So I ran

Code: Select all

mkdir build
cmake ..
ccmake . -> modified nvcc to ON
root@traxit-gpu-staging:/usr/local/eigen/build# ./check.sh cuda*

and here is the output:

Code: Select all

Building NVCC (Device) object test/CMakeFiles/cuda_basic.dir//./cuda_basic_generated_cuda_basic.cu.o /usr/local/eigen/Eigen/src/SVD/BDCSVD.h:1164:1: error: prototype for 'Eigen::BDCSVD<Eigen::Matrix<typename Eigen::internal::traits<T>::Scalar, Eigen::internal::traits<T>::RowsAtCompileTime, Eigen::internal::traits<T>::ColsAtCompileTime, (0 | ((Eigen::internal::traits<T>::Flags & 1u) ? (Eigen::._98)1u : (Eigen::._98)0u)), Eigen::internal::traits<T>::MaxRowsAtCompileTime, Eigen::internal::traits<T>::MaxColsAtCompileTime> > Eigen::MatrixBase<Derived>::bdcSvd(unsigned int) const' does not match any in class 'Eigen::MatrixBase<Derived>'/usr/local/eigen/Eigen/src/Core/MatrixBase.h:393:388: error: candidate is: Eigen::BDCSVD<Eigen::Matrix<typename Eigen::internal::traits<T>::Scalar, Eigen::internal::traits<T>::RowsAtCompileTime, Eigen::internal::traits<T>::ColsAtCompileTime, (0 | ((Eigen::internal::traits<T>::Flags & 1u) ? Eigen::RowMajor :  Eigen::ColMajor)), Eigen::internal::traits<T>::MaxRowsAtCompileTime, Eigen::internal::traits<T>::MaxColsAtCompileTime> > Eigen::MatrixBase<Derived>::bdcSvd(unsigned int) cons

CMake Error at cuda_basic_generated_cuda_basic.cu.o.cmake:252 (message):
  Error generating file
  /usr/local/eigen/build/test/CMakeFiles/cuda_basic.dir//./cuda_basic_generated_cuda_basic.cu.o

This is the default branch of the eigen dev repository. I am using CUDA 6.5, running on Ubuntu 12.04 (on an AWS GPU instance).
EDIT: gcc version is 4.6.3

Do you have any input on this error?

Workaround pushed: https://bitbucket.org/eigen/eigen/commits/31d2b16f5ef1/

What does the __CUDACC__ environment variable correspond to?

I removed the build folder, cmaked it again, and now ./check.sh cuda* returns

Code: Select all

make: *** No rule to make target `cuda_basic'.  Stop.

You have to redo: "ccmake . -> modified nvcc to ON"

Nice, good job! I had forgotten that hg pull does not update...
Here is the result of tests:

Code: Select all

Built target cuda_basic
Test project /usr/local/eigen/build
    Start 576: cuda_basic
1/1 Test #576: cuda_basic .......................   Passed    1.42 sec

100% tests passed, 0 tests failed out of 1

Label Time Summary:
Official    =   1.42 sec

Total Test time (real) =   1.88 sec


I will now try to use my code with CUDA, I'll keep you posted. By the way, quick question, can I still use files with .cpp extension? I see .cu everywhere, I guess that's just a semantic differentiation, and I read that there was no change needed in order to make it work (except for CMakeLists files).