[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #22 from aoliva at gcc dot gnu dot org 2007-10-15 17:07 --- Subject: Bug 33572 Author: aoliva Date: Mon Oct 15 17:07:20 2007 New Revision: 129356 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129356 Log: gcc/ChangeLog: PR tree-optimization/33735 PR tree-optimization/33572 * tree-inline.c (update_ssa_across_abnormal_edges): Revert 2007-10-09's change. * except.c (duplicate_eh_regions): Don't look for prev_try beyond ERT_ALLOWED_EXCEPTIONS with an empty list. gcc/testsuite/ChangeLog: PR tree-optimization/33735 * g++.dg/torture/pr33735.C: New. Added: trunk/gcc/testsuite/g++.dg/torture/pr33735.C Modified: trunk/gcc/ChangeLog trunk/gcc/except.c trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-inline.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #18 from danglin at gcc dot gnu dot org 2007-10-08 16:41 --- I see the same failure as HJL on hppa-unknown-linux-gnu. -- danglin at gcc dot gnu dot org changed: What|Removed |Added CC||danglin at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #19 from aoliva at gcc dot gnu dot org 2007-10-08 23:57 --- Subject: Bug 33572 Author: aoliva Date: Mon Oct 8 23:57:20 2007 New Revision: 129144 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129144 Log: PR tree-optimization/33572 * g++.dg/torture/pr33572.C: Replace with complete test. Modified: trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/g++.dg/torture/pr33572.C -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #20 from aoliva at gcc dot gnu dot org 2007-10-09 04:47 --- Subject: Bug 33572 Author: aoliva Date: Tue Oct 9 04:46:49 2007 New Revision: 129151 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129151 Log: PR tree-optimization/33572 * tree-inline.c (update_ssa_across_abnormal_edges): Tolerate the absence of a corresponding edge from the exit block. Modified: trunk/gcc/ChangeLog trunk/gcc/tree-inline.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #21 from aoliva at gcc dot gnu dot org 2007-10-09 04:48 --- Fixed. -- aoliva at gcc dot gnu dot org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #17 from hjl at lucon dot org 2007-10-07 14:39 --- (In reply to comment #16) Subject: Bug 33572 Author: aoliva Date: Sat Oct 6 11:43:56 2007 New Revision: 129051 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129051 Log: gcc/ChangeLog: PR tree-optimization/33572 * tree-cfg.c (verify_stmts): Check for missing PHI defs. * tree-inline.c (update_ssa_across_eh_edges): Renamed to... (update_ssa_across_abnormal_edges): ... this. Set slots in the return PHI node. (copy_edges_for_bb): Handle nonlocal label edges. (make_nonlocal_label_edges): Deleted. (optimize_inline_calls): Don't call it. gcc/testsuite/ChangeLog: PR tree-optimization/33572 * g++.dg/torture/pr33572.C: New. Added: trunk/gcc/testsuite/g++.dg/torture/pr33572.C Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-cfg.c trunk/gcc/tree-inline.c g++.dg/torture/pr33572.C failed on Linux/ia32: Executing on host: /export/build/gnu/gcc/build-i686-linux/gcc/testsuite/g++/../../g++ -B/export/build/gnu/gcc/build-i686-linux/gcc/testsuite/g++/../../ /net/gnu-13/export/gnu/src/gcc/gcc/gcc/testsuite/g++.dg/torture/pr33572.C -nostdinc++ -I/export/build/gnu/gcc/build-i686-linux/i686-pc-linux-gnu/libstdc++-v3/include/i686-pc-linux-gnu -I/export/build/gnu/gcc/build-i686-linux/i686-pc-linux-gnu/libstdc++-v3/include -I/net/gnu-13/export/gnu/src/gcc/gcc/libstdc++-v3/libsupc++ -I/net/gnu-13/export/gnu/src/gcc/gcc/libstdc++-v3/include/backward -I/net/gnu-13/export/gnu/src/gcc/gcc/libstdc++-v3/testsuite/util -fmessage-length=0 -O0-fno-show-column -L/export/build/gnu/gcc/build-i686-linux/i686-pc-linux-gnu/./libstdc++-v3/src/.libs -L/export/build/gnu/gcc/build-i686-linux/i686-pc-linux-gnu/./libstdc++-v3/src/.libs -L/export/build/gnu/gcc/build-i686-linux/i686-pc-linux-gnu/./libiberty -lm -o ./pr33572.exe(timeout = 300) /tmp/ccSsvTAh.o: In function `vector::deallocate(int*)':^M pr33572.C:(.text._ZN6vector10deallocateEPi[_ZN6vector10deallocateEPi]+0x24): undefined reference to `__gnu_cxx::__pooltrue::_M_reclaim_block(char*, unsigned long)'^M collect2: ld returned 1 exit status^M -- hjl at lucon dot org changed: What|Removed |Added CC||hjl at lucon dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #16 from aoliva at gcc dot gnu dot org 2007-10-06 11:44 --- Subject: Bug 33572 Author: aoliva Date: Sat Oct 6 11:43:56 2007 New Revision: 129051 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=129051 Log: gcc/ChangeLog: PR tree-optimization/33572 * tree-cfg.c (verify_stmts): Check for missing PHI defs. * tree-inline.c (update_ssa_across_eh_edges): Renamed to... (update_ssa_across_abnormal_edges): ... this. Set slots in the return PHI node. (copy_edges_for_bb): Handle nonlocal label edges. (make_nonlocal_label_edges): Deleted. (optimize_inline_calls): Don't call it. gcc/testsuite/ChangeLog: PR tree-optimization/33572 * g++.dg/torture/pr33572.C: New. Added: trunk/gcc/testsuite/g++.dg/torture/pr33572.C Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-cfg.c trunk/gcc/tree-inline.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #15 from aoliva at gcc dot gnu dot org 2007-10-03 01:04 --- Confirmed, the patch fixes it. -- aoliva at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |aoliva at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #13 from aoliva at gcc dot gnu dot org 2007-10-01 19:55 --- Patch at http://gcc.gnu.org/ml/gcc-patches/2007-10/msg00061.html is quite likely to fix this. -- aoliva at gcc dot gnu dot org changed: What|Removed |Added CC||aoliva at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #14 from aoliva at gcc dot gnu dot org 2007-10-02 01:38 --- Subject: Bug 33572 Author: aoliva Date: Tue Oct 2 01:37:59 2007 New Revision: 128939 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=128939 Log: PR tree-optimization/33572 * tree-cfg.c (verify_stmts): Check for missing PHI defs. * tree-inline.c (update_ssa_across_eh_edges): Renamed to... (update_ssa_across_abnormal_edges): ... this. Set slots in the return PHI node. (copy_edges_for_bb): Handle nonlocal label edges. (make_nonlocal_label_edges): Deleted. (optimize_inline_calls): Don't call it. Modified: branches/var-tracking-assignments-branch/gcc/ChangeLog.vta branches/var-tracking-assignments-branch/gcc/tree-cfg.c branches/var-tracking-assignments-branch/gcc/tree-inline.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #9 from rguenth at gcc dot gnu dot org 2007-09-30 12:41 --- Diego, we seem to have a general problem with the incremental SSA updater. If we rename foo$ptr in bb 6: # foo$ptr_16 = PHI foo$ptr_18(ab)(5), D.1758_3(4) p_12 = foo$ptr_16; foo$ptr_19(ab) = 0B; p_15 = foo$ptr_16; p_4 = foo$ptr_16; p_5 = foo$ptr_16; D.1781_6 = foo$ptr_16-_vptr.Foo; D.1782_7 = *D.1781_6; OBJ_TYPE_REF(D.1782_7;foo$ptr_16-0) (foo$ptr_16); goto bb 8; it apperantly does not deal correctly with the lifetimes of foo$ptr_16 and foo$ptr_19(ab) overlapping. Note that the SSA_NAME_OCCURS_IN_ABNORMAL_PHI flags are correct, foo$ptr_16 does not live across an EH edge (but just local in BB 6). Still, for this flag to prevent creating overlapping life ranges it would have need to be set on foo$ptr_16 as well. [In the testcase the final inliner is the first one to cause renaming of foo$ptr, the early SRA pass introduces the life-range overlap (but hidden by a copy chain) and the first copyprop makes it obviously visible (that's the dump from above)] -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||dnovillo at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #10 from rguenth at gcc dot gnu dot org 2007-09-30 12:51 --- Zdenek may also have an idea? -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||rakdver at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #11 from dnovillo at google dot com 2007-09-30 13:41 --- Subject: Re: [4.3 Regression] wrong code with -O On 30 Sep 2007 12:41:03 -, rguenth at gcc dot gnu dot org [EMAIL PROTECTED] wrote: --- Comment #9 from rguenth at gcc dot gnu dot org 2007-09-30 12:41 --- Diego, we seem to have a general problem with the incremental SSA updater. If we rename foo$ptr in bb 6: # foo$ptr_16 = PHI foo$ptr_18(ab)(5), D.1758_3(4) p_12 = foo$ptr_16; foo$ptr_19(ab) = 0B; p_15 = foo$ptr_16; p_4 = foo$ptr_16; p_5 = foo$ptr_16; D.1781_6 = foo$ptr_16-_vptr.Foo; D.1782_7 = *D.1781_6; OBJ_TYPE_REF(D.1782_7;foo$ptr_16-0) (foo$ptr_16); goto bb 8; Is the symbol foo$ptr being marked for renaming? If so, that's wrong. A gimple register symbol cannot be marked for renaming if there are overlapping live ranges in its SSA names. We don't have a general mechanism to prevent that. Mostly because we do not keep track when OLRs are created. The generic SSA updating mechanism has no cheap way of checking that. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #12 from rguenther at suse dot de 2007-09-30 14:01 ---
Subject: Re: [4.3 Regression] wrong code with
-O
On Sun, 30 Sep 2007, dnovillo at google dot com wrote:
--- Comment #9 from rguenth at gcc dot gnu dot org 2007-09-30 12:41
---
Diego, we seem to have a general problem with the incremental SSA updater.
If we rename foo$ptr in
bb 6:
# foo$ptr_16 = PHI foo$ptr_18(ab)(5), D.1758_3(4)
p_12 = foo$ptr_16;
foo$ptr_19(ab) = 0B;
p_15 = foo$ptr_16;
p_4 = foo$ptr_16;
p_5 = foo$ptr_16;
D.1781_6 = foo$ptr_16-_vptr.Foo;
D.1782_7 = *D.1781_6;
OBJ_TYPE_REF(D.1782_7;foo$ptr_16-0) (foo$ptr_16);
goto bb 8;
Is the symbol foo$ptr being marked for renaming? If so, that's wrong.
A gimple register symbol cannot be marked for renaming if there are
overlapping live ranges in its SSA names. We don't have a general
mechanism to prevent that. Mostly because we do not keep track when
OLRs are created. The generic SSA updating mechanism has no cheap way
of checking that.
Yes, it is marked for renaming by the inliner. Somehow the inliner
assumes there are no OLRs:
...
The function mark PHI_RESULT of such PHI nodes for renaming; it is
safe the EH edges are abnormal and SSA_NAME_OCCURS_IN_ABNORMAL_PHI
must be set. This means, that there will be no overlapping live ranges
for the underlying symbol.
This might change in future if we allow redirecting of EH edges and
we might want to change way build CFG pre-inlining to include
all the possible edges then. */
static void
update_ssa_across_eh_edges (basic_block bb)
{
edge e;
edge_iterator ei;
FOR_EACH_EDGE (e, ei, bb-succs)
if (!e-dest-aux
|| ((basic_block)e-dest-aux)-index == ENTRY_BLOCK)
{
tree phi;
gcc_assert (e-flags EDGE_EH);
for (phi = phi_nodes (e-dest); phi; phi = PHI_CHAIN (phi))
{
gcc_assert (SSA_NAME_OCCURS_IN_ABNORMAL_PHI
(PHI_RESULT (phi)));
mark_sym_for_renaming
(SSA_NAME_VAR (PHI_RESULT (phi)));
}
}
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #8 from rguenth at gcc dot gnu dot org 2007-09-28 12:59 --- Honza, as only foo$ptr is marked for renaming and that from update_ssa_across_eh_edges (), maybe this is a latent problem with inlining and EH info? Can you have a look? -- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||amacleod at redhat dot com, ||hubicka at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #7 from rguenth at gcc dot gnu dot org 2007-09-28 12:28 ---
After inlining deallocate into main we end up with:
int main(int, char**) (argc, D.2111)
{
static struct __pool pool;
static struct __pool pool;
int * p;
struct Foo * foo$ptr;
struct Foo * D.2185;
struct Foo * D.2182;
int save_filt.2;
void * save_eptr.1;
struct vector v;
int (*__vtbl_ptr_type) (void) D.2148;
int (*__vtbl_ptr_type) (void) * D.2147;
void * D.2145;
struct Foo * D.2124;
bb 2:
foo$ptr_18(ab) = 0B;
if (argc_1(D) != 0)
goto bb 3;
else
goto bb 5;
bb 3:
D.2145_2 = operator new (8);
bb 4:
D.2124_3 = (struct Foo *) D.2145_2;
D.2124_3-_vptr.Foo ={v} _ZTV3Foo[2];
goto bb 10;
bb 5:
bb 6:
if (0)
goto bb 7;
else
goto bb 8;
bb 7:
_M_reclaim_block (pool, 0B, 0);
bb 8:
bb 9:
bb 10:
# foo$ptr_16 = PHI foo$ptr_18(ab)(9), D.2124_3(4)
foo$ptr_19(ab) = 0B;
D.2147_6 = foo$ptr_16-_vptr.Foo;
D.2148_7 = *D.2147_6;
OBJ_TYPE_REF(D.2148_7;foo$ptr_16-0) (foo$ptr_16);
goto bb 12;
# foo$ptr_17(ab) = PHI foo$ptr_18(ab)(3), (7), foo$ptr_19(ab)(10)
L3:;
save_filt.2_10 = filter object;
save_eptr.1_11 = exception object;
D.2182_13 = foo$ptr_17(ab);
operator delete (D.2182_13);
exception object = save_eptr.1_11;
filter object = save_filt.2_10;
resx 1;
bb 12:
D.2185_14 = foo$ptr_19(ab);
operator delete (D.2185_14);
return 0;
}
after cleanup_cfg it still looks sane:
int main(int, char**) (argc, D.2111)
{
static struct __pool pool;
static struct __pool pool;
int * p;
struct Foo * foo$ptr;
struct Foo * D.2185;
struct Foo * D.2182;
int save_filt.2;
void * save_eptr.1;
struct vector v;
int (*__vtbl_ptr_type) (void) D.2148;
int (*__vtbl_ptr_type) (void) * D.2147;
void * D.2145;
struct Foo * D.2124;
bb 2:
foo$ptr_18(ab) = 0B;
if (argc_1(D) != 0)
goto bb 3;
else
goto bb 5;
bb 3:
D.2145_2 = operator new (8);
bb 4:
D.2124_3 = (struct Foo *) D.2145_2;
D.2124_3-_vptr.Foo ={v} _ZTV3Foo[2];
bb 5:
# foo$ptr_16 = PHI foo$ptr_18(ab)(2), D.2124_3(4)
foo$ptr_19(ab) = 0B;
D.2147_6 = foo$ptr_16-_vptr.Foo;
D.2148_7 = *D.2147_6;
OBJ_TYPE_REF(D.2148_7;foo$ptr_16-0) (foo$ptr_16);
goto bb 7;
# foo$ptr_17(ab) = PHI foo$ptr_18(ab)(3), foo$ptr_19(ab)(5)
L3:;
save_filt.2_10 = filter object;
save_eptr.1_11 = exception object;
D.2182_13 = foo$ptr_17(ab);
operator delete (D.2182_13);
exception object = save_eptr.1_11;
filter object = save_filt.2_10;
resx 1;
bb 7:
D.2185_14 = foo$ptr_19(ab);
operator delete (D.2185_14);
return 0;
}
the update_ssa (2048) call makes a mess out of it:
int main(int, char**) (argc, D.2111)
{
static struct __pool pool;
static struct __pool pool;
int * p;
struct Foo * foo$ptr;
struct Foo * D.2185;
struct Foo * D.2182;
int save_filt.2;
void * save_eptr.1;
struct vector v;
int (*__vtbl_ptr_type) (void) D.2148;
int (*__vtbl_ptr_type) (void) * D.2147;
void * D.2145;
struct Foo * D.2124;
bb 2:
foo$ptr_18(ab) = 0B;
if (argc_1(D) != 0)
goto bb 3;
else
goto bb 5;
bb 3:
D.2145_2 = operator new (8);
bb 4:
D.2124_3 = (struct Foo *) D.2145_2;
D.2124_3-_vptr.Foo ={v} _ZTV3Foo[2];
bb 5:
# foo$ptr_16 = PHI foo$ptr_18(ab)(2), D.2124_3(4)
foo$ptr_19(ab) = 0B;
D.2147_6 = foo$ptr_19(ab)-_vptr.Foo;
D.2148_7 = *D.2147_6;
OBJ_TYPE_REF(D.2148_7;foo$ptr_19(ab)-0) (foo$ptr_19(ab));
goto bb 7;
# foo$ptr_17(ab) = PHI foo$ptr_18(ab)(3), foo$ptr_19(ab)(5)
L3:;
save_filt.2_10 = filter object;
save_eptr.1_11 = exception object;
D.2182_13 = foo$ptr_17(ab);
operator delete (D.2182_13);
exception object = save_eptr.1_11;
filter object = save_filt.2_10;
resx 1;
bb 7:
D.2185_14 = foo$ptr_19(ab);
operator delete (D.2185_14);
return 0;
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #5 from rguenth at gcc dot gnu dot org 2007-09-27 21:03 ---
Confirmed. The reduced testcase gets optimized to
bb 3:
D.2145 = operator new (8);
bb 4:
((struct Foo *) D.2145)-_vptr.Foo = _ZTV3Foo[2];
bb 5:
D.2147 ={v} 0B-_vptr.Foo;
OBJ_TYPE_REF(*D.2147;0B-0) (0B);
where dereferencing NULL obviously is bogus. After 038t.sdse2 everything
looks ok:
bb 4:
D.2124_3 = (struct Foo *) D.2145_2;
D.2124_3-_vptr.Foo ={v} _ZTV3Foo[2];
goto bb 6;
bb 5:
deallocate (v, 0B);
bb 6:
# foo$ptr_16 = PHI foo$ptr_18(ab)(5), D.2124_3(4)
foo$ptr_19(ab) = 0B;
D.2147_6 = foo$ptr_16-_vptr.Foo;
D.2148_7 = *D.2147_6;
but after apply inline it's wrong!?
bb 3:
D.2145_2 = operator new (8);
bb 4:
D.2124_3 = (struct Foo *) D.2145_2;
D.2124_3-_vptr.Foo ={v} _ZTV3Foo[2];
bb 5:
# foo$ptr_16 = PHI foo$ptr_18(ab)(2), D.2124_3(4)
foo$ptr_19(ab) = 0B;
D.2147_6 = foo$ptr_19(ab)-_vptr.Foo;
D.2148_7 = *D.2147_6;
OBJ_TYPE_REF(D.2148_7;foo$ptr_19(ab)-0) (foo$ptr_19(ab));
goto bb 7;
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
CC||rguenth at gcc dot gnu dot
||org
Status|WAITING |NEW
Component|c++ |tree-optimization
Ever Confirmed|0 |1
Keywords||wrong-code
Last reconfirmed|-00-00 00:00:00 |2007-09-27 21:03:18
date||
Summary|wrong code with -O |[4.3 Regression] wrong code
||with -O
Target Milestone|--- |4.3.0
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
[Bug tree-optimization/33572] [4.3 Regression] wrong code with -O
--- Comment #6 from pcarlini at suse dot de 2007-09-27 21:19 --- I can see now from the reduced testcase that the library uses mt_alloc, not the default allocator. Next time, just say it... ;) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33572
