Is CUDA 10.2 still supported by the latest AMGX commit?

Description

Compilation errors when using NVIDIA's CUDA 10.2 Docker image (Container OS: Ubuntu 10.2; gcc 7.5) and the latest CMake (cmake 3.23.1). The error message is nvcc fatal : redefinition of argument 'optimize'. The cause of the error is the multiple -O3 flags in the compilation flags.

It seems nvcc 10.2 does not support multiple -O3 flags? Using the Docker image of CUDA 11.4.3/Ubuntu 20.04/gcc 9.4/Cmake 3.23.1 does not give any error even though the compilation flags also have multiple -O3.

To reproduce

Using NVIDIA's CUDA Docker image (CUDA 10.2; Ubuntu 18.04):

$ docker run -it --rm --gpus all nvidia/cuda:10.2-devel-ubuntu18.04 /bin/bash

Once inside the container, in the interactive shell:

# apt-get update && apt-get install -y curl
# curl -L https://github.com/Kitware/CMake/releases/download/v3.23.1/cmake-3.23.1-linux-x86_64.sh -o /cmake-installer.sh
# sh /cmake-installer.sh --prefix=/usr --exclude-subdir --skip-license
# curl -L https://github.com/NVIDIA/AMGX/tarball/6f14016e420617b4d57580f77b5a2ab4c28a162d -o /amgx.tar.gz
# mkdir /amgx
# tar -xf /amgx.tar.gz -C /amgx --strip-component=1
# cd /amgx
# mkdir build && cd build
# cmake -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_VERBOSE_MAKEFILE=ON ..
# make

It will show the error message (simplified and organized):

[  1%] Building CUDA object CMakeFiles/amgx_libs.dir/base/src/aggregation/coarseAgenerators/coarse_A_generator.cu.o
/usr/local/cuda/bin/nvcc
    -forward-unknown-to-host-compiler  -I/amgx/external/rapidjson/include -I/amgx/base/include -I/amgx/core/include
    -I/amgx/core/../base/include -I/amgx/template_plugin/include -I/amgx/template_plugin/../base/include
    -I/amgx/template_plugin/../core/include
    -O3 -DNDEBUG
    --generate-code=arch=compute_35,code=[compute_35,sm_35]
    --generate-code=arch=compute_52,code=[compute_52,sm_52]
    --generate-code=arch=compute_60,code=[compute_60,sm_60]
    --generate-code=arch=compute_70,code=[compute_70,sm_70]
    -O3 -DNDEBUG --Werror cross-execution-space-call -DNVTX_RANGES
    -Xcompiler=-rdynamic -Xcompiler=-fPIC
    -Xcompiler=-fvisibility=default -DDISABLE_MIXED_PRECISION -DCUSPARSE_GENERIC_INTERFACES
    -Xcompiler " -Wno-terminate -static-libgcc -fopenmp -DRAPIDJSON_DEFINED"
    -std=c++14 -MD
    -MT CMakeFiles/amgx_libs.dir/base/src/aggregation/coarseAgenerators/coarse_A_generator.cu.o
    -MF CMakeFiles/amgx_libs.dir/base/src/aggregation/coarseAgenerators/coarse_A_generator.cu.o.d
    -x cu -c /amgx/base/src/aggregation/coarseAgenerators/coarse_A_generator.cu
    -o CMakeFiles/amgx_libs.dir/base/src/aggregation/coarseAgenerators/coarse_A_generator.cu.o
nvcc fatal   : redefinition of argument 'optimize'

The error was caused by multiple -O3 flags.

Reproduce the build in NVIDIA's CUDA 10.2 Ubuntu 18.04 Docker image using CMake 3.23.1 and the referenced AMGX commit. Start with the verbose nvcc command shown in the report and trace where its duplicate -O3 flags are generated; done means the AMGX build completes without the nvcc optimize-argument error.