[llvm] r218101 - Optimize sext/zext insertion algorithm in back-end.

Kostya Serebryany kcc at google.com
Wed Sep 24 16:26:54 PDT 2014


On Wed, Sep 24, 2014 at 4:25 PM, Bob Wilson <bob.wilson at apple.com> wrote:

> I don’t think it would have shown up as a leak. The bad code was adding
> values to a DenseMap and never clever them out between functions. I don’t
> see how lsan could have detected that.
>

Yep, no way. We would need a memory profiler and compare the heap profiles
between revisions to detect spikes.


>
> On Sep 24, 2014, at 4:23 PM, Kostya Serebryany <kcc at google.com> wrote:
>
>
>
> On Wed, Sep 24, 2014 at 4:21 PM, Eric Christopher <echristo at gmail.com>
> wrote:
>
>> On Wed, Sep 24, 2014 at 4:20 PM, Kostya Serebryany <kcc at google.com>
>> wrote:
>> >
>> >
>> > On Wed, Sep 24, 2014 at 2:47 PM, Eric Christopher <echristo at gmail.com>
>> > wrote:
>> >>
>> >> On Wed, Sep 24, 2014 at 1:16 AM, Renato Golin <renato.golin at linaro.org
>> >
>> >> wrote:
>> >> > On 23 September 2014 18:56, Bob Wilson <bob.wilson at apple.com> wrote:
>> >> >> We’re seeing significant regressions in memory usage. I think you
>> >> >> forgot to clear the PreferredExtendType DenseMap that you added
>> here.
>> >> >
>> >> > Seems we're missing some valgrind / sanitizer buildbots to check that
>> >> > automatically?
>> >> >
>> >>
>> >> http://lab.llvm.org:8011/builders/clang-x86_64-linux-vg
>> >>
>> >> which appears to be having problems.
>> >>
>> >> I know that Kostya has some sanitizer bots, but I'm not sure what the
>> >> status is there.
>> >
>> >
>> > We have two bots:
>> > http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux
>> > http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/
>> > The second one is red at the moment with some (fresh?) msan reports.
>> > What is the problem here?
>> >
>>
>> Aha.
>>
>> Anyhow as of r218101 we were leaking quite a bit of memory. I don't
>> know if the sanitizers reported it or not.
>>
>
> lsan is silent. repro?
>
>
>
>>
>> -eric
>>
>> >>
>> >>
>> >> -eric
>> >
>> >
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140924/c30a60dc/attachment.html>


More information about the llvm-commits mailing list