[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 Markus Trippelsdorf trippels at gcc dot gnu.org changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #9 from Markus Trippelsdorf trippels at gcc dot gnu.org --- (In reply to Markus Trippelsdorf from comment #8) The original testcase still ICEs on ppc64. Sorry, it actually works fine now.
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 Martin Liška marxin at gcc dot gnu.org changed: What|Removed |Added CC||jiwang at gcc dot gnu.org --- Comment #6 from Martin Liška marxin at gcc dot gnu.org --- *** Bug 63721 has been marked as a duplicate of this bug. ***
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 --- Comment #7 from Martin Liška marxin at gcc dot gnu.org --- Author: marxin Date: Fri Nov 7 13:37:41 2014 New Revision: 217222 URL: https://gcc.gnu.org/viewcvs?rev=217222root=gccview=rev Log: PR ipa/63580 * cgraphunit.c (cgraph_node::create_wrapper): TREE_ADDRESSABLE is set to false for a newly created thunk. * g++.dg/ipa/pr63580.C: New test. Added: trunk/gcc/testsuite/g++.dg/ipa/pr63580.C Modified: trunk/gcc/ChangeLog trunk/gcc/cgraphunit.c trunk/gcc/testsuite/ChangeLog
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 --- Comment #8 from Markus Trippelsdorf trippels at gcc dot gnu.org --- The original testcase still ICEs on ppc64.
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580
--- Comment #4 from Martin Liška marxin at gcc dot gnu.org ---
diff --git a/gcc/cgraphunit.c b/gcc/cgraphunit.c
index 6f61f5c..89c96dc 100644
--- a/gcc/cgraphunit.c
+++ b/gcc/cgraphunit.c
@@ -2342,6 +2342,14 @@ cgraph_node::create_wrapper (cgraph_node *target)
cgraph_edge *e = create_edge (target, NULL, 0, CGRAPH_FREQ_BASE);
+tree arguments = DECL_ARGUMENTS (decl);
+
+while (arguments)
+ {
+ TREE_ADDRESSABLE (arguments) = false;
+ arguments = TREE_CHAIN (arguments);
+ }
+
expand_thunk (false, true);
e-call_stmt_cannot_inline_p = true;
Following patch fixed the issue, boostrap works without any regression.
But I guess it's just a partial fix. There's very similar issue: PR63573.
Thanks,
Martin
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580
Jan Hubicka hubicka at gcc dot gnu.org changed:
What|Removed |Added
CC||hubicka at gcc dot gnu.org
--- Comment #5 from Jan Hubicka hubicka at gcc dot gnu.org ---
Hmm, I think we may get confused by having ADDRESSABLE flag set because the
original function body took its address and it makes sense to clear it as
Martin suggest.
I am bit confused on why this leads to ICE - I think false positive on
ADDRESSABLE flag ought to make cgraphunit to introduce extra copy in:
if (nargs)
for (i = 1, arg = DECL_CHAIN (a); i nargs; i++, arg = DECL_CHAIN
(arg))
{
tree tmp = arg;
if (!is_gimple_val (arg))
{
tmp = create_tmp_reg (TYPE_MAIN_VARIANT
(TREE_TYPE (arg)), arg);
gimple stmt = gimple_build_assign (tmp, arg);
gsi_insert_after (bsi, stmt, GSI_NEW_STMT);
}
vargs.quick_push (tmp);
}
that may later end up on stack and prevent call from being a tail call.
I see that later optimizers may not be able to optimize out the extra copy
simply because it is bit hard to do: one needs to verify that the original
location is dead and possibly may be rewritten by a callee.
I did not find any code that would prevent clearning ADDRESSABLE flag on
parameters, so I think Martin's patch is correct.
Additionally I think it is not safe to set TAILCALL flag if the code above
introduced a local variable. This is becuase setting this flag by hand bypasses
logic in suitable_for_tail_call_opt_p that prevents tail call conversion when
any ADDRESSABLE vars exists.
I did not find any reasons why ADDRESSABLE flag should be set on parameter
besides the usual case that it has address taken, so I tink Martin's patch is
OK and correct fix.
Martin, does it fix the other PR too?
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 --- Comment #2 from marxin at gcc dot gnu.org --- (In reply to Richard Biener from comment #1) You miss to mark p1 addressable in the alias decl (that is, copy TREE_ADDRESSABLE). As I can see, ICF creates thunk to a localalias of foo. At which please I should preserve addressable attribute? Thanks, Martin
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580
--- Comment #3 from Martin Liška mliska at suse dot cz ---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580
Richard Biener rguenth at gcc dot gnu.org changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed||2014-10-20
Ever confirmed|0 |1
--- Comment #1 from Richard Biener rguenth at gcc dot gnu.org ---
You miss to mark p1 addressable in the alias decl (that is, copy
TREE_ADDRESSABLE).
Do you mean following change:
@@ -2334,6 +2334,14 @@ cgraph_node::create_wrapper (cgraph_node *target)
cgraph_edge *e = create_edge (target, NULL, 0, CGRAPH_FREQ_BASE);
+ tree arguments = DECL_ARGUMENTS (decl);
+
+ while (arguments)
+{
+ TREE_ADDRESSABLE (arguments) = false;
+ arguments = TREE_CHAIN (arguments);
+}
+
expand_thunk (false, true);
e-call_stmt_cannot_inline_p = true;
Thanks,
Martin
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 Richard Biener rguenth at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2014-10-20 Ever confirmed|0 |1 --- Comment #1 from Richard Biener rguenth at gcc dot gnu.org --- You miss to mark p1 addressable in the alias decl (that is, copy TREE_ADDRESSABLE).
[Bug ipa/63580] [5 Regression] ICE : error: invalid argument to gimple call
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63580 Markus Trippelsdorf trippels at gcc dot gnu.org changed: What|Removed |Added Target Milestone|--- |5.0
