r276977 - [OpenMP] Codegen for use_device_ptr clause.

[Hans Wennborg via cfe-commits]

It seems the compiler was getting thrown off by the in-class
initializers of the local MapInfo struct. I've tried to work around
this in r277256. Hopefully this will make the buildbots happier.

Cheers,
Hans

On Fri, Jul 29, 2016 at 2:20 PM, Samuel F Antao <sfantao at us.ibm.com> wrote:
> Hi Hans,
>
> Thanks for the troubleshooting. I'd prefer to avoid the 'this->', but if
> Alexey is okay with that I wouldn't mind. The best solution would be to
> update the compiler of the bot if there is a newer version and see if the
> issue goes away, but not sure if that is possible at all.
>
> Thanks again,
> Samuel
>
>
> ----- Original message -----
> From: Hans Wennborg <hans at chromium.org>
> Sent by: hwennborg at google.com
> To: Samuel F Antao/Watson/IBM at IBMUS
> Cc: Mike Aizatsky <aizatsky at google.com>, cfe-commits
> <cfe-commits at lists.llvm.org>, "Bataev, Alexey" <a.bataev at hotmail.com>
> Subject: Re: r276977 - [OpenMP] Codegen for use_device_ptr clause.
> Date: Fri, Jul 29, 2016 5:04 PM
>
> I agree this looks like a compiler bug, but I'm not sure what's
> causing it exactly. Putting "this->" in front of those accesses to
> CurDir and CGF seems to fix it, but it's also very ugly.
>
> I tried a couple of other things, like moving the method out-of-line,
> but no luck yet.
>
> On Fri, Jul 29, 2016 at 1:45 PM, Samuel F Antao <sfantao at us.ibm.com> wrote:
>> I don't have easy access to Windows machine to test this, but it seems
>> that
>> whichever compiler is being used is broken. This is about a member
>> function
>> accessing a variable of the same class. I don't think I am doing any
>> non-portable code pattern here.
>>
>> Alexey, do you have a easy way to test this? Can you give me a little help
>> here?
>>
>> Thanks!
>> Samuel
>>
>>
>> ----- Original message -----
>> From: Hans Wennborg <hans at chromium.org>
>> Sent by: hwennborg at google.com
>> To: Samuel F Antao/Watson/IBM at IBMUS
>> Cc: Mike Aizatsky <aizatsky at google.com>, cfe-commits
>> <cfe-commits at lists.llvm.org>
>> Subject: Re: r276977 - [OpenMP] Codegen for use_device_ptr clause.
>> Date: Fri, Jul 29, 2016 3:53 PM
>>
>> Yes, it still appears to be broken:
>>
>>
>> http://lab.llvm.org:8011/builders/sanitizer-windows/builds/26504/steps/run%20tests/logs/stdio
>>
>> On Thu, Jul 28, 2016 at 1:30 PM, Samuel F Antao via cfe-commits
>> <cfe-commits at lists.llvm.org> wrote:
>>> Hi Mike,
>>>
>>> I've already pushed r276981 and r276988 to fix those failures shortly
>>> after
>>> I pushed the patch. After that I didn't receive any bot message
>>> complaining
>>> about that code.
>>>
>>> Do you still see the issues?
>>>
>>> Thanks,
>>> Samuel
>>>
>>>
>>> ----- Original message -----
>>> From: Mike Aizatsky <aizatsky at google.com>
>>> To: Samuel F Antao/Watson/IBM at IBMUS, cfe-commits at lists.llvm.org
>>> Cc:
>>> Subject: Re: r276977 - [OpenMP] Codegen for use_device_ptr clause.
>>> Date: Thu, Jul 28, 2016 4:20 PM
>>>
>>> Samuel,
>>>
>>> I think your change breaks the build:
>>>
>>> http://lab.llvm.org:8011/builders/sanitizer-windows/builds/26397
>>>
>>> FAILED: