Re: [PR67891] drop is_gimple_reg test from set_parm_rtl

[Jeff Law <law@redhat.com>]

On 11/02/2015 06:11 PM, Alexandre Oliva wrote:
> On Oct 14, 2015, Richard Biener <richard.guenther@gmail.com> wrote:
>
>> On Wed, Oct 14, 2015 at 5:25 AM, Alexandre Oliva <aoliva@redhat.com> wrote:
>>> On Oct 12, 2015, Richard Biener <richard.guenther@gmail.com> wrote:
>>>
>>>> On Sat, Oct 10, 2015 at 3:16 PM, Alexandre Oliva <aoliva@redhat.com> wrote:
>>>>> On Oct  9, 2015, Richard Biener <richard.guenther@gmail.com> wrote:
>>>>>
>>>>>> Ok.  Note that I think emit_block_move shouldn't mess with the addressable flag.
>>>>>
>>>>> I have successfully tested a patch that stops it from doing so,
>>>>> reverting https://gcc.gnu.org/bugzilla/show_bug.cgi?id=49429#c11 but
>>>>> according to bugs 49429 and 49454, it looks like removing it would mess
>>>>> with escape analysis introduced in r175063 for bug 44194.  The thread
>>>>> that introduces the mark_addressable calls suggests some discomfort with
>>>>> this solution, and even a suggestion that the markings should be
>>>>> deferred past the end of expand, but in the end there was agreement to
>>>>> go with it.  https://gcc.gnu.org/ml/gcc-patches/2011-06/msg01746.html
>>>
>>>> Aww, indeed.  Of course the issue is that we don't track pointers to the
>>>> stack introduced during RTL properly.
>>>
>>>> Thanks for checking.  Might want to add a comment before that
>>>> addressable setting now that you've done the archeology.
>>>
>>> I decided to give the following approach a try instead.  The following
>>> patch was regstrapped on x86_64-linux-gnu and i686-linux-gnu.
>>> Ok to install?
>
>> It looks ok to me but lacks a comment in mark_addressable_1 why we
>> do this queueing when currently expanding to RTL.
>
>>> +/* Mark X as addressable or queue it up if called during expand.  */
>>> +
>>> +static void
>>> +mark_addressable_1 (tree x)
>
> How about this:
>
> /* Mark X as addressable or queue it up if called during expand.  We
>     don't want to apply it immediately during expand because decls are
>     made addressable at that point due to RTL-only concerns, such as
>     uses of memcpy for block moves, and TREE_ADDRESSABLE changes
>     is_gimple_reg, which might make it seem like a variable that used
>     to be a gimple_reg shouldn't have been an SSA name.  So we queue up
>     this flag setting and only apply it when we're done with GIMPLE and
>     only RTL issues matter.  */
Sounds good.
jeff