Question about building PlusServer with OpenCV on Ubuntu 20.04.1

[nist-anka]

Ubuntu 20.04.1 LTS, NVIDIA Corporation GP104 [GeForce GTX 1070], Driver Version: 460.27.04 CUDA Version: 11.2.
CMake 3.16.3.

I want to build PlusServer with options: PLUSBUILD_USE_OpenCV and PLUS_USE_OpenCV_VIDEO on.

I have gcc8 ((Ubuntu 8.4.0-3ubuntu2) 8.4.0) installed by Nvidia, and gcc9 ((Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0) on my system.
I set gcc8 as compiler in CMake and I am getting this building error:

error: #error -- unsupported GNU version! gcc versions later than 8 are not supported!
138 | #error -- unsupported GNU version! gcc versions later than 8 are not supported!
| ^~~~~
CMake Error at cuda_compile_1_generated_gpu_mat.cu.o.Release.cmake:220 (message):
Error generating
xxx/OpenCV-bin/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/./cuda_compile_1_generated_gpu_mat.cu.o

Please, advise what should I do.
I would like to use PlusServer with webcam.

[adamrankin]

Find the line of code with that #error message and comment it out. See if gcc8 works

[nist-anka]

You mean, I should change the crt file?

…

On Tue, Jan 19, 2021 at 2:24 PM Adam Rankin ***@***.***> wrote: Find the line of code with that #error message and comment it out. See if gcc8 works — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#767 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEWMRY3EP3IFAI3E6V56KY3S2XL7LANCNFSM4WJH3BNQ> .

[adamrankin]

Yes

[nist-anka]

Now , after commenting the line
//#error -- unsupported GNU version! gcc versions later than 8 are not supported!
in /usr/include/crt/host_config.h
it is:
[ 39%] Building NVCC (Device) object modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat.cu.o
/usr/include/c++/9/bits/stl_function.h(437): error: identifier "__builtin_is_constant_evaluated" is undefined

1 error detected in the compilation of "/tmp/tmpxft_00018adf_00000000-14_gpu_mat.compute_75.cpp1.ii".
CMake Error at cuda_compile_1_generated_gpu_mat.cu.o.Release.cmake:280 (message):
Error generating file
/home/anka/Plus/bin/OpenCV-bin/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/./cuda_compile_1_generated_gpu_mat.cu.o

make[5]: *** [modules/core/CMakeFiles/opencv_core.dir/build.make:484: modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat.cu.o] Error 1
make[4]: *** [CMakeFiles/Makefile2:2644: modules/core/CMakeFiles/opencv_core.dir/all] Error 2
make[3]: *** [Makefile:163: all] Error 2
make[2]: *** [CMakeFiles/OpenCV.dir/build.make:113: OpenCV-prefix/src/OpenCV-stamp/OpenCV-build] Error 2
make[1]: *** [CMakeFiles/Makefile2:233: CMakeFiles/OpenCV.dir/all] Error 2
make: *** [Makefile:84: all] Error 2

[adamrankin]

Ok. You'll have to use a newer version of Cuda or disable cuda for OpenCV

[markasselin]

Another idea would be to install an older version of gcc (5,6 or 7) and set it as default using update-alternatives. As far as I know PLUS still supports these, @adamrankin is this still the case?

[adamrankin]

It will probably compile. Whatever is default in ubuntu 18.04 and newer are supported.

[adamrankin]

A similar experience here:
espressomd/espresso#3654

[nist-anka]

Thanks, I tried gcc-7.5 but got different errors, I am trying your suggested: espressomd/espresso#3654
What should I set as CUDA_SDK_ROOT_DIR, I tried the top /usr/local/cuda-11.2 but it was rejected.

[adamrankin]

Why was it rejected?

[nist-anka]

I don't know, the CMake just replaced it with CUDA_SDK_ROOT_DIR_NOT_FOUND and I do not know what is it looking for in the SDK directory, I guess, I'll have to look closer at all the CmakeLists

…

On Wed, Jan 20, 2021 at 11:30 AM Adam Rankin ***@***.***> wrote: Why was it rejected? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#767 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEWMRY3KBSNQK644556RTR3S24AJBANCNFSM4WJH3BNQ> .

[nist-anka]

After looking around, my UDA_SDK_ROOT_DIR
is
/usr/local/cuda-11.2/samples/common
if anyone else ever looks at this problem.
Other issue is
fatal error: dynlink_nvcuvid.h: No such file or directory
#include <dynlink_nvcuvid.h>

It seems, that cuda video decoder is deprecated.

[nist-anka]

I added option
-DBUILD_opencv_cudacodec=OFF

to External_OpenCV.cmake and managed to build OpenCV. It was very slow comparing to building vtk.

[adamrankin]

Yes, there is a recommendation in the cmake configure output to set the cuda generation variable to only build for your current card (if building for yourself). Drastically reduces compile time.

[nist-anka]

I set it to Turing, wrong?

…

On Wed, Jan 20, 2021 at 4:22 PM Adam Rankin ***@***.***> wrote: Yes, there is a recommendation in the cmake configure output to set the cuda generation variable to only build for your current card (if building for yourself). Drastically reduces compile time. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#767 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEWMRY5COS7NMUSMWP4E2M3S25CTBANCNFSM4WJH3BNQ> .

[nist-anka]

Problem solved. Thank you for all your help.

[nist-anka]

I have made a new build to check the time - OpenCV build takes ~35 minutes, cuda 10, cuda generation variable set to Turing.

…

On Wed, Jan 20, 2021 at 4:22 PM Adam Rankin ***@***.***> wrote: Yes, there is a recommendation in the cmake configure output to set the cuda generation variable to only build for your current card (if building for yourself). Drastically reduces compile time. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#767 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEWMRY5COS7NMUSMWP4E2M3S25CTBANCNFSM4WJH3BNQ> .