thrust 1.8.3 exclusive_scan with custom temporary allocation, no default constructor error

[Mochimazui]

Using VS2013 with CUDA 8.0 + thrust 1.8.3.
Same error occurs with CUDA 7.5 + thrust 1.8.3.

See code at https://gist.github.com/Mochimazui/007b243e7be16527f5d5380d596c1993.
This is a simple modification from thrust/examples/cuda/custom_temporary_allocation.cu. Just add scan.h and replace the sort function with exclusive_scan.

nvcc shows following compile error:

1>C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include\thrust/detail/reference.inl(127): error : no default constructor exists for class "thrust::detail::execute_with_allocator<cached_allocator, thrust::system::cuda::detail::execute_on_stream_base>" 1>C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include\thrust/detail/reference.inl(98): error : no default constructor exists for class "thrust::detail::execute_with_allocator<cached_allocator, thrust::system::cuda::detail::execute_on_stream_base>"

[bramton]

I ran into the same error while trying to compile OpenCV in combination with the CUDA 8.0 libraries.
I am using VS2015 Community.
I checked out the thrust 1.8.3 branch from GitHub into the CUDA 8.0 include directory and this works fine for me.
@Mochimazui Could you try this as well ?

[Mochimazui]

@bramton tried same environment as you said, still getting them same error.

Using Microsoft Visual Studio Community 2015 Version 14.0.24720.00 Update 1, with thrust 1.8.3 from GitHub and copied to CUDA 8.0 include directory.

VS output window shows
1> D:\code\thrust_1.8.3_allocator\thrust_1.8.3_allocator>"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin\nvcc.exe" -gencode=arch=compute_20,code=\"sm_20,compute_20\" --use-local-env --cl-version 2015 -ccbin "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\x86_amd64" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include" --keep-dir x64\Release -maxrregcount=0 --machine 64 --compile -cudart static -DWIN32 -DNDEBUG -D_CONSOLE -D_LIB -D_UNICODE -DUNICODE -Xcompiler "/EHsc /W3 /nologo /O2 /Zi /MD " -o x64\Release\main.cu.obj "D:\code\thrust_1.8.3_allocator\thrust_1.8.3_allocator\main.cu"

Is that all the same with yours?

[bramton]

This is the error I get when I use the thrust library provided with CUDA 8.0:
C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0/include\thrust/detail/reference.inl(127): error : no default constructor exists for class "thrust::detail::execute_with_allocator<cv::cuda::device::ThrustAllocator, thrust::system::cuda::detail::execute_on_stream_base>"

What I did to get it working was the following:

• Renamed ...CUDA\v8.0\include\thrust\ to thrust_old
• Made a shortcut ..CUDA\v8.0\include\thrust\ to ...thrust-git-checkout\thrust

[bramton]

I made a bug report at NVIDIA with this issue, this is their reply:

Actually, this is a known issue and has been fixed now in our development builds. Unfortunately the CUDA 8.0 RC build doesn't contain this fix yet, it would be available for you in the next release of CUDA 8.0.

Could someone close this issue ?

[jaredhoberock]

Thanks @Mochimazui and @bramton! We'll keep this issue open until the fix is reflected in the repository to make others aware of the problem and let them know that a solution is in progress.

[3gx]

The solution is already in place. Please use "cuda-next-release" branch where the problem is fixed. The fix will also be in CUDA 8.0 GA.

[twmht]

@bramton 's solution works for me.

[zhanghang1989]

Thanks a lot! It works for me on ubuntu 16.