thrust device vector

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

why does declaring a device vector of a user-defined type (like a struct) cause a compile-time error?

This code doesn’t work because of it:

struct KcFeature

{

  float x;

  float y;

  float val;

};

thrust::device_vector<KcFeature> dv;

The error returned:

obj/release/kltcuda.cpp.o: In function `thrust::detail::vector_base<KcFeature, thrust::device_allocator >::resize(unsigned int, KcFeature)':

kltcuda.cpp:(.text._ZN6thrust6detail11vector_baseI9KcFeature

NS_16device_allocatorIS2_EEE6resizeEjS2_[thrust::detail::vect

or_base<KcFeature, thrust::device_allocator >::resize(unsigned int, KcFeature)]+0xd2): undefined reference to `void thrust::detail::device::cuda::for_each<thrust::detail::vector_base<KcFeature, thrust::device_allocator >::iterator::device_dereferenceable_type, thrust::detail::dispatch::detail::destroyer >(thrust::detail::vector_base<KcFeature, thrust::device_allocator >::iterator::device_dereferenceable_type, thrust::detail::vector_base<KcFeature, thrust::device_allocator >::iterator::device_dereferenceable_type, thrust::detail::dispatch::detail::destroyer)’

Any ideas?

2

[quote name=‘jrprak’ post=‘563741’ date=‘Jul 10 2009, 04:47 AM’]

Hi,

why does declaring a device vector of a user-defined type (like a struct) cause a compile-time error?

This code doesn’t work because of it:

[codebox]#include

struct KcFeature

{

float x;

float y;

float val;

};

int main(void)

{

thrust::device_vector<KcFeature> dv;

return 0;

}

[/codebox]

and it seems to work fine.

I would try the following:

  1. make sure that you’re using nvcc 2.2 (or 2.3 beta) by running nvcc --version.

  2. delete all object files and recompile everything

  3. try compiling the code snippet above

Jared and I read this board frequently, but in the future you should post questions to thrust-users [1] so we don’t miss anything.

[1] http://groups.google.com/group/thrust-users

3 
obj/release/kltcuda.cpp.o: In function `thrust::detail::vector_base<KcFeature, thrust::device_allocator<KcFeature> >::resize(unsigned int, KcFeature)':

kltcuda.cpp:(.text._ZN6thrust6detail11vector_baseI9KcFeature

It looks like you’re compiling with the MSVC compiler instead of nvcc, the CUDA compiler. Try giving your source file a ‘.cu’ extension and rename it to ‘kltcuda.cu’ to make sure that MSVC uses nvcc to compile.