search

LLNL / hiop

HPC solver for nonlinear optimization problems
Other
210 stars 42 forks source link

Cannot build with CUDA flag set ON #678

Closed finkeljos closed 8 months ago

finkeljos commented 9 months ago

Hi,

I am trying to build hiOp on an Nvidia GPU node on my local cluster.

After a fresh clone on the develop branch, I enter an empty build dir enter the command

cmake -DHIOP_USE_GPU=ON -DHIOP_USE_CUDA=ON -DCMAKE_BUILD_TYPE=Release ..

But when building the compiler complains that

Picture1.pdf

I don't understand why the .cpp code has any mention of RAJA if I am not explicitly enabling it. Is there a missing ifdef in this file? It is my understanding I should be able to build only with cuda in order to access cuBLAS, cuSolver etc.

I am using gcc, cuda and OpenMPI modules, all of which are found by Cmake, including my custom build of MAGMA.

cnpetra commented 9 months ago

raja is currently a dependence for cuda.

cnpetra commented 9 months ago

see also https://github.com/LLNL/hiop/pull/676

cnpetra commented 8 months ago

closing this since a partial fix has been done in PR #676. We will remove RAJA as a dependency in the not-so-far future.