[openmp][amdgpu] Remove dependencies clause [llvm,clang,lld] from project

[ronlieb]

Summary

There are other projects that can affect build of opnemp, libcxx for example.
Desire to have each commit trigger a buildbot.

[jdoerfert]

I don't understand how does this configuration depend on libcxx, or any other project for that matter?

[JonChesterfield]

Shared cmake iiuc. Does dropping the depend list mean build on any change to anything? In favour if so

[JonChesterfield]

Desire to have each commit trigger a buildbot.

[ronlieb]

i think the original intent was to limit what triggered the buildbot to just llvm, lld, clang, and openmp projects.
Yesterday we observed that changes specific to libcxx did not trigger a buildbot , rather it just rolled into a subsequent commit that did trigger a buildbot which failed.
Our desire is to fire a bot per commit for clarity. The current way we are structuring the depends creates a bit of confusion as to what patch may have caused a failure.

FYI: The patch that broke us, and Kruse's bot as well, is
407e07aa67ab [runtimes] Set more paths when building runtimes standalone
which was later reverted by Leonard Chan.

FYI: The revert also did not trigger a buildbot.

Bit of history:
The clause depends_on_projects was suggested in review of the original patch to implement a builder for amdgpu.
we added that, and tried to model ours closely to MKruse's "openmp-offload-cuda-project" and "openmp-offload-cuda-runtime".
Note: Neither of the cuda builders for openmp specify depends_on_projects

[Meinersbur]

I think originally, depends_on_projects existed before LLVM was a monorepository so these were the projects to be checked out. Since moving to the monorepository, depends_on_projects is used for the LLVM_ENABLE_PROJECTS argument, as well as a repository filter.

The getOpenMPCMakeBuildFactory by default uses llvm, clang and openmp as a minimum. Thus removing depends_on_projects only has an effect that lld is not considered anymore. The original patch D106928 uses LLVM_ENABLE_PROJECTS in extraCmakeArgs, which is in conflict the same parameter getOpenMPCMakeBuildFactory set as well. Therefore, I suggested to use depends_on_projects instead. Because it seems they want to use LLVM's own lld and therefore need to enable it. Somehow D107716 removed openmp as a dependent project, which would have cause such that OpenMP changes does not trigger a rebuild.

The CUDA OpenMP buildbots use a pre-installed lld instead (LLVM_ENABLE_LLD), no need to add lld as a dependent project.

rG407e07aa67ab56c92cdec1fdbf6b121afbceddaf changed the runtimes top-level directory that does not belong to any subproject, which is why it did not trigger a rebuild. If you goal is to trigger a build on any commit, this patch doesn't do that. Also consider that if it actually did that, it would also trigger a rebuild on any commit to the repository, such as flang, mlir, lldb, ...

buildbot/osuosl/master/config/builders.py

 # OpenMP AMDGPU Builders
     {'name' : "openmp-offload-amdgpu-project",
     'tags'  : ["openmp"],
     'workernames' : ["omp-vega20-0"],
     'builddir': "openmp-offload-amdgpu-project",
     'factory' : OpenMPBuilder.getOpenMPCMakeBuildFactory(
                         clean=False,
                         enable_runtimes=[],
-                        depends_on_projects=['llvm','clang','lld'],
                         extraCmakeArgs=[
                             "-DCMAKE_BUILD_TYPE=Release",
                             "-DCLANG_DEFAULT_LINKER=lld",
                             "-DLLVM_TARGETS_TO_BUILD=X86;AMDGPU",
                             "-DLLVM_ENABLE_ASSERTIONS=ON",
                             "-DLLVM_ENABLE_RUNTIMES=openmp",
                             ],
                         install=True,