[llvm-commits] [PATCH v2] Disable old-JIT and enable MCJIT tests on PowerPC

[Kaylor, Andrew]

That looks great, and I agree that the directory level skipping is better.  There may be some problems with 32-bit PowerPC and the MCJIT tests, but I don't believe any of the buildbots use that configuration.

Thanks, Ulrich.

-Andy

-----Original Message-----
From: Ulrich Weigand [mailto:Ulrich.Weigand at de.ibm.com] 
Sent: Monday, November 19, 2012 8:59 AM
To: Kaylor, Andrew; rdivacky at freebsd.org
Cc: Adhemerval Zanella; llvm-commits at cs.uiuc.edu
Subject: [PATCH v2] Disable old-JIT and enable MCJIT tests on PowerPC


Ulrich Weigand/Germany/IBM wrote on 19.11.2012 17:23:58:
> "Kaylor, Andrew" <andrew.kaylor at intel.com> wrote:
>
> > The buildbots are currently skipping the MCJIT ExecutionEngine tests 
> > on PowerPC64 and the legacy JIT ExecutionEngine tests are all failing.
> >
> > You can enable the MCJIT tests by modifying llvm/test/ 
> > ExecutionEngine/MCJIT/lit.local.cfg.  I think it's OK to add PowerPC 
> > to the XFAIL list for the legacy JIT tests.
> >
> > I added a few new tests to the MCJIT suite today to verify lli's 
> > remote-mcjit option, and it's possible that those will require some 
> > minor updates to the relocation handling.
>
> The attached patch implements the above suggestion (enable MCJIT tests 
> on PowerPC; XFAIL old-JIT tests on PowerPC).  Is this what you had in 
> mind?
>
> All MCJIT tests (as of rev. 168312) pass on PowerPC now.  (In fact, 
> with this patch, the test suite as a whole passes.)

Roman Divacky pointed out on IRC that it might be preferable to disable all old-JIT tests at the directory level instead of individually XFAILing them.  This alternative patch implements this suggestion.  All MCJIT tests still pass.

Either version OK to commit?

Bye,
Ulrich

(See attached file: diff-llvm-jit-test-v2)