Questions about cuda lib and include files

Accelerated Computing CUDA CUDA Programming and Performance
1

I got a CMakeLists.txt file from internet. I cannot understand the following CUDA lib and include directories.

cuda_include_directories(${local_CUDA_CUT_INCLUDE_DIRS})
cuda_include_directories(${CUDA_INCLUDE_DIRS} ${CUDA_CUT_INCLUDE_DIR} ${local_CUDA_CUT_INCLUDE_DIR})
link_directories(${local_CUDA_CUT_LIBRARY_DIRS})

cuda_add_library(cuda_stuff_library ${SrcCuda})
target_link_libraries(cuda_stuff_library
${CUDA_LIBRARIES}
${cutil_LIB}
)

what are they? Is there any document or instructions about tham?

2

CMake is a compilation tool used for relatively complex project/library compilation management. I suspect you do not need a CMakeList, and you’ll be fine reproducing one of the more simpler makefiles that come with the CUDA samples.

Edit: I see from your other posts (you should really keep it streamlined and put as much detail into a specific post as you can rather than spread them around) that you’re apparently trying to compile some sort of opencv+cuda application that was probably handed down to you and uses CMake, yes?

If so… here’s some CMake information:
[url]http://sbel.wisc.edu/Courses/ME964/2012/Lectures/lecture0214.pdf[/url]
[url]http://doktorandi.andreasherten.de/2013/07/24/an-introduction-to-cmake-and-how-to-use-cuda-in-pandaroot-with-it/[/url]

What I am guessing, and why you wanted CUDA 4.0, or 4.2, for that matter in your other post, is that because the paths in the CUDA toolkit have changed from v4.2 → v5/v6, your cmake files are not finding the correct directories files. It might not be all CUDA related though, so check the paths for any other built-in functions.

3

Yes, what you guessed is exactly correct!
I am trying to build a handed down from internet. The cource code is made up of many libraries including CUDA.

I can build CUDA sampler code in /usr/local/cuda-6.0/samples/ successfully, and the built code can also be executed correctly.

After I installed CUDA 6.0 on my ubuntu, I can build OpenCV 2.4.8 correctly.

From above experiments, I can ensure CUDA is correctly installed on my ubuntu, and I know how to set the default CUDA path.

The code I downloaded can also be built correctly. The problem I encountered is that some cudaMem related issues happened during linking.

In the code I downloaded, there are some other paths related with CUDA4.0. The paths are different with CUDA 6.0. I don’t know how to find the corresponding paths for them on CUDA 6.0.
Such as,

cuda/cuda_sdk/C/common/inc
cuda/cuda_sdk/C/lib
set(cutil_LIB “cutil”)
set(local_CUDA_LIB_DIR “/usr/lib64/nvidia”)
set(local_CUDA_LIB “/usr/lib64/nvidia/libcuda.so”)

In the CMakeLists.txt file of my code, I can’t find the paths listed above on CUDA 6.0.
Where are they on CUDA 6.0?