Re: [PATCH] IPA: do not release body if still needed

[Martin Jambor <mjambor@suse.cz>]

Hi,

On Mon, Jan 16 2023, Martin Liška wrote:
> On 1/14/23 22:36, Jan Hubicka wrote:
>>> Noticed during building of libbackend.a with the LTO partial linking.
>>>
>>> The function release_body is called even if clone_of is a clone
>>> of a another function and thus it shares tree declaration. We should
>>> preserve it in that situation.
>>>
>>> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
>>>
>>> Ready to be installed?
>>> Thanks,
>>> Martin
>>>
>>> 	PR ipa/107944
>>>
>>> gcc/ChangeLog:
>>>
>>> 	* cgraph.cc (cgraph_node::remove): Do not release body
>>> 	if a node is clone of another node.
>>> ---
>>>  gcc/cgraph.cc | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/gcc/cgraph.cc b/gcc/cgraph.cc
>>> index f15cb47c8b8..2e7d77ffd6c 100644
>>> --- a/gcc/cgraph.cc
>>> +++ b/gcc/cgraph.cc
>>> @@ -1893,7 +1893,7 @@ cgraph_node::remove (void)
>>>    else if (clone_of)
>>>      {
>>>        clone_of->clones = next_sibling_clone;
>>> -      if (!clone_of->analyzed && !clone_of->clones && !clones)
>>> +      if (!clone_of->analyzed && !clone_of->clones && !clones && !clone_of->clone_of)
>>>  	clone_of->release_body ();
>> 
>> It is interesting that the problem reproduced only after almost 20
>> years.  But I suppose it is because we materialize clones in parituclar
>> order.
>
> Well, it started with r13-48-g27ee75dbe81bb7 where Martin add a new code
> that calls the release_body function. So it's pretty new.
>
>> 
>> I think there are two ways to fix it.  Either declare release_body to be
>> applicable only to the master clone and avoid calling it here (as you
>> do) or make release_body do nothing when called on a clone.
>> I guess it makes sense to keep your approach but please add sanity check
>> to release_body that clone_of == NULL with a comment.
>
> I do support Martin's enhanced version of the patch.
>

I take that as an approval, so I am about to commit the following after
re-testing it on trunk.  Afterwards I'll backport it to the affected
release branches too.

Thanks,

Martin


The code removing function bodies when the last call graph clone of a
node is removed is too aggressive when there are nodes up the
clone_of chain which still need them.  Fixed by expanding the check.

gcc/ChangeLog:

2023-01-18  Martin Jambor  <mjambor@suse.cz>

	PR ipa/107944
	* cgraph.cc (cgraph_node::remove): Check whether nodes up the
	lcone_of chain also do not need the body.
---
 gcc/cgraph.cc | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/gcc/cgraph.cc b/gcc/cgraph.cc
index 5e60c2b73db..5f72ace9b57 100644
--- a/gcc/cgraph.cc
+++ b/gcc/cgraph.cc
@@ -1893,8 +1893,18 @@ cgraph_node::remove (void)
   else if (clone_of)
     {
       clone_of->clones = next_sibling_clone;
-      if (!clone_of->analyzed && !clone_of->clones && !clones)
-	clone_of->release_body ();
+      if (!clones)
+	{
+	  bool need_body = false;
+	  for (cgraph_node *n = clone_of; n; n = n->clone_of)
+	    if (n->analyzed || n->clones)
+	      {
+		need_body = true;
+		break;
+	      }
+	  if (!need_body)
+	    clone_of->release_body ();
+	}
     }
   if (next_sibling_clone)
     next_sibling_clone->prev_sibling_clone = prev_sibling_clone;
-- 
2.39.0