Bump to llvm commit 8b76aea8d8b1b71f6220bc2845abc749f18a19b7

[flemairen6]

Green LLVM commit: 8b76aea8d8b1b71f6220bc2845abc749f18a19b7
Green MHLO commit: cae5349f85b3b2f58d80ca788e9a8695d896c3a8

[jenkins-droid]

Can one of the admins verify this patch?

[jenkins-droid]

Can one of the admins verify this patch?

[philass]

@jenkins-droid test this please

[jenkins-droid]

Can one of the admins verify this patch?

[jenkins-droid]

Can one of the admins verify this patch?

[maxbartel]

@jenkins-droid test this please

[jenkins-droid]

Can one of the admins verify this patch?

[maxbartel]

@jenkins-droid test this please

[ttjost]

@AlexandreEichenberger I am seeing an issue with the powerpc pipeline that relates to a recent change in llvm-exegesis. Therefore, we don't even build LLVM for it. =(
LLVM buildbots for ppc pass the test. Do you have any idea how to solve this?
The error says that __builtin_thread_pointer' is not supported on this target, but I don't have a way to reproduce this.
Thank you!

[philass]

Ahh that looks really frustrating.

I wonder if we can play with CMake flags to not build with llvm-exegesis

[ttjost]

Ahh that looks really frustrating.

I wonder if we can play with CMake flags to not build with llvm-exegesis

I tried doing that. This led to disabling llvm-mca, obj2yaml, and yaml2obj, but lit tests all seem to depend on them, and I get a bunch of CMake dependency errors between these tools and tests.

In any case, it turns out __builtin_thread_pointer is not supported by gcc (https://godbolt.org/z/chEnh7Ton), but works with clang (https://godbolt.org/z/c9bdTqv5f), that is why LLVM buildbots for PPC work and onnx-mlir pipelines don't.

I am somewhat blocked now, since this only happens on PPC + GCC (it seems), and everything else is green.
Should we try building an in-tree clang for it?
Could you or anyone check this, please?

[sorenlassen]

maybe the problem has to do with this llvm code change:
llvm/llvm-project@46b5b85

but I can't think of any workaround, except build with clang as has been suggested

[AlexandreEichenberger]

@gongsu832

The LLVM bots succeed in building LLVM but we are not, for PPC. Reported error:

LLVM buildbots for ppc pass the test. Do you have any idea how to solve this?

Can you look into this? Is it because we use GCC to build LLVM and they are not?

[AlexandreEichenberger]

@flemairen6 @ttjost What about the s390 error, is it unrelated?

[jenkins-droid]

Can one of the admins verify this patch?

[sorenlassen]

@jenkins-droid test this please

[sorenlassen]

@ttjost please sign-off on the last commit to fix the DCO failure

I'm not 100% sure why the jenkins pipelines fail with ERROR: Exception: java.lang.Exception: admin authentication failed, maybe it's blocked by the DCO failure?

[jenkins-droid]

Can one of the admins verify this patch?

[sorenlassen]

@jenkins-droid test this please

[sorenlassen]

@AlexandreEichenberger can you tell why the jenkins pipelines are failing with ERROR: Exception: java.lang.Exception: admin authentication failed?

update: things rebuilt without this failure once I pushed a merge commit to update the branch

[jenkins-droid]

Can one of the admins verify this patch?

[maxbartel]

@jenkins-droid test this please

[gongsu832]

/workdir/llvm-project/llvm/tools/llvm-exegesis/lib/BenchmarkRunner.cpp: In member function 'llvm::Error llvm::exegesis::{anonymous}::SubProcessFunctionExecutorImpl::createSubProcessAndRunBenchmark(llvm::StringRef, llvm::SmallVectorImpl<long int>&) const':
/workdir/llvm-project/llvm/tools/llvm-exegesis/lib/BenchmarkRunner.cpp:336:61: error: '__builtin_thread_pointer' is not supported on this target
  336 |         syscall(SYS_rseq, (intptr_t)__builtin_thread_pointer() + __rseq_offset,
      |                                     ~~~~~~~~~~~~~~~~~~~~~~~~^~

Apparently ppc64le does not support __builtin_thread_pointer

[AlexandreEichenberger]

@gongsu832

Apparently ppc64le does not support __builtin_thread_pointer

the statement was made that the ppc build in LLVM is succesfull, so either they skip this target or do something different?

[maxbartel]

LGTM, the clang PR is also already up #2379

[maxbartel]

@AlexandreEichenberger @philass I will merge this if the pipelines are green, the clang fix landed

[jenkins-droid]

Jenkins Linux s390x Build #12026 [push] Bump to llvm commit 8b76... started at 08:53

[jenkins-droid]

Jenkins Linux ppc64le Build #11020 [push] Bump to llvm commit 8b76... started at 08:59

[jenkins-droid]

Jenkins Linux amd64 Build #12013 [push] Bump to llvm commit 8b76... started at 07:53

[AlexandreEichenberger]

@flemairen6 big thanks, much appreciated
@gongsu832 thanks also for your quick fix, worked perfectly

[jenkins-droid]

Jenkins Linux amd64 Build #12013 [push] Bump to llvm commit 8b76... passed after 2 hr 2 min

[jenkins-droid]

Jenkins Linux s390x Build #12026 [push] Bump to llvm commit 8b76... passed after 3 hr 8 min

[jenkins-droid]

Jenkins Linux ppc64le Build #11020 [push] Bump to llvm commit 8b76... passed after 3 hr 15 min