CUDA by Example - Compiler Error Compiler throws errors when trying to compile the examples of the b

_r00t · October 11, 2011, 8:30am

Good morning from Germany!

After setting up CUDA on Ubuntu 11.04 using these instructions (Samiux's Blog: HOWTO : nVidia CUDA Toolkit 4.0 on Ubuntu 11.04 Desktop) I can run the CUDA samples of the SDK, but when trying to compile the examples of the book ‘CUDA By Example’ I get the following error-code:

[indent]r00t@r00t-EG41MF-US2H:~/Documents/cuda_by_example/chapter03$ nvcc hello_world.cu

In file included from /usr/lib/nvidia-cuda-toolkit/include/cuda_runtime.h:59:0,

             from <command-line>:0:

/usr/include/host_config.h:82:2: warning: #warning – unsupported GNU version! gcc 4.5 and up are not supported!

In file included from /usr/lib/nvidia-cuda-toolkit/include/cuda_runtime.h:59:0,

             from <command-line>:0:

/usr/include/host_config.h:82:2: warning: #warning – unsupported GNU version! gcc 4.5 and up are not supported!

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `HandleError(cudaError, char const*, int)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x1e): undefined reference to `cudaGetErrorString’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `start_thread(void* ()(void), void*)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x198): undefined reference to `pthread_create’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `end_thread(unsigned long)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x1bb): undefined reference to `pthread_join’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `destroy_thread(unsigned long)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x1d5): undefined reference to `pthread_cancel’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `__cudaUnregisterBinaryUtil()':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x242): undefined reference to `__cudaUnregisterFatBinary’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `__device_stub__Z14float_to_colorPhPKf(unsigned char*, float const*)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x26a): undefined reference to `cudaSetupArgument’

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x289): undefined reference to `cudaSetupArgument’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `__device_stub__Z14float_to_colorP6uchar4PKf(uchar4*, float const*)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x2fe): undefined reference to `cudaSetupArgument’

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x31d): undefined reference to `cudaSetupArgument’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `__sti____cudaRegisterAll_46_tmpxft_00000954_00000000_4_hello_world_cpp1_ii_68d40520()':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x37e): undefined reference to `__cudaRegisterFatBinary’

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x3dc): undefined reference to `__cudaRegisterFunction’

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text+0x429): undefined reference to `__cudaRegisterFunction’

/tmp/tmpxft_00000954_00000000-13_hello_world.o: In function `cudaError cudaLaunch(char*)':

tmpxft_00000954_00000000-1_hello_world.cudafe1.cpp:(.text.Z10cudaLaunchIcE9cudaErrorPT[cudaError cudaLaunch(char*)]+0x14): undefined reference to `cudaLaunch’

collect2: ld returned 1 exit status[/indent]

To me it seems like I did not set up my environment the right way so gcc or nvcc does not now where to get the CUDA function code, but I am a Linux and CUDA beginner and do not have any idea how to solve this problem. Any help is appreciated! LG r00t

tera · October 11, 2011, 9:54am

You need to install gcc 4.4 (or 4.3). As the warning says, higher versions are not supported (yet).

_r00t · October 11, 2011, 11:23am

Thanks for your reply!

Did not think about it any further because when compiling the Nvidia-examples this warning was displayed, too.

But I though uninstalled it and then installed gcc-4.4. NVCC etc. had to be uninstalled, too, so I wanted to reinstall them but I get this message suggesting to reinstall gcc-4.5 etc.:

[indent]r00t@r00t-EG41MF-US2H:~$ sudo apt-get install nvidia-cuda-toolkit

Reading package lists… Done

Building dependency tree

Reading state information… Done

The following extra packages will be installed:

dkms g++ gcc nvidia-current nvidia-current-dev nvidia-opencl-dev

Suggested packages:

g+±multilib gcc-multilib autoconf automake1.9 libtool flex bison gcc-doc

The following NEW packages will be installed:

dkms g++ gcc nvidia-cuda-toolkit nvidia-current nvidia-current-dev

nvidia-opencl-dev

0 upgraded, 7 newly installed, 0 to remove and 28 not upgraded.

Need to get 0 B/76.5 MB of archives.

After this operation, 206 MB of additional disk space will be used.

Do you want to continue [Y/n]?[/indent]

Hope there is a way to prevent the installation of those packages, funnily enough that gcc-4.5 etc. are suggested for this package…

_r00t · October 11, 2011, 12:38pm

After following another tutorial (HDfpga: Install Cuda 4.0 on Ubuntu 11.04), I got nvcc using gcc-4.4, but same thing as before, just without the gcc-version warnings:
[indent]r00t@r00t-EG41MF-US2H:~/Documents/cuda_by_example/chapter03$ nvcc hello_world.cu
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function HandleError(cudaError, char const*, int)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x1e): undefined reference to cudaGetErrorString’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function start_thread(void* (*)(void*), void*)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x198): undefined reference to pthread_create’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function end_thread(unsigned long)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x1bb): undefined reference to pthread_join’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function destroy_thread(unsigned long)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x1d5): undefined reference to pthread_cancel’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function __cudaUnregisterBinaryUtil()': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x242): undefined reference to __cudaUnregisterFatBinary’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function __device_stub__Z14float_to_colorPhPKf(unsigned char*, float const*)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x26a): undefined reference to cudaSetupArgument’
tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x289): undefined reference to cudaSetupArgument' /tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function __device_stub__Z14float_to_colorP6uchar4PKf(uchar4*, float const*)‘:
tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x2fe): undefined reference to cudaSetupArgument' tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x31d): undefined reference to cudaSetupArgument’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function __sti____cudaRegisterAll_46_tmpxft_000060ce_00000000_4_hello_world_cpp1_ii_68d40520()': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x37f): undefined reference to __cudaRegisterFatBinary’
tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x3e0): undefined reference to __cudaRegisterFunction' tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text+0x430): undefined reference to __cudaRegisterFunction’
/tmp/tmpxft_000060ce_00000000-13_hello_world.o: In function cudaError cudaLaunch<char>(char*)': tmpxft_000060ce_00000000-1_hello_world.cudafe1.cpp:(.text._Z10cudaLaunchIcE9cudaErrorPT_[cudaError cudaLaunch<char>(char*)]+0x14): undefined reference to cudaLaunch’
collect2: ld returned 1 exit status[/indent]
What the hell could I be doing wrong?

_r00t · October 11, 2011, 1:09pm

Another question:
Does it matter, where I put “cuda_by_example”?

_r00t · October 13, 2011, 7:30am

After reading this : c++ - linking in library that contains reference to cuda kernel - Stack Overflow
I did the following:
//edited

call “nvcc -c hello_world.cu”
call “g+±4.4 -lcudart hello_world.o -o hello_world”
One gets an executable that one can run and it finally says “Hello World” GOD DAMMIT :D