I've been trying to link to the functions in the cutil.h ofthe GPU Computing SDK released by NVIDIA.
At the moment, I am simply trying to compile this simple piece of code:
#include <iostream>
#include <cuda.h>
#include <cutil.h>
using namespace std;
int main(){
unsigned int time_total;
cutCreateTimer(&time_total);
return 0;
}
using the following command:
nvcc -I/home/sj755/NVIDIA_GPU_Computing_SDK/C/common/inc/ -L/home/sj755/NVIDIA_GPU_Computing_SDK/C/lib/libcutil_x86_64.a cutiltest.cu
Only to get the following error:
/tmp/tmpxft_000077cc_00000000-13_cutiltest.o: In function `main':
tmpxft_000077cc_00000000-1_cutiltest.cudafe1.cpp:(.text+0x10): undefined reference to
`cutCreateTimer'
collect2: ld returned 1 exit status
ld also can't find -lcutil if I were to add the flag.
There is a static library that I'm supposed to link to, but for some reason this never works out. Here's what I tried:
I've changed my .bashrc file so that LD_LIBRARY_PATH includes the path to the static library
##########< CULA >
export CULA_ROOT=/usr/local/cula
export CULA_INC_PATH=$CULA_ROOT/include
export CULA_BIN_PATH_32=$CULA_ROOT/bin
export CULA_BIN_PATH_64=$CULA_ROOT/bin64
export CULA_LIB_PATH_32=$CULA_ROOT/lib
export CULA_LIB_PATH_64=$CULA_ROOT/lib64
##########< CUDA >
export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=:/usr/local/cuda/lib64
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CULA_LIB_PATH_64
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/libnvvp/
export CUDA_SDK_ROOT_DIR=/home/sj755/NVIDIA_GPU_Computing_SDK/C
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_SDK_ROOT_DIR/lib
I've also tried renaming libcutil_x86_64.a to libcutil.a, still nothing.
Tried extracting the archive, creating a shared object file, and linking to it:
ar -x libcutil_x86_64.a
gcc -I /usr/include/GL/ -L /usr/include/GL/ -lglut -lGL -lGLU -lX11 -lXmu -lXi -lm -lpthread -shared *.cpp.o -o libcutil.so
nvcc -lcutil -I /home/sj755/NVIDIA_GPU_Computing_SDK/C/common/inc/ -L /home/sj755/NVIDIA_GPU_Computing_SDK/C/lib/libcutil.so cutiltest.cu
Only to get the following /usr/bin/ld: cannot find -lcutil
What step am I forgetting here?
Your compilation statement is incorrect. It should look something like this:
nvcc -I$SDKROOT/C/common/inc -L$SDKROOT/C/lib cutiltest.cc -lcutil_x86_64
where SDKROOT holds the root path to the SDK, which looks to be
/home/sj755/NVIDIA_GPU_Computing_SDK
in your case. The key things to note here are that the library must be passed by name as a -l option after the code and objects that require it. A concrete example on OS X using your code snippet:
$ cat cutiltest.cc
#include <iostream>
#include <cuda.h>
#include <cutil.h>
using namespace std;
int main(){
unsigned int time_total;
cutCreateTimer(&time_total);
return 0;
}
$ nvcc -I/Developer/GPU\ Computing/C/common/inc -L /Developer/GPU\ Computing/C/lib -o cutiltest cutiltest.cc -lcutil_i386
$ ls -l cutiltest
-rwxr-xr-x 1 talonmies talonmies 117548 May 25 07:57 cutiltest
But as a last remark, you really should rethink your choice of using the SDK cutils library at all. It is only intended for use with the SDK examples. It isn't part of CUDA, it has no documentation, it isn't guaranteed to work or not contain bugs, and isn't guaranteed to be consistent (or even present) from one SDK release to another.
This is not only your problem. Are you tried to link with cutil as with shared library? This solution was posted on NVIDIA dev. zone forum
Related
At page 14 of the book "Introduction to GCC" by Brian Gough, the author wants to show a linker error due to not supplying gcc with the libm library, where the code for the function sqrt resides:
$ gcc -Wall calc.c -o calc
/tmp/ccbR6Ojm.o: In function ‘main’:
/tmp/ccbR6Ojm.o(.text+0x19): undefined reference to ‘sqrt’
The file calc.c (where the functionsqrt is invoked) is this:
#include <math.h>
#include <stdio.h>
int main (void)
{
double x = sqrt (2.0);
printf ("The square root of 2.0 is %f\n", x);
return 0;
}
This book is from 2003.
On my current Ubuntu Linux 18, I can't reproduce the link error: it links and works, printing the correct result:
$ ./calc
1.414214
I found with ldd calc that the libm.so shared library is not invoked at runtime. Nor of course is the static library libm.a involved here.
So how does gcc deal with the function sqrt? I found that in this case it uses the sqrt GCC built-in function.
Its code gets inserted in the calc.o object file at compile time. So no "undefined reference" error.
First question: this the whole story or am I missing something?
Second question: why did this behavior regarding the built-in functions of GCC change so much between 2003 (when the book was written) and now? (Practically invalidating the whole example, it seems to me)
Third question: why does the author make his examples (e.g.$ gcc -Wall calc.c -lm -o cal) implying that the static library libc.a will be used, when in reality in Linux that syntax invokes the dynamic library libm.so? Is this specific to Linux and not to GNU GCC? What am I mising?
I think this is due to optimization of a constant value. Modern GCC can compute exact value of sqrt (2.0). If you force it not to use the built-ins with -fno-builtin, it will still fail to link. Also, if you change the code a little bit so that the argument to sqrt() is not literal, it will fail to link:
#include <math.h>
#include <stdio.h>
double t = 2.0;
int main (void)
{
double x = sqrt (t);
printf ("The square root of 2.0 is %f\n", x);
return 0;
}
This produces link error:
> gcc -o test test.c
/usr/bin/ld: /tmp/ccLjHnQx.o: in function `main':
test.c:(.text+0x11): undefined reference to `sqrt'
collect2: error: ld returned 1 exit status
Regarding your 3rd question, -lm does not imply static library, AFAIK.
I have compiled a c++ code using g++ -std=c++11 -o main main.cpp -pthread and it compiled fine however if I compile the same code using gcc -std=c++11 -o main main.cpp -pthread it does not compile and throws error. The program uses threading which properly taken care of using -pthread option while compiling. For the reference I am attaching the code below. Any help is highly appreciated.
#include <iostream>
#include <thread>
class foo
{
public:
void bar(int loop_num)
{
for (int i = 0; i < loop_num; ++i) {
std::cout << "Thread executing\n";
++n;
std::this_thread::sleep_for(std::chrono::milliseconds(10));
}
}
int n = 0;
};
int main()
{
int n = 0;
foo f;
std::thread t1(&foo::bar, &f, 5);
t1.join();
}
If you’ve written C++ code with GCC, you’ll know that you need to use the program g++, both for compilation and linking. For multi-module programs, this means every .cpp file gets compiled with g++, and then the entire program must be linked separately using g++. If you try to link the program using gcc, it will almost work, but you’ll get a lot of “undefined reference” errors, like this:
test.cpp:(.text+0x11): undefined reference to `std::cout'
The need to use g++ to link the entire program causes trouble when you have a very complicated build process you don’t have full control of. For instance, I’m trying to link C++ code with Mercury, and I have to use the Mercury linker, which in turn calls gcc.
So just a quick tip: If you are forced to use gcc to link the program, just add the library “stdc++”, as you would any other library, and it will work. That is, add the option “-lstdc++” to your GCC linker command line. For example:
g++ -c hello.cpp
gcc -lstdc++ -o hello hello.o
I assume the error you get looks something like this:
/tmp/ccUKAq0K.o: In function `main':
main.cpp:(.text+0x59): undefined reference to `std::thread::join()'
/tmp/ccUKAq0K.o: In function `__static_initialization_and_destruction_0(int, int)':
main.cpp:(.text+0xb6): undefined reference to `std::ios_base::Init::Init()'
main.cpp:(.text+0xcb): undefined reference to `std::ios_base::Init::~Init()'
/tmp/ccUKAq0K.o: In function `std::thread::~thread()':
main.cpp:(.text._ZNSt6threadD2Ev[_ZNSt6threadD5Ev]+0x1d): undefined reference to `std::terminate()'
(And so on.)
C++ programs which use the standard library (so most of them) need to be linked using g++, not gcc. Only the g++ compiler driver links in most of the standard library. The gcc compiler driver compiles C++ programs, just like g++, but when the linker is invoked, the program is treated as a C program, which usually leads to linker errors.
Try adding the -lrt flag after the pthread one.
The question here gives an example of how to use xgboost - a machine learning library written in C++. I want to run the example, and therefore install the library for C++. I added the /lib files to /usr/local/lib and /src files to /usr/local/src
I am able to compile this much part of the example :
#include <iostream>
#include <xgboost/c_api.h>
using namespace std;
int main(){
int cols=3,rows=5;
float train[rows][cols];
for (int i=0;i<rows;i++)
for (int j=0;j<cols;j++)
train[i][j] = (i+1) * (j+1);
float train_labels[rows];
for (int i=0;i<rows;i++)
train_labels[i] = 1+i*i*i;
DMatrixHandle h_train[1];
XGDMatrixCreateFromMat((float *) train, rows, cols, -1, &h_train[0]);
}
However, I am getting a linking error:
/tmp/ccuBacNh.o: In function `main':
TerrainPredict.cpp:(.text+0x278): undefined reference to `XGDMatrixCreateFromMat'
collect2: error: ld returned 1 exit status
How can I resolve this. I concluded that cpp files are missing, but I don't know where to put them. I want help regarding how to proceed with installation.
I also tried using a makefile:
LDLIBSOPTIONS=xgboost/lib/libxgboost.a xgboost/rabit/lib/librabit.a xgboost/dmlc-core/libdmlc.a -lpthread
CFLAGS=-I xgboost/include -I xgboost/rabit/include -I dmlc-core/include
all:
g++ main.cpp $(CFLAGS) $(LDLIBSOPTIONS)
I put Makefile, main.cpp and xgboost directory in the same folder.
In that case I am getting a longer linking error. Error is too long to include fully. Here it is.
I am trying to compile a source given a .so file libfoo.so. The only thing in this library is a function that just returns a number (yeah, I know, advanced stuff). The header file equivalent (I was provided with both, but am only supposed to use the .so) is named foo.h and the function is named int foo().
My source file is main.c:
#include <stdio.h>
#include "foo.h"
int main()
{
int x = foo();
printf("%d", x);
return 0;
}
Now, when trying to compile I have the following commands:
gcc -Wall -fPIC -c main.c -o main.o
gcc -Wall -fPIC main.o -o main -lfoo -L.
The first command fails to create the object file, outputting the following error:
fatal error: foo.h: No such file or directory
I am using Ubuntu 16.04.
I have also tried exporting the current location to LD_LIBRARY_PATH as I've seen suggested on a few other answers.
export LD_LBIRARY_PATH=$LD_LIBRARY_PATH:machine/Desktop/lib_test
You need to have the interface definition from the .h file and that file must be in the current directory or a directory on the include search path.
Note that on some systems filenames and paths are case dependent.
When I try to build the following program:
#include <stdio.h>
int main(void)
{
printf("hello world\n");
return 0;
}
On OS X 10.6.4, with the following flags:
gcc -static -o blah blah.c
It returns this:
ld: library not found for -lcrt0.o
collect2: ld returned 1 exit status
Has anyone else encountered this, or is it something that noone else has been affected with yet? Any fixes?
Thanks
This won’t work. From the man page for gcc:
This option will not work on Mac OS X unless all libraries (including libgcc.a) have also been compiled with -static. Since neither a static version of libSystem.dylib nor crt0.o are provided, this option is not useful to most people.
Per Nate's answer, a completely static application is apparently not possible - see also man ld:
-static Produces a mach-o file that does not use the dyld. Only used building the kernel.
The problem in linking with static libraries is that, if both a static and a dynamic version of a library are found in the same directory, the dynamic version will be taken in preference. Three ways of avoiding this are:
Do not attempt to find them via the -L and -l options; instead, specify the full paths, to the libraries you want to use, on the compiler or linker command line.
$ g++ -Wall -Werror -o hi /usr/local/lib/libboost_unit_test_framework.a hi.cpp
Create a separate directory, containing symbolic links to the static libraries, use the -L option to have this directory searched first, and use the -l option to specify the libraries you want to use.
$ g++ -Wall -Werror -L ./staticBoostLib -l boost_unit_test_framework -o hi hi.cpp
Instead of creating a link of the same name in a different directory, create a link of a different name in the same directory, and specify that name in a -l argument.
$ g++ -Wall -Werror -l boost_unit_test_framework_static -o hi hi.cpp
You may also try LLVM LLD linker - I did prebuilt version for my two major OSes - https://github.com/VerKnowSys/Sofin-llds
This one allows me to link for exmple: "Qemu" properly - which is impossible with ld preinstalled by Apple.
And last one is - to build GCC yourself with libstdc++ (don't).