[pcl] Add cuda and tools option #2775
Conversation
|
I found a way to workaround the CUDA problem, the codes are like following set(WITH_CUDA OFF)
if("cuda" IN_LIST FEATURES)
set(WITH_CUDA ON)
if(MSVC_VERSION GREATER 1900)
get_filename_component(CUDA_HOST_COMPILER "../../VC/bin/cl.exe" ABSOLUTE BASE_DIR "$ENV{VS140COMNTOOLS}"})
endif()
endif()
vcpkg_configure_cmake(... -DCUDA_HOST_COMPILER=${CUDA_HOST_COMPILER})However, I found that |
| -DWITH_LIBUSB=OFF | ||
| -DWITH_OPENNI2=${WITH_OPENNI2} | ||
| -DWITH_PCAP=${WITH_PCAP} | ||
| -DWITH_PNG=OFF | ||
| -DWITH_PNG=ON |
There was a problem hiding this comment.
Should libpng be added as a dependency because of this?
There was a problem hiding this comment.
It could be, but pcl depends on vtk which takes libpng as a dependency, so adding it as a dependency is not necessary
|
Now pcl can be built with cuda almost successfully. The problem occured during |
|
Is it able to merge this branch now? |
|
Yep, thanks for the PR and sorry for the delay! |
With this portfile, pcl can be theoretically compiled against cuda. However, I met problems when building it with cuda like #2749. How I solved this is exactly described here at StackOverflow which was caused by CUDA cmake bug.
In addition, with current cuda version (8.0), only toolset along with VS2015 can be used to compile .cu files (while I am using VS2017). The problem is also describe in another post at StackOverflow, workarounds are mentioned in both posts. I am able to build pcl with cuda when specifying CUDA_HOST_COMPILER as the cl compiler in toolset v140. So the problem is with cuda rather than with pcl/opencv in my opinion, which could be solved or at least warned in the portfile of cuda.