NVidia Thrust device_vector of strings

Question

I started using the NVidia Thrust library which comes as part of the CUDA 4.0 toolkit and wanted to verify something before digging deeper. I can perform the following and have no issues during build:

thrust::host_vector <int> iVec;
thrust::device_vector <int> iVec2;
thrust::host_vector <std::string> sVec;

When I try the following I get a compilation error:

    thrust::device_vector <std::string> sVec2;

What I would like to know is can I assume any data type I can use in an STL vector should be useable in a thrust vector irrespective of whether it's device or host? Or are there limitations here and I shouldn't expect this to work?

The error I get is the following:

c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\device\cuda\for_each.inl(93): error C2027: use of undefined type 'thrust::detail::STATIC_ASSERTION_FAILURE' 1> with 1> [ 1> x=false 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\device\dispatch\for_each.h(56) : see reference to function template instantiation 'RandomAccessIterator thrust::detail::device::cuda::for_each_n(RandomAccessIterator,Size,UnaryFunction)' being compiled 1> with 1> [ 1> RandomAccessIterator=thrust::detail::normal_iterator>, 1> Size=__w64 int, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\device\for_each.inl(43) : see reference to function template instantiation 'RandomAccessIterator thrust::detail::device::dispatch::for_each_n(RandomAccessIterator,Size,UnaryFunction,thrust::detail::cuda_device_space_tag)' being compiled 1> with 1> [ 1> RandomAccessIterator=thrust::detail::normal_iterator>, 1> OutputIterator=thrust::detail::normal_iterator>, 1> Size=__w64 int, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\device\for_each.inl(54) : see reference to function template instantiation 'OutputIterator thrust::detail::device::for_each_n(OutputIterator,Size,UnaryFunction)' being compiled 1> with 1> [ 1> OutputIterator=thrust::detail::normal_iterator>, 1> InputIterator=thrust::detail::normal_iterator>, 1> UnaryFunction=thrust::detail::device_destroy_functor, 1> Size=__w64 int 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\dispatch\for_each.h(72) : see reference to function template instantiation 'InputIterator thrust::detail::device::for_each(InputIterator,InputIterator,UnaryFunction)' being compiled 1> with 1> [ 1> InputIterator=thrust::detail::normal_iterator>, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\for_each.inl(51) : see reference to function template instantiation 'InputIterator thrust::detail::dispatch::for_each(InputIterator,InputIterator,UnaryFunction,thrust::device_space_tag)' being compiled 1> with 1> [ 1> InputIterator=thrust::detail::normal_iterator>, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\for_each.inl(67) : see reference to function template instantiation 'InputIterator thrust::detail::for_each(InputIterator,InputIterator,UnaryFunction)' being compiled 1> with 1> [ 1> InputIterator=thrust::detail::normal_iterator>, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\dispatch\destroy.h(59) : see reference to function template instantiation 'void thrust::for_each>(InputIterator,InputIterator,UnaryFunction)' being compiled 1> with 1> [ 1> ForwardIterator=thrust::detail::normal_iterator>, 1> T=value_type, 1> InputIterator=thrust::detail::normal_iterator>, 1> UnaryFunction=thrust::detail::device_destroy_functor 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\destroy.h(42) : see reference to function template instantiation 'void thrust::detail::dispatch::destroy(ForwardIterator,ForwardIterator,thrust::detail::false_type)' being compiled 1> with 1> [ 1> ForwardIterator=thrust::detail::normal_iterator> 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\vector_base.inl(442) : see reference to function template instantiation 'void thrust::detail::destroy>(ForwardIterator,ForwardIterator)' being compiled 1> with 1> [ 1> Pointer=thrust::device_ptr, 1> ForwardIterator=thrust::detail::normal_iterator> 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\detail\vector_base.inl(440) : while compiling class template member function 'thrust::detail::vector_base::~vector_base(void)' 1> with 1> [ 1> T=std::string, 1> Alloc=thrust::device_malloc_allocator 1> ] 1> c:\program files\nvidia gpu computing toolkit\cuda\v4.0\include\thrust\device_vector.h(55) : see reference to class template instantiation 'thrust::detail::vector_base' being compiled 1> with 1> [ 1> T=std::string, 1> Alloc=thrust::device_malloc_allocator 1> ] 1> c:\users\fsquared\mydata\idata\main.cpp(119) : see reference to class template instantiation 'thrust::device_vector' being compiled 1> with 1> [ 1> T=std::string 1> ] ========== Build: 0 succeeded, 1 failed, 0 up-to-date, 0 skipped ==========

I am using MSCV 2010 here.

I have no experience with either CUDA or thrust (therefore this is not an answer), but I think the datatype has to be supported on the device, which I doubt for a C++ class to be supported in CUDA. Also keep in mind that std::string is not a simple standard data type, but a C++ class, also not a simple one (much template stuff). It could work with pointers to strings or with plain old char arrays, but all that I'm just guessing. — Christian Rau
– Christian Rau, Commented May 9, 2011 at 1:12

talonmies · Accepted Answer · 2018-01-15 11:24:51Z

2

CUDA does not support standard C++ container types in device code, it is basically limited to C++ POD types only. You can define your own classes for use on the GPU, but the constructor and member functions must be defined as CUDA __device__ functions, and there are still a number of limitations on what language features are support in device code.

edited Jan 15, 2018 at 11:24

answered May 9, 2011 at 2:22

talonmies

72.8k35 gold badges204 silver badges296 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

NVidia Thrust device_vector of strings

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related