Build OpenCV with CUDA support

Question

Im use CMake to generate visual studio 2013 solution. Next im try to build it, but get follow error:

Building NVCC (Device) object modules/core/CMakeFile

Answer 1

When using cmake to do configurations, set the option CUDA_GENERATION to specific your GPU architecture. I ran across the same error and tried this to work out the problem.

Answer 2

You can use CUDA_GENERATION to specify the corresponding generation code name for your GPU architecture.

Here's the relevant opencv cmake code that parses the CUDA_GENERATION value:

  set(__cuda_arch_ptx "")
  if(CUDA_GENERATION STREQUAL "Fermi")
    set(__cuda_arch_bin "2.0")
  elseif(CUDA_GENERATION STREQUAL "Kepler")
    set(__cuda_arch_bin "3.0 3.5 3.7")
  elseif(CUDA_GENERATION STREQUAL "Maxwell")
    set(__cuda_arch_bin "5.0 5.2")
  elseif(CUDA_GENERATION STREQUAL "Pascal")
    set(__cuda_arch_bin "6.0 6.1")
  elseif(CUDA_GENERATION STREQUAL "Volta")
    set(__cuda_arch_bin "7.0")
  elseif(CUDA_GENERATION STREQUAL "Auto")
    execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
                     WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
                     RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
                     ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
    if(NOT _nvcc_res EQUAL 0)
      message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
    else()
      set(__cuda_arch_bin "${_nvcc_out}")
      string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
    endif()
endif()

And the wikipedia CUDA page has a nice table for mapping your video card to the right microarchitecture code name (sorry, it's too large to reproduce here):

https://en.wikipedia.org/wiki/CUDA#GPUs_supported

For example, my middling-2012 Macbook Pro has an antique GeForce GT 650M, which the wikipedia table indicates uses the Kepler microarchitecture. Therefore, I use this in my cmake command line:

cmake -D CUDA_GENERATION="Kepler" ...

and the opencv script converts that to "3.0 3.5 3.7" when it displays the config summary, and passes on the corresponding flags to nvcc.

In my case, before setting this properly, I was getting errors about compute_70 not supported. Apparently, there is still an open issue in the opencv tracker as of today (2017-10-07) about auto-detection not working properly.

Answer 3

Following up on Yun's answer (could not leave comment), this worked for me and shows a possible value for CUDA_GENERATION:

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D CUDA_GENERATION=Kepler ..

(Ubuntu 12.04 and 14.04, GTX Titan, and OpenCV 2.4.11 and 3.0.0.)

Answer 4

You should set with cmake these entries CUDA_ARCH_BIN = 3.2 and CUDA_ARCH_PTX = 3.2

Hope it helps.

Regards

Answer 5

Another option. Ubuntu 14.04, GTX Titan X, opencv-2.4.10

cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local -D BUILD_TIFF=ON -D BUILD_EXAMPLES=ON -D CUDA_GENERATION=Auto -D BUILD_NEW_PYTHON_SUPPORT=ON  .. 

I also applied the patch, but I'm not sure whether it ended up being needed. I had tried with and withoutCUDA_GENERATION=Maxwell but Maxwell isn't detected. I did not try CUDA_GENERATION=Auto prior to the patch, that's why I don't know for sure.

Answer 6

Thank you,

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D CUDA_GENERATION=Kepler ..

This let me install opencv-2.4.9.

If you want to know more detaills check this link.