[zorg] Add buildbot for HIP

[ashi1]

Summary

Build HIP tests from llvm-test-suite using AMDGPU.
There will be an upcoming patch to add HIP test to llvm-test-suite/External/HIP.

[ashi1]

buildbot/osuosl/master/config/builders.py

1697

@gkistanova - is it okay to temporarily use an external script, while we are bringing up the build? Or does it need to go into zorg first? Thank you.

[gkistanova]

Everything else looks Ok.

buildbot/osuosl/master/config/builders.py

1697

This is fine, but please provide here a real path where the script will be placed to when ready. Just have it as a local uncommitted patch in the builder source tree for now.

[ashi1]

This HIP builder is dependent on two other patches:

1. hip-build.sh script found at https://reviews.llvm.org/D100060
2. llvm-test-suite - adding HIP tests found at https://reviews.llvm.org/D99997

buildbot/osuosl/master/config/builders.py

1697

That sounds good. I've opened a patch for that script (although still under-development):
https://reviews.llvm.org/D100060

[ashi1]

buildbot/osuosl/master/config/builders.py

1697

AnnotatedBuilder.py seems to enforce running python on scripts in relative local paths. Should I wrap my bash script with a python script so that it can run the bash script? I noticed cuda also using full path /buildbot/cuda-build outside of zorg. Thanks.

[ashi1]

@gkistanova , is it okay to submit this with a full system path temporarily? /buildbot/hip-build.sh? Or do we have to switch to a python script?

[gkistanova]

No. Cuda builders got that in the code base as a temporary solution while they are working on their annotated script. Exactly the same arguments as yours. And I see that being confusing, and don't think it worth it after all.

To get it working in the production you need to add the support for shell scripts anyway. With that in place you could put your script in the right place and keep working on that while your builder is staged. It seems the right thing to do.

Please feel free to ask if you have questions or will need help with adding the shell script support.

[ashi1]

In D99894#2693229, @gkistanova wrote:

No. Cuda builders got that in the code base as a temporary solution while they are working on their annotated script. Exactly the same arguments as yours. And I see that being confusing, and don't think it worth it after all.

To get it working in the production you need to add the support for shell scripts anyway. With that in place you could put your script in the right place and keep working on that while your builder is staged. It seems the right thing to do.

Please feel free to ask if you have questions or will need help with adding the shell script support.

Hi Galina, I've opened a small patch that should execute .py scripts with python, and all other scripts directly. Is there any public buildbot test that ensures the change works, I won't have access to all the various environments using AnnotatedBuilder.py. Thank you.
https://reviews.llvm.org/D100666

[ashi1]

Hi @gkistanova, I've changed the buildbot to use a local relative path (in annotated/hip-build.sh dir).
I will keep it as a local uncommitted script during bring-up. D100666 was submitted, and script_interpreter="python" is disabled for this bash script.

[gkistanova]

Looks good with a nitpick. Please see my comment inline.

I will keep it as a local uncommitted script during bring-up.

This is fine, but please make sure everything is complete and committed before you will be ready to move the bot to the production.

buildbot/osuosl/master/config/builders.py

1699

How about script_interpreter=None instead?

[ashi1]

Revised to @gkistanova 's comments.

[gkistanova]

LGTM

buildbot/osuosl/master/config/builders.py

'factory' : UnifiedTreeBuilder.getCmakeWithNinjaBuildFactory(
                         "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=VE",
                         "-DLLVM_ENABLE_RUNTIMES=compiler-rt",
                         "-DLLVM_BUILD_LLVM_DYLIB=On",
                         "-DBUILD_SHARED_LIBS=Off",
                         "-DLLVM_LINK_LLVM_DYLIB=On",
                         "-DCLANG_LINK_CLANG_DYLIB=On",
                     ])},
 
+# HIP builders.
+
+    {'name' : "clang-hip-vega20",
+    'tags'  : ["clang"],
+    'workernames' : ["hip-vega20-0"],
+    'builddir': "clang-hip-vega20",
+    'factory' : AnnotatedBuilder.getAnnotatedBuildFactory(
+                    script="hip-build.sh",
+                    checkout_llvm_sources=False,
+                    script_interpreter=None)},
+
 # Latest stable fedora running on Red Hat internal OpenShift cluster (PSI).
 
     {'name' : 'x86_64-fedora-clang',
     'tags'  : ['mlir'],
     'collapseRequests': False,
     'workernames': ['fedora-llvm-x86_64'],
     'builddir': 'x86_64-fedora-clang',
     'factory' : UnifiedTreeBuilder.getCmakeWithNinjaBuildFactory(

buildbot/osuosl/master/config/status.py

reporters.MailNotifier(
         sendToInterestedUsers = False,
         extraRecipients=[
             "[e-mail]"],
         subject = "MLIR Build Failure: %(builder)s",
         mode = "failing",
         builders = ["mlir-nvidia",
                     "mlir-windows",
                     "ppc64le-mlir-rhel-clang"]),
+    reporters.MailNotifier(
+        fromaddr="[e-mail]",
+        sendToInterestedUsers = False,
+        extraRecipients=["[e-mail]"],
+        subject = "Build Failure: %(builder)s",
+        mode = "failing",
+        builders = ["clang-hip-vega20"]),
 ]
 
 
 from twisted.python import log
 
 # Returns a list of Status Targets. The results of each build will be
 # pushed to these targets. buildbot.plugins reporters has a variety
 # to choose from, including email senders, and IRC bots.
 def getReporters():
     if config.options.getboolean('Master Options', 'is_production'):
         log.msg(">>> getReporters: Production mode. All reporters are registered.")
         return all
     else:
         # Staging buildbot does not report issues.
         log.msg(">>> getReporters: Staging buildbot does not report issues.")
         return []

buildbot/osuosl/master/config/workers.py

return [
         # Windows 10 Pro x86_64-scei-ps4, 2 x Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
         create_worker("as-worker-93", properties={'jobs': 36}, max_builds=1),
 
         # WIP migration of the CUDA buildbot to GCE.
         create_worker("cuda-k80-0", max_builds=1),
         create_worker("cuda-p4-0", max_builds=1),
         create_worker("cuda-t4-0", max_builds=1),
 
+        # HIP on Ubuntu 18.04.5,  Intel(R) Xeon(R) Gold 5218 @ 2.30GHz, Vega20 GPU
+        create_worker("hip-vega20-0", max_builds=1),
+
         # X86_64 AVX2, Ubuntu 16.04.2, Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz
         create_worker("avx2-intel64", properties={'jobs': 80}, max_builds=1),
 
         # X86_64 with SDE, Ubuntu 16.04.2, Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz
         create_worker("sde-avx512-intel64", properties={'jobs': 80}, max_builds=1),
 
         # Debian 9.0 x86_64 64-core GCE instances
         create_worker("fuchsia-debian-64-us-central1-a-1", properties={'jobs': 64}, max_builds=1),