[llvm-commits] [llvm] r144188 - in /llvm/trunk: lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp test/CodeGen/X86/fold-pcmpeqd-0.ll test/CodeGen/X86/multiple-libcalls-and-twoaddr-deps-scheduling.ll

[Duncan Sands]

Hi Evan,

> Are you sure this is the patch that caused the problem? It's disabling an optimization. Unless the patch is somehow introducing non-determinism, I find it hard to believe it's the problem.

I agree it seems pretty unlikely, yet it seems to be the one: the buildbot
recovered on this build:

http://lab.llvm.org:8011/builders/dragonegg-i386-linux/builds/358

The commit list contains one commit: the reversion of Dan's patch.

I speculatively reverted Dan's patch because while it seemed unlikely to
be the culprit, I thought it more likely than the execution domain changes
(since the builder targets i386) or Bill's compact unwind info changes
(as the difference in the object files is due to register allocator changes
and maybe blocks being moved around) which were the other possible culprits.

Ciao, Duncan.

>
> Evan
>
> On Nov 9, 2011, at 1:53 PM, Duncan Sands wrote:
>
>> On 09/11/11 20:11, David Meyer wrote:
>>> :(
>>>
>>> Without this fix, our integration bot asserts while doing bootstrap
>>> build (compiling llvm again with clang).
>>
>> I understand that Dan is fixing something.  Unfortunately it breaks
>> something else, and the LLVM rule is that changes should be reverted
>> in that case.  Hopefully Dan will identify the problem and reapply
>> with a fix.
>>
>> Ciao, Duncan.
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>