This PR brakes the build on Jetson TX1 and TX2

• cmake

-- General configuration for OpenCV 4.4.0-pre =====================================
--   Version control:               4.3.0-464-g5ed6546
-- 
--   Extra modules:
--     Location (extra):            /opencv_contrib/modules
--     Version control (extra):     4.3.0-69-g0f56e6d
 :
-- CUDA detected: 8.0
-- Automatic detection of CUDA generation failed. Going to build for all known architectures.
-- CUDA NVCC target flags: -ccbin;/usr/bin/cc;-gencode;arch=compute_53,code=sm_53;-gencode;arch=compute_62,code=sm_62;-gencode;arch=compute_72,code=sm_72;-D_FORCE_INLINES

• It usually detects the native CC (i.e. sm53 for TX1, sm62 for TX2)

  • Then, sm72 is not in the list of CUDA 8.0. (i.e. The fall safe path is not safe at all)

• make

nvcc fatal   : Unsupported gpu architecture 'compute_72'
CMake Error at cuda_compile_1_generated_gpu_mat.cu.o.Release.cmake:219 (message):
  Error generating
  /opencv/build/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/./cuda_compile_1_generated_gpu_mat.cu.o

This change has broken CUDA builds. Please undo it.

@NHarishGit , I appreciate your comment.
It would be a great help if you could share what "CUDA builds" you are talking about. like this

* Jetson TX2 (TX1)
* Ubuntu 16.04 (Aarch64)
* CUDA 8.0
* GCC 5.4.0
* CMake 3.15.1

OS, architecture, CUDA version, compiler version, cmake version.
That helps the discussion.
Thanks

@NHarishGit , I appreciate your comment.
It would be a great help if you could share what "CUDA builds" you are talking about. like this

* Jetson TX2 (TX1)
* Ubuntu 16.04 (Aarch64)
* CUDA 8.0
* GCC 5.4.0
* CMake 3.15.1

OS, architecture, CUDA version, compiler version, cmake version.
That helps the discussion.
Thanks

OS: WIN 10 x64 2004 build
VS 2019 community v 16.6.2
CUDA v 10.2
CMAKE v 1.17.3
GSTREAMER v 1.16.2
MKL, TBB 2020 versions
CUDA DNN v 7.6.5

I spent quite a few hours to figure out that the generated CMAKE file for NVCCC has NVCC_FLAGS set wrongly without quotes around the path for cl.exe, which breaks on windows system on C:/program files (x86)/blah blah blah.

@cyyever any comments ?
#17598 (comment)

@cyyever , could you comment on what's the situation you were trying to fix ?

@cyyever any comments ?
#17598 (comment)

@cyyever , could you comment on what's the situation you were trying to fix ?

My default g++ is g++-10, without this patch, nvcc would use g++-10 and because no CUDA supports g++-10, it would failed to detect my CUDA arch even if I specify CUDA_HOST_COMPILER.
@NHarishGit
Could it work by changing
LIST(APPEND CUDA_NVCC_FLAGS -ccbin ${CUDA_HOST_COMPILER})
to
LIST(APPEND CUDA_NVCC_FLAGS -ccbin "${CUDA_HOST_COMPILER}") ?
I feel strange that if space leads to your failure, then

  get_filename_component(host_compiler_bindir ${CMAKE_LINKER} DIRECTORY)
  LIST(APPEND CUDA_NVCC_FLAGS -ccbin ${host_compiler_bindir})

would fail before my PR. So should we change it too?

@cyyever, @NHarishGit
I appreciate if you could have a look on #17671
If my understanding is correct, this patch should satisfy all three of us.

@cyyever, @NHarishGit
I appreciate if you could have a look on #17671
If my understanding is correct, this patch should satisfy all three of us.

Looks good to me @tomoaki0705. Do you know tentative plan for 4.4.0 release ?

@NHarishGit , thank you for checking. I don't know the release schedule of 4.4.0. Best