We are trying to build a TensorFlow test case with debug flag:
bazel build -c dbg
//tensorflow/python/kernel_tests:sparse_matmul_op_test
However the build is failing with below error:
/usr/include/features.h:330:4: error: #warning _FORTIFY_SOURCE
requires compiling with optimization (-O) [-Werror=cpp]
warning _FORTIFY_SOURCE requires compiling with optimization (-O)
cc1: all warnings being treated as errors
Target //tensorflow/python/kernel_tests:sparse_matmul_op_test failed
to build
We have tried below options to resolve this:
built by exporting export CFLAGS and CXXFLAGS to "-Wno-error"
bazel build -c dbg --cxxopt="-Wno-all" --cxxopt="-Wno-error" //tensorflow/python/kernel_tests:sparse_matmul_op_test
Tried commenting compiler_flag from third_party/gpus/crosstool/CROSSTOOL.tpl
What is the correct way to suppress these warnings for the build to proceed?
We are using gcc v5.4.0.
I've had the same issue recently. It got solved by adding --copt=-O and -c opt to the build command.
Example:
bazel build --copt=-O -c dbg -c opt //tensorflow/python/kernel_tests:sparse_matmul_op_test
Solution by #BernardoGO does not work when CUDA build is enabled:
$ bazel build --copt=-O -c opt --config cuda -c dbg --strip=never //tensorflow/tools/pip_package:build_pip_package -s
/usr/include/c++/6/bits/stl_pair.h(327): error: calling a __host__ function("std::_Rb_tree_const_iterator< ::tensorflow::NcclManager::NcclStream *> ::_Rb_tree_const_iterator") from a __device__ function("std::pair< ::std::_Rb_tree_const_iterator< ::tensorflow::NcclManager::NcclStream *> , bool> ::pair< ::std::_Rb_tree_iterator< ::tensorflow::NcclManager::NcclStream *> &, bool &, (bool)1> ") is not allowed
/usr/include/c++/6/bits/stl_pair.h(327): error: identifier "std::_Rb_tree_const_iterator< ::tensorflow::NcclManager::NcclStream *> ::_Rb_tree_const_iterator" is undefined in device code
/usr/include/c++/6/bits/stl_algobase.h(1009): error: calling a __host__ function("__builtin_clzl") from a __device__ function("std::__lg") is not allowed
3 errors detected in the compilation of "/tmp/tmpxft_00007abb_00000000-6_nccl_manager.cpp1.ii".
Works only if --copt=-O is replaced with --copt=-O1, but -O1 is too much for comfortable debugging.
Related
I was trying to build a llvm-slicer from:
https://github.com/IAIK/ios-analysis-llvmslicer
and I follow the instructions:
cd llvm-slicer
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE="Release" -DLLVM_TARGETS_TO_BUILD="AArch64;X86" -DLLVM_ENABLE_EH=YES -DLLVM_ENABLE_RTTI=ON ..
make -j4 opt
make -j4 llvm-slicer
But when I execute the last command make -j4 llvm-slicer, I got an error:
[100%] Built target LLVMAnalysis
[100%] Linking CXX executable ../../bin/llvm-slicer
/usr/bin/ld: ../../lib/libLLVMSlicer.a(FunctionStaticSlicer.cpp.o): in function `llvm::Pass* llvm::callDefaultCtor<(anonymous namespace)::FunctionSlicer>()':
FunctionStaticSlicer.cpp:(.text._ZN4llvm15callDefaultCtorIN12_GLOBAL__N_114FunctionSlicerEEEPNS_4PassEv+0x1c): undefined reference to `vtable for (anonymous namespace)::FunctionSlicer'
collect2: error: ld returned 1 exit status
make[3]: *** [tools/llvm-slicer/CMakeFiles/llvm-slicer.dir/build.make:116: bin/llvm-slicer] Error 1
make[2]: *** [CMakeFiles/Makefile2:9647: tools/llvm-slicer/CMakeFiles/llvm-slicer.dir/all] Error 2
make[1]: *** [CMakeFiles/Makefile2:9654: tools/llvm-slicer/CMakeFiles/llvm-slicer.dir/rule] Error 2
make: *** [Makefile:2740: llvm-slicer] Error 2
I have no idea about how to fix it and I couldn't find any similar issues on Google, I hope someone can help me to figure it out, many thx.
There can be several scenarios for this issue, but in my opinion, here the scenario is for the given information that your default compiler (_ZN...something errors mostly because of compilers or linkers) is earlier version of gnu (gcc for c, g++ for c++) in your host pc (or maybe another compiler other than clang) and struggling about resolving a c++ feature anonymous namespace, as, a part of given error says:
/usr/bin/ld: ../../lib/libLLVMSlicer.a(FunctionStaticSlicer.cpp.o): in function `llvm::Pass* llvm::callDefaultCtor<(anonymous namespace)::FunctionSlicer>()':
FunctionStaticSlicer.cpp:(.text._ZN4llvm15callDefaultCtorIN12_GLOBAL__N_114FunctionSlicerEEEPNS_4PassEv+0x1c): undefined reference to `vtable for (anonymous namespace)::FunctionSlicer'
For this, there are several things you can do:
1-) You can change your compiler for building. In command window, you can export clang and clang++ as your compiler before cmake configuration. Here I assume that you installed new versions of clang compiler from anywhere, even github clone, I can give several examples here how to export and you can adapt one of them to your case easily:
export CC=clang
export CXX=clang++
another version:
export CC=clang-11
export CXX=clang++-11
another version:
export CC=/usr/bin/clang
export CXX=/usr/bin/clang++
another version:
export CC=/usr/bin/clang-12
export CXX=/usr/bin/clang++-12
another github clone version:
export CC=~/llvm/llvm-project/build/bin/clang
export CXX=~/llvm/llvm-project/build/bin/clang++
Even if this does not change your compiler, you can change your compiler while configuring with cmake by -DCMAKE_C_COMPILER={your-c-compiler} and -DCMAKE_CXX_COMPILER={your-c++-compiler} cmake flags. Try either with gcc or clang. Here is an example configuration in your case:
cmake \
-DCMAKE_C_COMPILER=clang \
-DCMAKE_CXX_COMPILER=clang++ \
-DCMAKE_BUILD_TYPE="Release" \
-DLLVM_TARGETS_TO_BUILD="AArch64;X86" \
-DLLVM_ENABLE_EH=YES \
-DLLVM_ENABLE_RTTI=ON ..
2-) Error can be because of c++ standard, you can add -DCMAKE_CXX_STANDARD=14 (Default C++ standard using by LLVM) flag as cmake configuration like:
cmake -DCMAKE_BUILD_TYPE="Release" -DLLVM_TARGETS_TO_BUILD="AArch64;X86" -DLLVM_ENABLE_EH=YES -DLLVM_ENABLE_RTTI=ON -DCMAKE_CXX_STANDARD=14 ..
3-) Even if didn't work above cases, you can remove anonymous namespace from FunctionStaticSlicer.cpp file (given in the error message) and try to build again. You can find these lines like this:
// some code here and do not delete
namespace { // delete this line
// as you can see there is no name of namespace that's why called 'anonymous'
// if it was it would be like 'namespace nmspcName {'
// some code here and do not delete
} // and delete this line, too
// some code here and do not delete
But this is really bad practice, even if it can solve your problem, I do not recommend.
Also you can try to do similar changes in CMakeLists.txt file or adding other cmake configuration flags, however in my opinion, you could solve your problem easily by 1. option which is changing your default compiler.
I'm triing to compile the superoptimizer on windows. (https://github.com/bonzini/superopt)
Unfortunatly my knowledge of make is very limited. I tried:
make CPU=-D386 superopt
Which gives me the error:
cc superopt.c -o superopt
process_begin: CreateProcess(NULL, cc superopt.c -o superopt, ...) failed.
make (e=2): The system cannot find the file specified.
make: *** [superopt] Fehler 2
After setting the compiler to gcc manually with:
make CPU=-D386 superopt CC=gcc
I get the following error:
gcc superopt.c -o superopt
In file included from superopt.c:27:0:
superopt.h:104:2: error: #error You have to choose target CPU type (--with-arch).
#error You have to choose target CPU type (--with-arch).
^
In file included from superopt.h:130:0,
from superopt.c:27:
longlong.h:1465:14: error: unknown type name 'UQItype'
extern const UQItype __clz_tab[];
^
superopt.c:32:21: fatal error: version.h: No such file or directory
compilation terminated.
make: *** [superopt] Fehler 1
It seems it does't properly selects i386.
Any hints would be greatly appreciated.
That is not the source code of
GNU superopt. It is the source code of someone's project
to patch GNU superopt, last updated 2008, and seemingly
abandoned a hard-hat area.
This is the source code of
GNU superopt. Extract the tarball and build as you have attempted
with:
make CC=gcc CPU=-DI386 superopt
Note: I386, not 386.
You will see warnings like:
warning: incompatible implicit declaration of built-in function 'foo'
because the 20 yearold C code does not diligently include the standard
headers that prototype the standard functions that it calls, but superopt will build successfully.
I got an issue when compiling glog by running 'make' after running './configure'
Then I got an error:
Undefined symbols for architecture x86_64:
"testing::internal::StrStreamToString(std::__1::basic_stringstream, std::__1::allocator >)", referenced from:
testing::internal::String testing::internal::StreamableToString(void const const&) in logging_unittest-logging_unittest.o
testing::internal::String testing::internal::StreamableToString(int const&) in logging_unittest-logging_unittest.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make: * [logging_unittest] Error 1
I am using glog-0.3.3 on Mac OS X.
SO how can i turn of testing while compiling glog?
In another context, i installed glog and gflags by using Macport, then i run a small program. It will generate a error :
"ERROR: unknown command line flag 'logtostderr'"
I believe that's the problem with linking to gflags. So how can i fix it. Thanks
GLog needs GFlags compiled in the "google" namespace instead of the now default "gflags" namespace.
In order to set this namespace you must compile and install gflags from source and set the GFLAGS_NAMESPACE variable to "google".
Here are the steps I followed in Kubuntu 14.04 and should be similar to what you should do in Mac OSX. These will place the GFlags source in /usr/local/src and install the library in the /usr/local/lib&include directories. The last command (ldconfig) registers the library in the system.
cd /usr/local/src/
cp /path/to/downloaded/gflags-2.1.1.tar.gz .
sudo tar xzf gflags-2.1.1.tar.gz
cd /tmp
mkdir buildgflags
cd buildgflags
cmake -DCMAKE_INSTALL_PREFIX=/usr/local -DBUILD_SHARED_LIBS=ON \
-DGFLAGS_NAMESPACE=google -G"Unix Makefiles" /usr/local/src/gflags-2.1.1/
make
sudo make install
sudo ldconfig
Alternatively you can apply the following patch in the GLog source (attached in the last reply):
https://code.google.com/p/google-glog/issues/detail?id=194
It basically uses the namespace of gflags after the includes on the GLogs unit test source files like so:
#ifdef HAVE_LIB_GFLAGS
#include <gflags/gflags.h>
using namespace gflags;
#endif
I'm trying to test this script from oracle to get active NFS clients on Ubuntu 10.04, but I can' get it to run.
To achieve that, I first installed dtrace following these instructions. This is what I've done exactly:
apt-get install bison flex zlib1g-dev libelf-dev binutils-dev libdw-dev libc6-dev-i386
wget ftp://crisp.dyndns-server.com/pub/release/website/dtrace/dtrace-20121009.tar.bz2
tar xfj dtrace-20121009.tar.bz2
cd dtrace-20121009
make all
make install
make load
However, I get this warning when compiling:
=================================================================
=== You need /usr/lib/libdwarf.a and /usr/lib/libbfd.a installed to build.
===
=== apt-get install binutils-dev
=== apt-get install libdw-dev
===
=== Without these, we will not build ctfconvert (needed for
=== SDT structure definitions).
=================================================================
cd cmd/instr ; make --no-print-directory
cd usdt/c ; make --no-print-directory
tools/mkdriver.pl all
Executing: /usr/src/dtrace/dtrace-20121009/tools/make-me
make -C /lib/modules/2.6.38-16-server/build M=/usr/src/dtrace/dtrace-20121009/build-2.6.38-16-server/driver
CC [M] /usr/src/dtrace/dtrace-20121009/build-2.6.38-16-server/driver/systrace.o
LD [M] /usr/src/dtrace/dtrace-20121009/build-2.6.38-16-server/driver/dtracedrv.o
Building modules, stage 2.
MODPOST 1 modules
LD [M] /usr/src/dtrace/dtrace-20121009/build-2.6.38-16-server/driver/dtracedrv.ko
tools/mkctf.sh
build/ctfconvert not available - so not building the linux.ctf file
NOTE: The build is complete, but build/ctfconvert is not available.
This means you will get run time errors from the io.d and sched.d files
due to undefined kernel structure definitions. Simply delete or rename
these files until a fix can be put in place to handle older
distros which do not have the required libdwarf dependencies.
(Typical error is references to undefined struct definitions such
as dtrace_cpu_t).
sync
I've installed libdw-dev and binutils-dev, but taking a look at the makefile, it seems it looks for libdwarf.so, and libdw on my system is named libdw.so.
To circunvent this, I create a symlink with ln -s /usr/lib/libdw.so /usr/lib/libdwarf.so. After doing so, compiling fails.
cd cmd/ctfconvert ; make --no-print-directory
gcc -g -I. -I../../ -I../../libctf -I../../common -I../../uts/common -I../../linux -I/usr/include/libdwarf -c dwarf.c
In file included from dwarf.c:94:
/usr/include/dwarf.h:56: error: expected identifier before numeric constant
/usr/include/dwarf.h:136: error: expected identifier before numeric constant
/usr/include/dwarf.h:321: error: expected identifier before numeric constant
/usr/include/dwarf.h:461: error: expected identifier before numeric constant
/usr/include/dwarf.h:517: error: expected identifier before numeric constant
make[3]: *** [../../build/ctfconvert.obj/dwarf.o] Error 1
make[2]: *** [all] Error 2
make[1]: *** [do_cmds] Error 2
tools/bug.sh
make: *** [all] Error 1
So, let's undo that. I remove the symlink, compile again, run make install and make load and hope everything is fine. And everything seems to be fine.
But, then I try to run the script mentioned above, and it fails:
# ./get_ngs_clients.d
dtrace: failed to compile script ./get_ngs_clients.d: Preprocessor not found
I have no clue on what's going on. I have gcc installed, just in case.
# dpkg -l | grep gcc
ii gcc 4:4.4.3-1ubuntu1 The GNU C compiler
ii gcc-4.4 4.4.3-4ubuntu5.1 The GNU C compiler
ii gcc-4.4-base 4.4.3-4ubuntu5.1 The GNU Compiler Collection (base package)
ii gcc-4.4-multilib 4.4.3-4ubuntu5.1 The GNU C compiler (multilib files)
ii gcc-multilib 4:4.4.3-1ubuntu1 The GNU C compiler (multilib files)
ii lib32gcc1 1:4.4.3-4ubuntu5.1 GCC support library (32 bit Version)
ii libgcc1 1:4.4.3-4ubuntu5.1 GCC support library
If you do not have libdwarf.a on your system, the ctfconvert tool will not build. (libdwarf.a and libdw.a are not the same).
If ctfconvert does not build, then any of your own, or the dtrace etc/*.d scripts may not load. (Dtrace force loads these scripts for you automatically, which is annoying). Any script which relies on structure definitions will then fail.
As of May 2013, I am looking at seeing what it takes to update to libdw.a since this seems to be the modern replacement for libdwarf.
(posted by the 'author' of DTrace/Linux).
Have you tried to add --enable-dtrace=false to /.configure?
Or maybe --with-dtrace=false?
That should do the trick I think...
I have just started learning OpenCL. I received the following error when I typed in make to compile the HelloCL program:
mkdir -p depends/x86_64
perl ../../../../../make/fastdep.pl -I. -I../../../../../include -I../../../../../samples/opencl/SDKUtil/include --obj-suffix='.o' --obj-prefix='build/debug/x86_64//' HelloCL.cpp > depends/x86_64/HelloCL.depend
mkdir -p build/debug/x86_64/
Building build/debug/x86_64//HelloCL.o
g++ -Wpointer-arith -Wfloat-equal -g3 -ffor-scope -I ../../../../../samples/opencl/SDKUtil/include -I "/opt/AMDAPP/include" -I ../../../../../include -o build/debug/x86_64//HelloCL.o -c HelloCL.cpp
In file included from HelloCL.cpp:106:0:
/opt/AMDAPP/include/CL/cl.hpp: In function ‘cl_int cl::UnloadCompiler()’:
/opt/AMDAPP/include/CL/cl.hpp:1826:12: error: ‘::clUnloadCompiler’ has not been declared
/opt/AMDAPP/include/CL/cl.hpp: In member function ‘cl_int cl::CommandQueue::enqueueMarker(cl::Event*) const’:
/opt/AMDAPP/include/CL/cl.hpp:4842:13: error: ‘::clEnqueueMarker’ has not been declared
/opt/AMDAPP/include/CL/cl.hpp: In member function ‘cl_int cl::CommandQueue::enqueueWaitForEvents(const std::vector<cl::Event>&) const’:
/opt/AMDAPP/include/CL/cl.hpp:4849:13: error: ‘::clEnqueueWaitForEvents’ has not been declared
/opt/AMDAPP/include/CL/cl.hpp: In member function ‘cl_int cl::CommandQueue::enqueueBarrier() const’:
/opt/AMDAPP/include/CL/cl.hpp:4985:13: error: ‘::clEnqueueBarrier’ has not been declared
make: *** [build/debug/x86_64//HelloCL.o] Error 1
Can someone help me understand the errors I am seeing, and suggest the corrections that need to be made?
Looks like cl.hpp you use is not adopted for the version of OpenCL your SDK supports. Try the cl.hpp from the Khronos site.
I advise downloading latest AMD APP SDK from : http://developer.amd.com/sdks/AMDAPPSDK/downloads/Pages/default.aspx
And try running the samples again, this SDK supports: OpenCL 1.2, so you shouldn't see any issues.