Cannot build with CUDA

[yathAg]

I am trying to build Openroad with cmake .. -DGPU=true and I get the following output

-- GPU is enabled
-- CUDA is found
-- The CUDA compiler identification is NVIDIA 10.1.243
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: /usr/bin/nvcc - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
-- Found re2: /opt/or-tools/lib/cmake/re2/re2Config.cmake (found version "9.0.0") 
-- Found Clp: /opt/or-tools/lib/cmake/Clp/ClpConfig.cmake (found version "1.17.7") 
-- Found Cbc: /opt/or-tools/lib/cmake/Cbc/CbcConfig.cmake (found version "2.10.7") 
-- Found Eigen3: /usr/local/share/eigen3/cmake/Eigen3Config.cmake (found version "3.4.0") 
-- Found SCIP: /opt/or-tools/lib/cmake/scip/scip-config.cmake (found version "8.0.1") 
-- GUI is enabled
-- Found Boost: /usr/local/lib/cmake/Boost-1.80.0/BoostConfig.cmake (found version "1.80.0") found components: serialization 
-- Found OpenMP_CXX: -fopenmp (found version "4.5") 
-- Found OpenMP: TRUE (found version "4.5")  
-- Could NOT find VTune (missing: VTune_LIBRARIES VTune_INCLUDE_DIRS) 
-- Found Boost: /usr/local/lib/cmake/Boost-1.80.0/BoostConfig.cmake (found suitable version "1.80.0", minimum required is "1.78")  
-- Found Boost: /usr/local/lib/cmake/Boost-1.80.0/BoostConfig.cmake (found version "1.80.0") found components: serialization system thread 
-- TCL readline enabled
-- Tcl Extended disabled
-- Python3 enabled
-- Configuring done
CMake Warning (dev) in src/gpl/CMakeLists.txt:
  Policy CMP0104 is not set: CMAKE_CUDA_ARCHITECTURES now detected for NVCC,
  empty CUDA_ARCHITECTURES not allowed.  Run "cmake --help-policy CMP0104"
  for policy details.  Use the cmake_policy command to set the policy and
  suppress this warning.

  CUDA_ARCHITECTURES is empty for target "gpl".
This warning is for project developers.  Use -Wno-dev to suppress it.

CMake Error in src/gpl/CMakeLists.txt:
  Target "gpl" requires the language dialect "CUDA17" .  But the current
  compiler "NVIDIA" does not support this, or CMake does not know the flags
  to enable it.


-- Generating done
CMake Generate step failed.  Build files cannot be regenerated correctly.

I have Cuda installed and on nvidia-smi I get

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.85.12    Driver Version: 525.85.12    CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0  On |                  N/A |
| N/A   51C    P5    18W / 115W |    187MiB /  6144MiB |     19%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1211      G   /usr/lib/xorg/Xorg                 52MiB |
|    0   N/A  N/A      1875      G   /usr/lib/xorg/Xorg                133MiB |
+-----------------------------------------------------------------------------+

and on nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

Any help on how to get the setup working is really appreciated and thanks in advanced.

[vvbandeira]

Please install a newer version of nvcc and try again. As per NVIDIA docs, you will require at least v11; see more here.

[rovinski]

Is there a way in Cmake to enforce nvcc >= 11?

[Pavelevich]

Yes, is possible to enforce a min version of NVCC using CMake,
por ejemplo:
min required version of NVCC
set(CMAKE_CUDA_COMPILER_VERSION 11.0)
then:
find the CUDA package
find_package(CUDA ${CMAKE_CUDA_COMPILER_VERSION} REQUIRED)
and:
check that the version of NVCC is >= 11.0
if(CUDA_VERSION LESS ${CMAKE_CUDA_COMPILER_VERSION})
message(FATAL_ERROR "Minimum required version of nvcc is ${CMAKE_CUDA_COMPILER_VERSION}")
endif()

[maliberty]

If we bump to cmake 3.10 then https://cmake.org/cmake/help/latest/module/FindCUDA.html suggests we can use the usual VERSION keyword.

FYI - the use of CUDA is quite minimal and probably not worth the bother at this point.

[vvbandeira]

This looks like a cleaner option. We already install cmake 3.24 from source on our dependency installer, and this dependency can be installed without root. @yathAg could you confirm that the version update is enough?

[yathAg]

Updating the nvcc version to 12.1 did the trick. Thank you for the help.

[vvbandeira]

Thanks @yathAg. @maliberty I will look into adding the check and adding a pipeline to test the code with gpu.