld.exe:
Compiling:
clang -target i686-pc-win32 -fms-compatibility-version=19 -fms-extensions -fdelayed-template-parsing -fexceptions -mthread-model posix -fno-threadsafe-statics -Wno-msvc-not-found -DWIN32 -D_WIN32 -D_MT -D_DLL -Xclang -disable-llvm-verifier -Xclang '--dependent-lib=msvcrt' -Xclang '--dependent-lib=ucrt' -Xclang '--dependent-lib=oldnames' -Xclang '--dependent-lib=vcruntime' -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE -U__GNUC__ -U__gnu_linux__ -U__GNUC_MINOR__ -U__GNUC_PATCHLEVEL__ -U__GNUC_STDC_INLINE__ -I/mnt/d/source/windeps/LLVM/include -I/mnt/d/source/windeps/MSVC/14.22.27905/include -I/mnt/d/source/windeps/ucrt -I/mnt/d/source/windeps/shared -I/mnt/d/source/windeps/winrt -c hello.cc -o hello.o
Linking:
clang -fuse-ld=lld-link.exe -target i686-pc-win32 -Wl,-machine:x86 -fmsc-version=1923 -o hello.exe hello.o -L/mnt/d/source/windeps/MSVC/14.22.27905/lib/x86/msvcrt.lib -nostdlib -lmsvcrt -Wno-msvc-not-found
Naturally, I have attempted to express this first command via my CMake toolchain:
add_compile_options(
-W
-Wall
-std=c++17
-stdlib=libc++
-fcoroutines-ts
-fms-extensions
-fdelayed-template-parsing
-fexceptions
-fdeclspec
-mthread-model posix
-fno-threadsafe-statics
-Wno-msvc-not-found
-DWIN32
-D_WIN32
-D_MT
-D_DLL
-Xclang
-disable-llvm-verifier
# These are commented out currently, but I have linked them to the proper CMakeList
# -Xclang '--dependent-lib=msvcrt'
# -Xclang '--dependent-lib=ucrt'
# -Xclang '--dependent-lib=oldnames'
# -Xclang '--dependent-lib=vcruntime'
-D_CRT_SECURE_NO_WARNINGS
-D_CRT_NONSTDC_NO_DEPRECATE
-U__GNUC__
-U__gnu_linux__
-U__GNUC_MINOR__
-U__GNUC_PATCHLEVEL__
-U__GNUC_STDC_INLINE__
-I/mnt/d/source/windeps/LLVM/include
-I/mnt/d/source/windeps/MSVC/14.22.27905/include
-I/mnt/d/source/windeps/um
-I/mnt/d/source/windeps/ucrt
-I/mnt/d/source/windeps/shared
-I/mnt/d/source/windeps/winrt
)
This is throwing an error from within the ucrt library when one of our files (event_logger.cpp) does #include <array>.
In file included from src/client/services/src/event_logger.cpp:10:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/array:6:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/algorithm:6:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/xmemory:8:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/limits:8:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/cwchar:8:
In file included from /mnt/d/source/windeps/MSVC/14.22.27905/include/cstdio:8:
In file included from /mnt/d/source/windeps/ucrt/stdio.h:13:
/mnt/d/source/windeps/ucrt/corecrt_wstdio.h:581:9: error: use of undeclared identifier '__crt_va_end'
__crt_va_end(_ArgList);
^
/mnt/d/source/windeps/ucrt/corecrt_wstdio.h:597:9: error: use of undeclared identifier '__crt_va_start_a'
__crt_va_start(_ArgList, _Locale);
^
/mnt/d/source/windeps/MSVC/14.22.27905/include/vadefs.h:156:99: note: expanded from macro '__crt_va_start'
#define __crt_va_start(ap, x) ((void)(__vcrt_assert_va_start_is_not_reference<decltype(x)>(), __crt_va_start_a(ap, x)))
Because of this behavior, I'm suspicious that perhaps we're not using libc++ as indicated by the -stdlib flag. I'm also not sure how to correct this, as this seems to be kind of a newer process and there's not a ton of documentation in the wild yet. Any advice is appreciated.
Ubuntu 17.10
GCC Version: 5.4
Bazel Version: 0.9.0
TensorFlow: r1.5
CUDA 8.0 / cuDNN 6 / GTX 1080 Ti
How do I make Bazel use gcc for building TensorFlow from source?
While building, its running into compiler errors like:
error: 'errno' was not declared in this scope
while (nanosleep(&ts, &ts) != 0 && errno == EINTR) {}
Setting --verbose_failures flag, it shows that its not using /usr/bin/gcc-5 or /usr/bin/gcc for compiling
external/local_config_cuda/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc -U_FORTIFY_SOURCE '-D_FORTIFY_SOURCE=1' -fstack-protector -fPIE -Wall -Wunused-but-set-parameter -Wno-free-nonheap-object -fno-omit-frame-pointer -g0 -O2 -DNDEBUG -ffunction-sections -fdata-sections -g0 -DGEMMLOWP_ALLOW_SLOW_SCALAR_FALLBACK '-march=native' '-std=c++11' -g0 -MD -MF
Searching online, I found that the path to gcc and CC, CXX variables must be set in tools/cpp/CROSSTOOL. But where exactly is tools/cpp/CROSSTOOL?? How do I force bazel to use gcc-5?
I know next to nothing about cuda and tensorflow, but tensorflow doesn't use the same C++ toolchain that bazel autoconfigures when compiling with cuda, so the CC env variable trick won't work. The file crosstool_wrapper_driver_is_not_gcc is just a shell wrapper that could in theory still call your gcc (or it will be cuda). I'd run bazel with --subcommands to see the complete invocation of the failing action, then reproduce without bazel, and then go from there.
I have Eclipse CDT Oxygen on Ubuntu 16.04, with the LLVM support installed. LLVM 5.0.0 is installed, and I want to experiment with libc++. The problem is building a project with libc++, but without libstdc++.
The first step is building a simple "Hello World" executable using c++ and libc++. For the project properties -> C/C++ Build -> Settings I have the following:
LLVM Clang (all options): -O0 -emit-llvm -g3 -Wall -c -fmessage-length=0 -std=c11
LLVM Clang++ (all options): -O0 -emit-llvm -g3 -Wall -c -fmessage-length=0 -stdlib=libc++ -std=c++17
LLVM Clang linker (all options): -v -L/usr/local/lib -L/usr/lib -L/usr/lib/gcc/x86_64-linux-gnu/5/ -nodefaultlibs
Note that "-L/usr/lib/gcc/x86_64-linux-gnu/5/" is added automatically and I cannot get rid of it.
The linked libraries are (in order): c++, c++abi, m, c, gcc_s, gcc, stdc++
The stdc++ lib is added automatically and seems impossible to strip out. This results in an executable that contains both libc++ and libstdc++ which we don't want. Does anyone know of a way to build a project without libstdc++ but with libc++ ?