I am using eclipse to execute a cuda program. I have downloaded a CUDA PLUGIN for
eclipse. When I execute sample cuda program given by plugin its fine but when I try
to execute any other program I am getting error undefined reference to main...
make
Building target: Add_cuda
Invoking: NVCC Linker
nvcc -L/export/trainee3/dinesh/cuda5.0/lib64 -o "Add_cuda" ./mycuda.o -lcudart
/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/crt1.o: In function `_start':
(.text+0x20): undefined reference to `main'
collect2: ld returned 1 exit status
I found in stackoverflow that some times we get this problem because of system startup file
I used flag -nostartfiles but it is not working in my case..
I have included lib64 and include path in c++ build.
So any suggestion to over come this problem....
We've been over this already.
Use nsight eclipse edition instead.
If you have cuda 5.0 or 5.5 installed, just type nsight in a terminal session.
If you really want to use that Eclipse CUDA plugin (which is no longer supported, I don't believe), then start with the C++ sample project, which you agree now and back then would work. Then modify the source code in that project. Don't create your own project.
Install CUDA 5. It comes with Nsight Eclipse version. Very Elegant to use. NV Visual profiler is integrated with Nsight. Syntax highlighting and debug mode are very easy to use.
Related
I am building Swift compiler from source on CentOS 6, and am running into a library issue. After fighting with the build script for a while I have got where running ./utils/build-script eventually gives:
+ /home/src/cmake-3.4.1-Linux-x86_64/bin/cmake --build /home/src/swift/build/Ninja-DebugAssert/cmark-linux-x86_64 -- all
ninja: no work to do.
llvm: using standard linker
+ cd /home/src/swift/build/Ninja-DebugAssert/llvm-linux-x86_64
+ /home/src/cmake-3.4.1-Linux-x86_64/bin/cmake -G Ninja -DCMAKE_C_COMPILER:PATH=clang -DCMAKE_CXX_COMPILER:PATH=clang++ '-DCMAKE_C_FLAGS= ' '-DCMAKE_CXX_FLAGS= ' -DCMAKE_BUILD_TYPE:STRING=Debug -DLLVM_ENABLE_ASSERTIONS:BOOL=TRUE -DLLVM_TOOL_SWIFT_BUILD:BOOL=NO '-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64' -DLLVM_INCLUDE_TESTS:BOOL=TRUE -LLVM_INCLUDE_DOCS:BOOL=TRUE -DCMAKE_INSTALL_PREFIX:PATH=/usr -DINTERNAL_INSTALL_PREFIX=local /home/src/swift/llvm
CMake Error at cmake/modules/CheckAtomic.cmake:36 (message):
Host compiler appears to require libatomic, but cannot find it.
Call Stack (most recent call first):
cmake/config-ix.cmake:296 (include)
CMakeLists.txt:403 (include)
-- Configuring incomplete, errors occurred!
See also "/home/src/swift/build/Ninja-DebugAssert/llvm-linux-x86_64/CMakeFiles/CMakeOutput.log".
See also "/home/src/swift/build/Ninja-DebugAssert/llvm-linux-x86_64/CMakeFiles/CMakeError.log".
./utils/build-script: command terminated with a non-zero exit status 1, aborting
(gcc-4.8.2 was what I compiled llvm with)
libatomic is there:
$ locate libatomic
/opt/gcc-4.8.2/lib64/libatomic.a
/opt/gcc-4.8.2/lib64/libatomic.la
/opt/gcc-4.8.2/lib64/libatomic.so
/opt/gcc-4.8.2/lib64/libatomic.so.1
/opt/gcc-4.8.2/lib64/libatomic.so.1.0.0
I just don't know how to tell the build system where to look. I have tried the usual CMAKE_LIBRARY_PATH (exporting on the command line - I am not sure if cmake works like the way LD_LIBRARY_PATH, LIBRARY_PATH work) but it can't seem to find it.
I also don't have root on the machine.
I had not tried building from source on CentOS 6 until I saw this question, but I have been able to build Swift 2.2 on CentOS 7.1 and Ubuntu 14.04, with partial success. A few things to think about:
You will need numerous dependencies required to build Swift, and unless
they happen to be already on the system, you will need root access to
install them.
Use -R flag with the build-script to create a release build.
Building in DebugAssert (the default) will require a lot of memory. In my case even 14 GB was not sufficient. A release build
can be done with about 6 GB.
As for your specific problem, it is related to Clang's dependency on GCC-related packages for headers and libraries. See, for example, Fedora 21 with clang, without gcc.
Even if you installed GCC 4.8.2 and adjusted the path to use gcc and g++ from 4.8.2, Clang may still be looking in the old GCC directories for headers and libraries. CMake first tries to compile a C++ test file that includes the header atomic, which does not exist in the old GCC. So, it then tries to link a C test program that uses the library libatomic, which again doesn't exist in the old GCC. You can see this by looking at llvm/cmake/modules/CheckAtomic.cmake mentioned by usr1234567. CMakeError.log and CMakeOutput.log can also provide valuable insight. BTW, when I was building Swift on CentOS 7.1, I didn't run into this problem because GCC 4.8.2 was used by Clang for headers and libraries and the atomic header was found, so the C++ file got compiled. However, had the libatomic check been done, it would have failed, because libatomic.so in the repository-provided 4.8.2 has INPUT ( <name of some non-existent file> ), so trying to link with libatomic errors out.
I'm sure there are various ways of dealing with this issue, but what solved the problem for me was setting the following environment variables, please adjust to your specific setup:
export CPLUS_INCLUDE_PATH=/opt/gcc-4.8.2/include/c++/4.8.2:/opt/gcc-4.8.2/include/c++/4.8.2/x86_64-unknown-linux-gnu
export LIBRARY_PATH=/opt/gcc-4.8.2/lib64:/opt/gcc-4.8.2/lib/gcc/x86_64-unknown-linux-gnu/4.8.2
Also make sure that your 4.8.2 version of libstdc++.so is available to the dynamic linker at runtime. Since you don't have root, do
export LD_LIBRARY_PATH=/opt/gcc-4.8.2/lib64
If you had root, you could use ldconfig.
Before you start building Swift, you may want to try building, using Clang, a simple C program linking it with libatomic (the code doesn't actually have to use any symbols from the lib) and a simple C++ program that includes the <atomic> header. When compiling the C++ program, use the -std=c++11 compiler flag. If the C++ program compiles successfully, then it is not necessary for the libatomic linking test to be successful.
Interestingly, the CMakeOutput.log file still did not report finding GCC 4.8.2 as a candidate GCC installation, but the configuration/build worked well past the error.
Hopefully this helps. Please let us know if you run into something else.
CheckAtomic.cmake seems to be part of LLVM. I found a file at Github and it tries to find '__atomic_fetch_add_4' from libatomic
check_library_exists(atomic __atomic_fetch_add_4 "" HAVE_LIBATOMIC)
This fails for you. Check CMakeFiles/CMakeError.log to get more details why this test failed. Or try this line in a new project.
I am using Mac OSX (yosemite V 10.10.1) and running MATLAB 2014a on it.
I wanted to use SPAM library (sparse modeling software by J. Mairal) on MATLAB and for that I have to install XCode6.1 (that has gcc). First I type in command window mex -setup and result is shown below:
mex -setup
MEX configured to use 'Xcode with Clang' for C language compilation.
Warning: The MATLAB C and Fortran API has changed to support MATLAB
variables with more than 2^32-1 elements. In the near future
you will be required to update your code to utilize the
new API. You can find more information about this at:
http://www.mathworks.com/help/matlab/matlab_external/upgrading-mex-files-to-use-64-bit-api.html.
To choose a different language, select one from the following:
mex -setup C++
mex -setup FORTRAN
So after that I run the compile.m file in SPAM library and suddenly I saw an error that was:
add_flag =
-mmacosx-version-min=10.6
Warning: Directory already exists.
> In compile at 144
compilation of: -I./linalg/ -I./decomp/ -I./prox/ -I./dictLearn/ dictLearn/mex/mexArchetypalAnalysis.cpp
Building with 'Xcode Clang++'.
clang: warning: argument unused during compilation: '-fopenmp'
Error using mex
ld: warning: directory not found for option '-L/usr/lib/gcc/x86_64-linux-gnu/4.8/'
ld: library not found for -lgomp
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Error in compile (line 439)
mex(args{:});
I don't understand what to do. please help me!
It appears that the actual error is occurring at the point of linking with precompiled libraries. The 2 issues are as follows:
The compile code apparently included in the compile.m file looks like it is intended to compile with gcc (it is trying to include files installed by GCC, possibly even linux-specific ones, are you sure that it's an OSX-compatible toolbox?), and yet the error strongly suggests that you are in fact using clang to compile it - you will either need to change the compiler (easy) or rewrite compile.m (not so easy).
One of the libraries that the code needs to have installed in order to be properly linked hasn't been found. On OSX I think this file should be called libgomp.dylib (any mac afficionados want to confirm this?). If you have it on your computer, then it's not in a directory that clang is looking in. You can confirm the library is installed by running find / | grep libgomp.dylib from the terminal - if it is there, you can add it into the compiler argument in compile.m using the -I /DIRECTORY_HOLDING_LIBRARY syntax.
It is entirely possible that fixing 1. will also remediate 2. - I have never tried using SPAM
I have 64-bit Mac, OS X 10.8.5, and I have xcode installed. I can also verify gcc works from the command line. When I type mex -setup I get
The options files available for mex are:
1: /Applications/MATLAB_R2013a.app/bin/mexopts.sh :
Template Options file for building MEX-files
0: Exit with no changes
This is unhelpful. And when I type make, with all of the relevant libsvm files in my folder of choice, I get
make
xcodebuild: error: SDK "macosx10.7" cannot be located.
xcrun: error: unable to find utility "clang", not a developer tool or in PATH
mex: compile of ' "libsvmread.c"' failed.
If make.m fails, please check README about detailed instructions.
Is anyone able to help me with this?
The quickest thing is to edit the mexopts.sh file directly, using your favorite text editor (you may need to do this with "Administrator Privileges"). The file:
/Applications/MATLAB_R2013a.app/bin/mexopts.sh
defines a bunch of paths and flags for invoking the C/C++ compiler on your system. It tends not to keep up with revisions to the MacOS.
On my system, I had to make the following changes:
lines 258-260
CC='gcc'
SDKROOT='/Developer/SDKs/MacOSX10.6.sdk'
MACOSX_DEPLOYMENT_TARGET='10.6'
line 273
CXX=g++
There will be many references to "CC=" in the file; you're looking for the ones that follow the line
maci64)
But the correct values for your system depend on which gcc/g++ you have and where they are installed. As you can see, I have the MacOS 10.6 Developer tools installed under /Develop. You will need an install of the Developer tools (XCode) - see
How to use/install gcc on Mac OS X 10.8 / Xcode 4.4
In more recent versions of the XCode tools, the path might look more like:
/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk
But compiling MEX code with more recent versions of XCode might cause other problems - I had issues with char16_t, see:
MEX compile error: unknown type name 'char16_t'
I am using Eclipse Juno parallel version. I have installed
cuda plugin. But while building a sample cuda helloworld
program I am getting error -
make all
make: *** No rule to make target `exe_cuda', needed by `all'. Stop.
******Makefile snapshot**************
all: exe_cuda
# Tool invocations
#echo 'No tool found that can build the extension specified with the build
artifact name $#'
# Other Targets
clean:
-$(RM) $(OBJS)$(C_DEPS) exe_cuda
-#echo ' '
Any suggestion......
I agree that the toolchain does not appear to be set up properly for C projects.
My suggestions are:
Create a C++ project instead. That seems to work.
Use Nsight Eclipse Edition instead. It is installed automatically with the linux CUDA 5 package (just type nsight from a terminal window).
You could report the problem to fixstars corporation, the developer of that cuda plugin. It seems they have not updated it since 2011, so I don't know if it's actively maintained. But the help page is here which includes a link for a mailing list you could use.
I'm trying to use CUDA 5 RC1 on OS X Mountain Lion 10.8.1. When debugging from Nsight Eclipse Edition I get the error:
No source available for main()
I've verified that nvcc is set to use -g and -G to emit device and host debug symbols and that -O0 is set to disable optimizations.
Any ideas what the issue is or some other diagnostics I can carry out?
The issue, for me at least, turned out to be that the architecture needed to be set to x64.
To do this, right click on your project and hit Properties.
Then go to Build --> Settings --> Tool Settings.
Under NVCC Compiler --> Code Generation, set Architecture to x64.
Under NVCC Linker --> Miscellaneous set Architecture to x64.
Finally, clean the project and rebuild.