COMPUTE_API_BUFFER_TYPE()

[Eric Johnson]

I came across this problem after running 'make'. It seems that a series of problem starts with COMPUTE_API_BUFFER_TYPE() not being treated as an operator by the compiler nvcc. The system is Ubuntu and there was no problem with installing dependencies, installing nvidia_driver_ubuntu and running configure-cuda. The error info is attached in the end. I suppose I'm making some sinple mistakes, and just wonder if anyone else has run into the same problem. Thank you!

[Eric Johnson]

I just noticed the pictures attached may not show on this page. Please 'right click' and open in another tab. I'm sorry about this.

[sandezi]

In file included from cocolib++/compute_api/compute_engine.h:25:0,
from cocolib++/compute_api/compute_grid.h:25,
from cocolib++/compute_api/compute_array.h:25,
from cocolib++/compute_api/compute_array.cpp:24:
cocolib++/compute_api/compute_buffer.h:58:14: error: expected type-specifier before ‘COMPUTE_API_BUFFER_TYPE’
operator COMPUTE_API_BUFFER_TYPE();
^
cocolib++/compute_api/compute_buffer.h:59:20: error: ISO C++ forbids declaration of ‘type name’ with no type [-fpermissive]
operator const COMPUTE_API_BUFFER_TYPE() const;
^
cocolib++/compute_api/compute_buffer.h:59:5: error: ‘operator const int’ does not name a type
operator const COMPUTE_API_BUFFER_TYPE() const;
^
cocolib++/compute_api/compute_buffer.h:62:5: error: ‘COMPUTE_API_BUFFER_TYPE’ does not name a type
COMPUTE_API_BUFFER_TYPE _data;
^

I met the same problem, and have no idea about this. Looking forward to your reply. Thanks!

[Simão Marto]

That issue is probably caused because the driver isn't being detected. Replace the contents of ./config_tests/test_compute_capability.cu by the following:

#include <iostream>
using namespace std;

#define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t code, const char *file, int line, bool abort=true)
{
   if (code != cudaSuccess) 
   {
      cout << "GPUassert: " << cudaGetErrorString(code) << " " << file << " " << line << endl;
      if (abort) exit(0);
   }
}

int main()
{
  cudaDeviceProp deviceProp;
  gpuErrchk(cudaGetDeviceProperties(&deviceProp,0))
  cout << "  detected compute capability " << deviceProp.major << "." << deviceProp.minor << endl;
  return deviceProp.major * 10 + deviceProp.minor;
}

And then run configure-cuda.sh. In my case it said

GPUassert: no CUDA-capable device is detected ./config_tests/test_compute_capability.cu 39

I solved this by installing the nvidia driver manually. For more information check this answer

Ignore everything I wrote bellow this line

---

Hello. I had the exact same problem, and hacked together a solution. I went in to file compute_buffer.h, and added afterline 28 the following:

typedef float* COMPUTE_API_BUFFER_TYPE;

This allowed the make process to continue, but eventually I had an error (after a very long list of warnigns).:

make[1]: Entering directory `/home/simao/Documents/cocolib-release-6/tools/parse_lfa_results'
g++ -c -m64 -pipe -I../../cocolib -pipe -g -Wall -g -D_REENTRANT -Wall -W -DQT_WEBKIT -DQT_GUI_LIB -DQT_CORE_LIB -DQT_SHARED -I/usr/share/qt4/mkspecs/linux-g++-64 -I. -I/usr/include/qt4/QtCore -I/usr/include/qt4/QtGui -I/usr/include/qt4 -I. -o obj/main.o main.cpp
main.cpp: In function ‘int main(int, char**)’:
main.cpp:247:38: error: too many arguments to function ‘herr_t H5Eset_auto1(H5E_auto1_t, void*)’
/usr/include/H5Epublic.h:216:15: note: declared here
make[1]: *** [obj/main.o] Error 1
make[1]: Leaving directory `/home/simao/Documents/cocolib-release-6/tools/parse_lfa_results'
make: *** [sub-tools-parse_lfa_results-make_default-ordered] Error 2

I then tried to solve this by removing (without really knowing what I'm doing) the third parameter (a NULL) in the function call at main.cpp:247:38. I got a new and different error:

cd examples/ && make -f Makefile 
make[1]: Entering directory `/home/simao/Documents/cocolib-release-6/examples'
g++ -c -m64 -pipe -I../cocolib -pipe -g -Wall -g -D_REENTRANT -Wall -W -DQT_WEBKIT -DQT_GUI_LIB -DQT_CORE_LIB -DQT_SHARED -I/usr/share/qt4/mkspecs/linux-g++-64 -I. -I/usr/include/qt4/QtCore -I/usr/include/qt4/QtGui -I/usr/include/qt4 -I. -o obj/main.o main.cpp
g++ -c -m64 -pipe -I../cocolib -pipe -g -Wall -g -D_REENTRANT -Wall -W -DQT_WEBKIT -DQT_GUI_LIB -DQT_CORE_LIB -DQT_SHARED -I/usr/share/qt4/mkspecs/linux-g++-64 -I. -I/usr/include/qt4/QtCore -I/usr/include/qt4/QtGui -I/usr/include/qt4 -I. -o obj/config.o config.cpp
config.cpp: In function ‘bool coco::init_config_data(coco::config_data*)’:
config.cpp:622:15: error: ‘rand’ was not declared in this scope
make[1]: *** [obj/config.o] Error 1
make[1]: Leaving directory `/home/simao/Documents/cocolib-release-6/examples'
make: *** [sub-examples-make_default-ordered] Error 2

To try and solve this, I added after line 27 of config.h the following:
#include <stdlib.h>

I also deleted all .o files just in case.
Unfortunately, this didn't put an end to my nightmare, and more exciting errors appeared. This time I got more than three thousand lines of errors, all about undefined references. Here's a small sample:

../cocolib/libcocolib.a(cuda_interface_cuda.obj): In function `__sti____cudaRegisterAll_49_tmpxft_0000513b_00000000_4_cuda_interface_cpp1_ii_1ebb158b':
/tmp/tmpxft_0000513b_00000000-1_cuda_interface.cudafe1.stub.c:4: undefined reference to `__cudaRegisterFatBinary'
../cocolib/libcocolib.a(cuda_interface_cuda.obj): In function `coco::cuda_device_name(unsigned long)':
/home/simao/Documents/cocolib-release-6/cocolib/cuda/cuda_interface.cu:45: undefined reference to `cudaGetDeviceProperties'
../cocolib/libcocolib.a(cuda_interface_cuda.obj): In function `coco::cuda_device_count()':
/home/simao/Documents/cocolib-release-6/cocolib/cuda/cuda_interface.cu:36: undefined reference to `cudaGetDeviceCount'
../cocolib/libcocolib.a(cuda_interface_cuda.obj): In function `__cudaUnregisterBinaryUtil':
/usr/include/crt/host_runtime.h:172: undefined reference to `__cudaUnregisterFatBinary'
../cocolib/libcocolib.a(cuda_interface_cuda.obj): In function `coco::cuda_set_device(unsigned long)':
/home/simao/Documents/cocolib-release-6/cocolib/cuda/cuda_interface.cu:53: undefined reference to `cudaSetDevice'
../cocolib/libcocolib.a(tv_l2_cuda.obj): In function `__sti____cudaRegisterAll_40_tmpxft_00005860_00000000_4_tv_l2_cpp1_ii_c1e4d73a':
/tmp/tmpxft_00005860_00000000-1_tv_l2.cudafe1.stub.c:4: undefined reference to `__cudaRegisterFatBinary'
../cocolib/libcocolib.a(tv_l2_cuda.obj): In function `coco::tv_l2_energy(coco::tv_l2_data*)':
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:192: undefined reference to `cudaMemcpy'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:193: undefined reference to `cudaMemcpy'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:194: undefined reference to `cudaThreadSynchronize'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:192: undefined reference to `cudaGetLastError'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:192: undefined reference to `cudaGetErrorString'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:194: undefined reference to `cudaGetLastError'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:194: undefined reference to `cudaGetErrorString'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:193: undefined reference to `cudaGetLastError'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:193: undefined reference to `cudaGetErrorString'
../cocolib/libcocolib.a(tv_l2_cuda.obj): In function `coco::tv_l2_get_dual_xi(coco::tv_l2_data*, std::vector<coco::gsl_matrix*, std::allocator<coco::gsl_matrix*> >&)':
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:178: undefined reference to `cudaThreadSynchronize'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:178: undefined reference to `cudaGetLastError'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:178: undefined reference to `cudaGetErrorString'
../cocolib/libcocolib.a(tv_l2_cuda.obj): In function `coco::tv_l2_get_solution(coco::tv_l2_data*, coco::gsl_matrix*)':
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:160: undefined reference to `cudaThreadSynchronize'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:160: undefined reference to `cudaGetLastError'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:160: undefined reference to `cudaGetErrorString'
../cocolib/libcocolib.a(tv_l2_cuda.obj): In function `coco::tv_l2_initialize(coco::tv_l2_data*, coco::gsl_matrix*)':
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:142: undefined reference to `cudaMemset'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:143: undefined reference to `cudaMemset'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:144: undefined reference to `cudaMemset'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:145: undefined reference to `cudaMemset'
/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:146: undefined reference to `cudaMemset'
../cocolib/libcocolib.a(tv_l2_cuda.obj):/home/simao/Documents/cocolib-release-6/cocolib/tv/tv_l2.cu:147: more undefined references to `cudaMemset' follow

Now I have no idea what to do. Maybe all I've done so far only made it worse, who knows.