[llvm] r308078 - [ThinLTO] Ensure we always select the same function copy to import

Teresa Johnson via llvm-commits llvm-commits at lists.llvm.org
Sat Jul 15 07:34:50 PDT 2017


Thanks. I'm about 99.9% sure I know the cause, it was a big oversight that
I needed to disable multiple threads since I am parsing the ThinLTO backend
messages.
Teresa

On Sat, Jul 15, 2017 at 7:30 AM, Chandler Carruth <chandlerc at gmail.com>
wrote:

> On Sat, Jul 15, 2017 at 9:39 AM Chandler Carruth <chandlerc at gmail.com>
> wrote:
>
>> On Sat, Jul 15, 2017 at 12:53 AM Teresa Johnson via llvm-commits <
>> llvm-commits at lists.llvm.org> wrote:
>>
>>> Author: tejohnson
>>> Date: Fri Jul 14 21:53:05 2017
>>> New Revision: 308078
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=308078&view=rev
>>> Log:
>>> [ThinLTO] Ensure we always select the same function copy to import
>>>
>>> Summary:
>>> Check if the first eligible callee is under the instruction threshold.
>>> Checking this on the first eligible callee ensures that we don't end
>>> up selecting different callees to import when we invoke this routine
>>> with different thresholds due to reaching the callee via paths that
>>> are shallower or hotter (when there are multiple copies, i.e. with
>>> weak or linkonce linkage). We don't want to leave the decision of which
>>> copy to import up to the backend.
>>>
>>> Reviewers: mehdi_amini
>>>
>>> Subscribers: inglorion, fhahn, llvm-commits
>>>
>>> Differential Revision: https://reviews.llvm.org/D35436
>>>
>>> Added:
>>>     llvm/trunk/test/Transforms/FunctionImport/Inputs/
>>> funcimport_resolved1.ll
>>>     llvm/trunk/test/Transforms/FunctionImport/Inputs/
>>> funcimport_resolved2.ll
>>>     llvm/trunk/test/Transforms/FunctionImport/funcimport_resolved.ll
>>>
>>
>> This newly added test appears to be flaky, showing some nondeterminism in
>> the system. For example, this command will run the tests 100 times, and
>> doing this I see roughly 5 failures:
>>
>> % repeat 100 ./dev/bin/llvm-lit -s ../test/Transforms/FunctionImport
>>
>> The build bots have been failing and then recovering because of this for
>> some time, so I'm going to revert until we can get this fixed to lessen the
>> build bot spam.
>>
>
> Reverted in r308095, let me know if i can help tracking it down.
>



-- 
Teresa Johnson |  Software Engineer |  tejohnson at google.com |  408-460-2413
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170715/224a1a1d/attachment.html>


More information about the llvm-commits mailing list