I can't compile this project by Cmake at all for many day and nights about amgx

[ConduciveMocha]

Is your path set correctly? It looks like CMAKE can't find your toolkit. Let me know if you get this working. I've been trying for hours to build this with the same configuration with no luck

[ningjiafei]

I have set path but I can't sure that is complete. It seem that CMAKE can't find something about MPI and CUDA.
CUDA_PATH_V10_0 = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0
CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0
MSMPI_BIN = C:\Program Files\Microsoft MPI\Bin\

[ningjiafei]

Here is another question that why every folder has a CMAKELIST.TXT？ Does that indicate that every folder could be compile separately?

[Doublove]

The error "CMake Error at CMakeLists.txt:247 (STRING):" means cmake didn't select any CUDA toolkit version because you only installed CUDA10.0 when the cmake only got 9.x max. I found one solution to write your own CUDA toolkit version's cmake just by changing the 9 to 10.

[ningjiafei]

for the error :
CMake Error at CMakeLists.txt:247 (STRING):
STRING sub-command REGEX, mode REPLACE needs at least 6 arguments total to
command.
I change the command 247
from
STRING(REGEX REPLACE " " ";" CUDA_ARCH ${CUDA_ARCH})
to
STRING(REGEX REPLACE " " ";" CUDA_ARCH $"{CUDA_ARCH}")

AND setup the MPICH2
Now I compile AMGX with cmake successfully .
BUT the sln file couldn't work fine, I can't build it in VS2017.

[ningjiafei]

The error "CMake Error at CMakeLists.txt:247 (STRING):" means cmake didn't select any CUDA toolkit version because you only installed CUDA10.0 when the cmake only got 9.x max. I found one solution to write your own CUDA toolkit version's cmake just by changing the 9 to 10.

OH I SEE!!! THANK YOU !!

[ningjiafei]

The error "CMake Error at CMakeLists.txt:247 (STRING):" means cmake didn't select any CUDA toolkit version because you only installed CUDA10.0 when the cmake only got 9.x max. I found one solution to write your own CUDA toolkit version's cmake just by changing the 9 to 10.
“ write your own CUDA toolkit version's cmake just by changing the 9 to 10.” build successfully,but the solution couldn't work. We may rewrite the CMakeLists.txt:242-244 that:
IF(CUDA_VERSION_MAJOR MATCHES 10)
SET(CUDA_ARCH "35 52 60 75" CACHE STRING "Target Architectures (SM35 SM52 SM60 SM75), multiple are allowed")
ENDIF(CUDA_VERSION_MAJOR MATCHES 10)