This is an archive of the discontinued LLVM Phabricator instance.

Differential D75105

mlir-nvidia builder: using traget X86 instead of host
AbandonedPublic

Authored by kuhnel on Feb 25 2020, 12:27 AM.

Details

Reviewers
gkistanova
Summary

Target "host" works from the command line but not from buildbot.
Not sure why that is.

Diff Detail

Event Timeline

kuhnel created this revision.Feb 25 2020, 12:27 AM
Herald added a subscriber: rriddle. · View Herald TranscriptFeb 25 2020, 12:27 AM
kuhnel added a comment.Feb 25 2020, 12:28 AM

@gkistanova can you please land this one again?
Thx!

Harbormaster completed remote builds in B47185: Diff 246381.Feb 25 2020, 12:32 AM
gkistanova added a comment.Mar 3 2020, 7:34 PM

Hi Christian.

It looks like the problem is in the quotation marks, not in using host target.
This patch wouldn't do either.

I could take a look if you will connect your bot to the staging (port 9994), let me know.

kuhnel added a comment.Mar 19 2020, 3:43 AM

Hi @gkistanova,

sorry for the late reply, I was out of office for a while an it hasn't gotten much better recently. I just reconfigured mlir-nvidia to port 9994.
Let me know what I need to do to get the bot working.

Thx,
Christian

gkistanova added a comment.EditedMar 19 2020, 8:42 PM

Hi Christian,

The https://github.com/llvm/llvm-zorg/commit/fbf5b2d2a35622f6a8758d0273f8ef2222cc8e76 takes care of the cmake params.

Now the bot has a few failing tests you may want to take care of before moving it back to the production:
http://lab.llvm.org:8014/builders/mlir-nvidia/builds/45/steps/cmake-configure/logs/stdio

Testing Time: 78.25s
********************
Failing Tests (6):
    MLIR :: mlir-cuda-runner/all-reduce-and.mlir
    MLIR :: mlir-cuda-runner/all-reduce-max.mlir
    MLIR :: mlir-cuda-runner/all-reduce-min.mlir
    MLIR :: mlir-cuda-runner/all-reduce-or.mlir
    MLIR :: mlir-cuda-runner/all-reduce-xor.mlir
    MLIR :: mlir-cuda-runner/multiple-all-reduce.mlir

  Expected Passes    : 21511
  Expected Failures  : 67
  Unsupported Tests  : 15435
  Unexpected Failures: 6
FAILED: CMakeFiles/check-all
gkistanova added a comment.Mar 28 2020, 5:33 PM

Hi Christian,

http://lab.llvm.org:8014/builders/mlir-nvidia is reliably green on the staging.

Please feel free to move it to the production buildbot. There is no need in this patch.

Thanks

Galina

kuhnel abandoned this revision.Mar 29 2020, 11:53 PM

@gkistanova thx for your help. The bot is back in production!

Revision Contents

PathSize
buildbot/
osuosl/
master/
config/
2 lines

Diff 246381

buildbot/osuosl/master/config/builders.py

Show First 20 Lines • Show All 1,027 Lines • ▼ Show 20 Lines return [
'factory': UnifiedTreeBuilder.getCmakeWithNinjaBuildFactory( 'factory': UnifiedTreeBuilder.getCmakeWithNinjaBuildFactory(
llvm_srcdir="llvm.src", llvm_srcdir="llvm.src",
obj_dir="llvm.obj", obj_dir="llvm.obj",
clean=True, clean=True,
depends_on_projects=['llvm','mlir'], depends_on_projects=['llvm','mlir'],
extra_configure_args=[ extra_configure_args=[
'-DLLVM_BUILD_EXAMPLES=ON', '-DLLVM_BUILD_EXAMPLES=ON',
'-DLLVM_ENABLE_CXX1Y=Y', '-DLLVM_ENABLE_CXX1Y=Y',
'-DLLVM_TARGETS_TO_BUILD="host;NVPTX"', '-DLLVM_TARGETS_TO_BUILD="X86;NVPTX"',
'-DLLVM_ENABLE_PROJECTS=mlir', '-DLLVM_ENABLE_PROJECTS=mlir',
'-DMLIR_CUDA_RUNNER_ENABLED=1', '-DMLIR_CUDA_RUNNER_ENABLED=1',
'-DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc', '-DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc',
], ],
env={ env={
'CC':'clang', 'CC':'clang',
'CXX': 'clang++', 'CXX': 'clang++',
'LD': 'lld', 'LD': 'lld',
▲ Show 20 Lines • Show All 554 Lines • Show Last 20 Lines