[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED --- Comment #11 from Jakub Jelinek jakub at gcc dot gnu.org 2013-02-08 11:29:40 UTC --- Honza, please be more careful with PR numbers ;) I believe this is fixed now by: Author: hubicka Date: Tue Feb 5 09:11:53 2013 New Revision: 195750 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=195750 Log: PR tree-optimization/r55789 * cgraphclones.c (cgraph_remove_node_and_inline_clones): Remove the dead call anyway. * g++.dg/torture/pr55789.C: New testcase. Added: trunk/gcc/testsuite/g++.dg/torture/pr55789.C Modified: trunk/gcc/ChangeLog trunk/gcc/cgraphclones.c trunk/gcc/testsuite/ChangeLog Author: hubicka Date: Tue Feb 5 09:13:48 2013 New Revision: 195751 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=195751 Log: PR tree-optimization/55789 * g++.dg/tree-ssa/inline-1.C: Update max-inliner-iterations. * g++.dg/tree-ssa/inline-2.C: Update max-inliner-iterations. * g++.dg/tree-ssa/inline-3.C: Update max-inliner-iterations. * g++.dg/ipa/inline-1.C: New testcase. * g++.dg/ipa/inline-2.C: New testcase. * g++.dg/ipa/inline-3.C: New testcase. * params.def (PARAM_EARLY_INLINER_MAX_ITERATIONS): Drop to 1. Added: trunk/gcc/testsuite/g++.dg/ipa/inline-1.C trunk/gcc/testsuite/g++.dg/ipa/inline-2.C trunk/gcc/testsuite/g++.dg/ipa/inline-3.C Modified: trunk/gcc/ChangeLog trunk/gcc/params.def trunk/gcc/testsuite/g++.dg/tree-ssa/inline-1.C trunk/gcc/testsuite/g++.dg/tree-ssa/inline-2.C trunk/gcc/testsuite/g++.dg/tree-ssa/inline-3.C Author: hubicka Date: Tue Feb 5 15:23:56 2013 New Revision: 195758 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=195758 Log: PR tree-optimization/55789 * g++.dg/tree-ssa/inline-1.C: Update max-inliner-iterations. * g++.dg/tree-ssa/inline-2.C: Update max-inliner-iterations. * g++.dg/tree-ssa/inline-3.C: Update max-inliner-iterations. * g++.dg/ipa/inline-1.C: New testcase. * g++.dg/ipa/inline-2.C: New testcase. * g++.dg/ipa/inline-3.C: New testcase. Modified: trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/g++.dg/ipa/inline-1.C trunk/gcc/testsuite/g++.dg/tree-ssa/inline-3.C Author: jakub Date: Thu Feb 7 10:45:12 2013 New Revision: 195844 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=195844 Log: PR tree-optimization/55789 * g++.dg/ipa/inline-3.C: Use cleanup-ipa-dump instead of cleanup-tree-dump. * gcc.dg/tree-ssa/inline-3.c: Add --param max-early-inliner-iterations=2 option. Modified: trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/g++.dg/ipa/inline-3.C trunk/gcc/testsuite/gcc.dg/tree-ssa/inline-3.c
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Jan Hubicka hubicka at gcc dot gnu.org changed: What|Removed |Added Status|NEW |ASSIGNED AssignedTo|unassigned at gcc dot |hubicka at gcc dot gnu.org |gnu.org | --- Comment #5 from Jan Hubicka hubicka at gcc dot gnu.org 2013-01-23 13:28:35 UTC --- Path to disable early inliner iteation is posted to gcc.gnu.org/ml/gcc-patches/2013-01/msg01138.html The ICE happens in IPA inlining rather than in early inliner. I am looking into it.
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 --- Comment #6 from Jan Hubicka hubicka at gcc dot gnu.org 2013-01-23 13:38:16 UTC --- The patch in Comment #4 should not have any effect (and indeed the test does not fire for me on the testcase). can_early_inline predicate already test that the callee is in SSA form and we do into-ssa just before early inlining. So the functions not processed yet in the topological order are not in SSA form.
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #7 from Jakub Jelinek jakub at gcc dot gnu.org 2013-01-23 13:49:17 UTC --- For 4.9, wouldn't it be better to get all functions through the very early passes (up to and including build_ssa (or one or three passes after it, but before pass_inline_parameters)), then in another loop run the rest of early passes (i.e. inline_parameters/einline, ..., eipa_sra, ..., pass_inline_parameters) and then the normal IPA queue? The amount of issues we have with functions not in SSA form yet, whether it is in early inliner, or eipa_sra, etc. is big.
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797
--- Comment #8 from Jan Hubicka hubicka at gcc dot gnu.org 2013-01-23
13:56:44 UTC ---
This is not really issue with early inliner confused with function not being in
SSA form. But for aid of esra, we can do that at expense of increasing of peak
memory use - the SSA form is quite bloated just after built and early passes
gets it noticeably smaller, so we may need to schedule another DCE pass or so.
The problem here is however saving body of
file_info::~file_info() (struct file_info * const this)
{
struct intrusive_ptr * _2;
struct intrusive_ptr * _5;
struct intrusive_ptr * _7;
struct section_info * _9;
bb 2:
_2 = this_1(D)-switched_section;
_9 = _2-px;
if (_9 != 0B)
goto bb 3;
else
goto bb 5 (L2);
bb 3:
section_info::~section_info (_9);
bb 4:
operator delete (_9);
L2:
_5 = this_1(D)-parent;
intrusive_ptrfile_info::~intrusive_ptr (_5);
return;
L1:
_7 = this_1(D)-parent;
intrusive_ptrfile_info::~intrusive_ptr (_7);
resx 1
}
The basic block L1 in unreachable, but for some reason it is not removed
prior inlining. save_function_body must run delete_unreachable_blocks in order
to update SSA after copying and that one gets rid of the call. It updates the
node itself, but not the clones. I will add code to update clones.
I suppose this is because
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 --- Comment #9 from Jan Hubicka hubicka at gcc dot gnu.org 2013-01-23 14:00:23 UTC --- Just for record, I do not recall any issues with early inliner being run in parallel with into-SSA. As a simple inliner working in topological order, it really does not care about functions not processed yet.
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797
--- Comment #10 from Jan Hubicka hubicka at gcc dot gnu.org 2013-01-23
14:19:51 UTC ---
I am testing the following patch. It is a side case where we save function body
but the function we save the body of becomes unnecesary as a result of dead
block removal during inlining that is caused by ipa-pure-const proving function
nothorw.
We get it almost right by removing the unnecesary clone after saving, but we
forget about the edge.
Index: cgraphclones.c
===
--- cgraphclones.c (revision 195370)
+++ cgraphclones.c (working copy)
@@ -570,7 +570,10 @@ cgraph_remove_node_and_inline_clones (st
bool found = false;
if (node == forbidden_node)
-return true;
+{
+ cgraph_remove_edge (node-callers);
+ return true;
+}
for (e = node-callees; e; e = next)
{
next = e-next_callee;
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797
--- Comment #3 from Richard Biener rguenth at gcc dot gnu.org 2013-01-08
13:50:58 UTC ---
Eh, we do totally crazy (recursive) inlining here ...
struct section_info
{
intrusive_ptr section_info parent;
};
struct file_info
{
intrusive_ptr file_info parent;
intrusive_ptr section_info switched_section;
};
so the simple
void
start_file (void)
{
intrusive_ptr file_info parent;
}
creates and destroys the graph of file_info / section_info nodes
with the edges represented by intrusive_ptr's.
void start_file() ()
{
...
bb 2:
_5 = parent.px;
if (_5 != 0B)
goto bb 3;
else
goto bb 1041 (L3);
bb 3:
_6 = _5-switched_section;
_7 = _6-px;
if (_7 != 0B)
goto bb 4;
else
goto bb 6 (L1);
bb 4:
section_info::~section_info (_7);
bb 5:
operator delete (_7);
...
and 1000 calls follow.
I wonder why we need such high early-inlin-insns number and for lower we hit:
else if ((n = num_calls (callee)) != 0
growth * (n + 1) PARAM_VALUE (PARAM_EARLY_INLINING_INSNS))
{
if (dump_file)
fprintf (dump_file, will not early inline: %s/%i-%s/%i,
growth %i exceeds --param early-inlining-insns
divided by number of calls\n,
xstrdup (cgraph_node_name (e-caller)), e-caller-uid,
xstrdup (cgraph_node_name (callee)), callee-uid,
growth);
want_inline = false;
}
of which I cannot make very much sense. Why should the number of calls
in callee(!) times the growth matter? Shouldn't this be the number
of times the caller calls callee? And why even that? We've gone completely
away from the consider only if all calls can be inlined way of early
inline operation!
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797
--- Comment #4 from Richard Biener rguenth at gcc dot gnu.org 2013-01-08
14:24:59 UTC ---
Solution to put off recursive inlining through iteration:
Index: ipa-inline.c
===
--- ipa-inline.c(revision 195014)
+++ ipa-inline.c(working copy)
@@ -1951,7 +1951,9 @@ early_inline_small_functions (struct cgr
if (!can_early_inline_edge_p (e))
continue;
- if (cgraph_edge_recursive_p (e))
+ if (cgraph_edge_recursive_p (e)
+ || !DECL_STRUCT_FUNCTION
+ (callee-symbol.decl)-always_inline_functions_inlined)
{
if (dump_file)
fprintf (dump_file, Not inlining: recursive call.\n);
as we process functions in DFS order any not already processed function
is in a callgraph cycle. This doesn't fix the overzealeous inlining via
the iteration for the testcase though as we still grow in calls quadratically
(we never hit a direct recursive call when inlining the recursive
sub-callgraph).
That is, early inlining completely misses the graph topology hints
(never inline a SCC entry edge, etc.)
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Richard Biener rguenth at gcc dot gnu.org changed: What|Removed |Added Priority|P3 |P1
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Marek Polacek mpolacek at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2012-12-30 CC||mpolacek at gcc dot gnu.org Ever Confirmed|0 |1 --- Comment #1 from Marek Polacek mpolacek at gcc dot gnu.org 2012-12-30 10:54:03 UTC --- Confirmed.
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Marek Polacek mpolacek at gcc dot gnu.org changed: What|Removed |Added CC||hubicka at gcc dot gnu.org --- Comment #2 from Marek Polacek mpolacek at gcc dot gnu.org 2012-12-30 12:16:15 UTC --- Started with http://gcc.gnu.org/viewcvs?view=revisionrevision=193157
[Bug middle-end/55797] [4.8 Regression] ICE: verify_cgraph_node failed: edge has no corresponding call_stmt
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55797 Andrew Pinski pinskia at gcc dot gnu.org changed: What|Removed |Added Target Milestone|--- |4.8.0
