Thrust and CUDA 2.3a problem

Accelerated Computing CUDA CUDA Programming and Performance
1

everything is updated to the new version.

CUDA toolkit 2.3.a
CUDA DRiver 2.3.1a
on Snow Leopard:

I have problem with thrust:

I got this error on all examples using thrust:

/usr/include/c++/4.2.1/ext/atomicity.h(51): error: identifier “__sync_fetch_and_add” is undefined

/usr/include/c++/4.2.1/ext/atomicity.h(55): error: identifier “__sync_fetch_and_add” is undefined

2 errors detected in the compilation of “/tmp/tmpxft_00002fed_00000000-4_version.cpp1.ii”.

If I use the old gcc (g++) 4.0.1 than everything is fine…

For the examples in the SDK I had to modify the common.mk to use -m32, otherwise all compiled and working…
So my question is: What is the problem with thrust?

Any help, or just wait for the new release?

Thanks,

TRS

2

Thanks for bringing this to our attention. This is probably some incompatibility with gcc-4.2 + nvcc 2.3a that we can work around. We’re looking into it.

3

Thrust v1.1 should now work with CUDA 2.3a & Snow Leopard. Please let us know if you have any further difficulty.

4

What was your workaround there. I’m getting the same error “__sync_add” message when try to compile a mix of .cpp and .cu files using FindCUDA on the release candidate three of cmake. I’m using fedora 11 with gcc 4.4.1. Cheers

5

Use this incantation before any standard headers get included:

#if defined(__APPLE__)

#if _GLIBCXX_ATOMIC_BUILTINS == 1

#undef _GLIBCXX_ATOMIC_BUILTINS

#endif // _GLIBCXX_ATOMIC_BUILTINS

#endif // __APPLE__
6

I want to use reduce from thrust and made this code

device_vector<float> Y1(3*iH*iW,0);

int sum;

sum = thrust::reduce(Y1.begin(), Y1.end(), (float) 0, thrust::plus<float>());

and encountered error in lint of reduce function while linking.

In function `float thrust::detail::device::detail::reduce_device<thrust::detail::normal_iterator<thrust::device_ptr<float> >, float, thrust::plus<float> >(thrust::detail::normal_iterator<thrust::device_ptr<float> >, thrust::detail::normal_iterator<thrust::device_ptr<float> >, float, thrust::plus<float>, thrust::detail::integral_constant<bool, false>)':

main.cpp:(.text._ZN6thrust6detail6device6detail13reduce_deviceINS0

_15normal_iteratorINS_10device_ptrIfEEEEfNS_4plusIfEEEET0_T_S

B_SA_T1_NS0_17integral_constantIbLb0EEE[float thrust::detail::device::detail::reduce_device<thrust::detail::normal_iterator<thrust::device_ptr<float> >, float, thrust::plus<float> >(thrust::detail::normal_iterator<thrust::device_ptr<float> >, thrust::detail::normal_iterator<thrust::device_ptr<float> >, float, thrust::plus<float>, thrust::detail::integral_constant<bool, false>)]+0x6b): undefined reference to `float thrust::detail::device::cuda::reduce<thrust::detail::device::detail::reduce_functor<thrust::detail::normal_iterator<thrust::device_ptr<float> >, float>, float, thrust::plus<float> >(thrust::detail::device::detail::reduce_functor<thrust::detail::normal_iterator<thrust::device_ptr<float> >, float>, unsigned int, float, thrust::plus<float>)'

I use Eclipse to edit, CUDA 2.3 and findCUDA for CUDA, all headers of thrust folder are included

7

Rename your source to main.cu and compile with nvcc.