ppc64 clang LNT buildbot / reference_output / test failures

[Hal Finkel]

----- Original Message -----
> From: "Daniel Dunbar" <daniel at zuster.org>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: "will schmidt" <will_schmidt at vnet.ibm.com>, "Michael Gottesman" <mgottesman at apple.com>, "Galina Kistanova"
> <gkistanova at gmail.com>, "Commit Messages and Patches for LLVM" <llvm-commits at cs.uiuc.edu>
> Sent: Thursday, April 25, 2013 2:33:05 PM
> Subject: Re: ppc64 clang LNT buildbot / reference_output / test failures
> 
> 
> Hi,
> 
> 
> I believe that almost all of the work to make reference outputs
> portable is done. I believe the current status is that the reference
> outputs largely work across Darwin x {x86, ARM, PPC} and Linux x {
> x86?, ARM }, which makes it increasingly unlikely that the reference
> outputs are incorrect for your platform.

But running without reference outputs vs. gcc yields no failures, so I'd tend to conclude that the problem is with the reference outputs.

> 
> 
> There isn't currently a way from the LNT command line to disable the
> use of reference outputs, and I personally would prefer this serve
> as a forcing function to fix them in the few remaining places they
> don't work.

I understand, but this is a bad rationale. We'll be branching for 3.3 soon, and having buildbot feedback is an important part of the quality control system. We have multiple ppc64 buildbots, and we can certainly keep some with reference outputs and some without.

Thanks again,
Hal

> 
> 
> - Daniel
> 
> 
> 
> On Fri, Apr 19, 2013 at 4:03 PM, Hal Finkel < hfinkel at anl.gov >
> wrote:
> 
> 
> Michael, Galina, (or anyone else),
> 
> Can you please provide some guidance here? I know that people are
> working on fixing up the reference outputs so that they work across
> different platforms, however, in the mean time, we'd like to have
> (at least some of) the PPC64/Linux buildbots reflect the true state
> of the compiler. We can do this if we can turn off using the
> reference outputs. How can this be done?
> 
> Thanks again,
> Hal
> 
> 
> 
> ----- Original Message -----
> > From: "Will Schmidt" < will_schmidt at vnet.ibm.com >
> > To: llvm-commits at cs.uiuc.edu
> > Sent: Tuesday, February 12, 2013 3:29:29 PM
> > Subject: ppc64 clang LNT buildbot / reference_output / test
> > failures
> > 
> > Hi All,
> > 
> > I've been continuing to poke at the ppc64-elf-linux LNT/clang
> > buildbot,
> > and think I now understand part of whats causing the current
> > failures,
> > but am not sure how to properly resolve.
> > 
> > A number of the tests indicate FAILED, with output such as:
> > 
> > /home/clangbuild/clang-ppc64-1/test-suite/DiffOutput.sh
> > "/home/clangbuild/clang-ppc64-1/tests/nt/build/sample-0/tools/fpcmp
> > -r
> > 0.002" simple alti.stepfft
> > ******************** TEST (simple) 'alti.stepfft' FAILED!
> > ********************
> > Execution Context Diff:
> > /home/clangbuild/clang-ppc64-1/tests/nt/build/sample-0/tools/fpcmp:
> > FP Comparison failed, not a numeric difference between 'e' and 'f'
> > 
> > 
> > In this and most(?) of the other failure cases, there is no attempt
> > to build or run
> > the native version of the testcase. The result comparison is
> > occurring between the
> > output of testcase.simple and the contents of
> > test-suite/default.reference_output
> > which contains "exit 0". This is in contrast to when we are running
> > the test-suite by
> > hand, and seeing both .native and .simple versions of each test.
> > 
> > On the surface, this seems to be related to USE_REFERENCE_OUTPUT=1
> > (set by lnt/nt.py ),
> > but a local hack to set that to zero is not sufficient. Nor is
> > resetting
> > USE_REFERENCE_OUTPUT:=0 in Makefile.programs.
> > 
> > ... ideas ?
> > 
> > Thanks,
> > -Will
> > 
> > 
> > _______________________________________________
> > llvm-commits mailing list
> > llvm-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> > 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
>