Closed cinterloper closed 5 years ago
Is it possible to get more information about the system you're building on? OS, distro, compiler, CMake version, CMake configure line, etc.
Does Terra build without CUDA? CUDA is optional for Terra; do you actually need it?
Can you build other CMake projects with CUDA? We're using CMake's built in FindCUDA module, so in theory this should be a problem for any other CMake projects as well.
Its nvidia L4T, an ubuntu 18.04 derivative for arm64. It didnt work with no arguments (is there a way to explicitly disable cuda?) but I got cmake to work with the following command line:
cmake -DCMAKE_VERBOSE_MAKEFILE=On -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER=/usr/local/cuda-10.0/bin/nvcc -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.0/ -DCLANG_INCLUDE_DIRS=/usr/local/cuda-10.0/include/ .
However it wont build because aarch64 isint in luajit until 2.1 and terra is currently including 2.0.5
Terra does work with LuaJIT 2.1, we just need to add some logic to allow this to be configurable in CMake. In the mean time, you can edit this to set the LuaJIT version to anything you want, e.g. 2.1.0-beta3
should do what you want. (I.e. set LUAJIT_VERSION_MINOR
to 1
and LUAJIT_VERSION_EXTRA
to .0-beta3
.)
Looking again at your log from before, I think the actual error is that CLANG_INCLUDE_DIRS
isn't being set properly, which may indicate that the LLVM build isn't being found, or perhaps doesn't have what we need. E.g. are the -dev
packages for clang
installed? Usually I'd recommend installing clang-X.Y libclang-X.Y-dev llvm-X.Y-dev
if you're using the Ubuntu package repos.
Forgot to add this link in the last email:
https://github.com/zdevito/terra/blob/master/cmake/Modules/GetLuaJIT.cmake#L3-L6
when building on a nvidia xavier with cuda 10, cmake fails with the following error: