[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #1 from pinskia at gcc dot gnu dot org 2007-12-10 00:23 ---
This is a vectorizer vs not being able to run may_alias after it.
Confirmed.
(gdb) p debug_generic_expr (tag_type)
vector short int
(gdb) p debug_tree(tag)
unit size
align 16 symtab 0 alias set 5 canonical type 0x4277ff50
fields
HI file t.c line 6 col 18
size
unit size
align 16 offset_align 128
offset
bit offset context
chain > context
pointer_to_this chain >
addressable used file t.c line 11 col 1 context >
$3 = void
(gdb) p debug_generic_expr (0x4277ff50)
struct
{
short int s1;
short int s2;
short int s3;
short int s4;
}
(gdb) p tag_set
$5 = 6
(gdb) p get_alias_set (tag)
$6 = 5
--
pinskia at gcc dot gnu dot org changed:
What|Removed |Added
CC||dorit at gcc dot gnu dot org
Status|UNCONFIRMED |NEW
Ever Confirmed|0 |1
GCC build triplet|i686-pc-linux-gnu |
GCC host triplet|i686-pc-linux-gnu |
GCC target triplet|i686-*-*, x86_64-*-*|
Keywords||ice-on-valid-code
Last reconfirmed|-00-00 00:00:00 |2007-12-10 00:23:44
date||
Summary|-ftree-parallelize-loops=4 |-ftree-parallelize-loops=4
|ICE |ICE with the vectorizer also
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
-- rakdver at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |rakdver at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2007-12-10 00:23:44 |2007-12-16 22:07:18 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #2 from rakdver at gcc dot gnu dot org 2007-12-18 18:51 --- I do not see a way how to fix this in 4.3, other than disabling vectorizer when parallelization is enabled, or vice versa. -- rakdver at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|rakdver at gcc dot gnu dot |unassigned at gcc dot gnu |org |dot org Status|ASSIGNED|NEW http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #4 from jakub at gcc dot gnu dot org 2007-12-19 09:46 --- We could even TODO_rebuild_alias only if vectorizer changed anything... -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #3 from dorit at gcc dot gnu dot org 2007-12-19 09:38 --- > This is a vectorizer vs not being able to run may_alias after it can you please remind me why we can't run may_alias after the vectorizer? (and what do you think can be done about it?) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #5 from rakdver at gcc dot gnu dot org 2008-01-21 17:35 --- (In reply to comment #3) > > This is a vectorizer vs not being able to run may_alias after it > > can you please remind me why we can't run may_alias after the vectorizer? (and > what do you think can be done about it?) As for this ICE: may_alias rebuilds the smt's, and finds that the smt for the pointers created by vectorizer should be different (based on their type, not the artificial ones created by vectorizer). If we removed the hacks to update the alias info from vectorizer and run may_alias instead, we might get type-based alias analysis related misscompilations, as we create code of form vector_type *p = &object_of_arbitrary_type; *p = something; And TBA might decide that references through vector_type and arbitrary_type cannot alias. I actually tried this some time ago, and I was not able to get misscompilations this way (I am not really sure why). However, I got many fails in the testsuite (that I did not have time to investigate), so there may be some other problems with alias analysis after vectorizer. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #6 from razya at il dot ibm dot com 2008-05-12 09:13 ---
(In reply to comment #0)
> /* { dg-do compile } */
> /* { dg-options "-O3 -ftree-parallelize-loops=4" } */
> struct T
> {
> int t;
> struct { short s1, s2, s3, s4 } *s;
> };
> void
> foo (int *a, int *b, int *c, int *d, struct T *e)
> {
> int i;
> for (i = 0; i < e->t; i++)
> {
> e->s[i].s1 = a[i];
> e->s[i].s2 = b[i];
> e->s[i].s3 = c[i];
> e->s[i].s4 = d[i];
> }
> }
> ICEs on x86_64-linux with
> test.c: In function 'foo':
> test.c:11: internal compiler error: in get_smt_for, at tree-ssa-alias.c:3203
> Please submit a full bug report,
> with preprocessed source if appropriate.
> See for instructions.
I'm getting the same ICE (with r134722) when running spec2006/gobmk on power6.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #7 from razya at il dot ibm dot com 2008-05-12 09:42 --- (In reply to comment #6) > I'm getting the same ICE (with r134722) when running spec2006/gobmk on power6. The same failure appears also for other benchmarks from spec2006: h264ref, gromacs, calculix, tonto, wrf -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #8 from rguenth at gcc dot gnu dot org 2008-05-13 09:07 --- To fix potential TBAA miscompilations the vectorizer could mark all pointer (types) it creates with TYPE_REF_CAN_ALIAS_ALL. But I guess I have a more pragmatic fix for this. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |rguenth at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2007-12-16 22:07:18 |2008-05-13 09:07:52 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #9 from rguenth at gcc dot gnu dot org 2008-05-15 08:22 --- Fixed on the trunk. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Known to fail||4.3.0 Known to work||4.4.0 Target Milestone|--- |4.4.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #10 from rguenth at gcc dot gnu dot org 2008-05-15 08:22 --- Subject: Bug 34330 Author: rguenth Date: Thu May 15 08:21:32 2008 New Revision: 135329 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135329 Log: 2008-05-15 Richard Guenther <[EMAIL PROTECTED]> PR tree-optimization/34330 * tree-ssa-alias.c (get_smt_for): Only assert that accesses through the pointer will alias the SMT. * gcc.dg/torture/pr34330.c: New testcase. Added: trunk/gcc/testsuite/gcc.dg/torture/pr34330.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-ssa-alias.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #14 from razya at il dot ibm dot com 2008-05-15 09:41 --- (In reply to comment #13) > I see this also even without the patch. Can you open a new PR for this? (In reply to comment #12) > int q, x; > is missing in the testcase. I suppose you tried on the 4.3 branch? No, actually it's 4.4 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #12 from rguenth at gcc dot gnu dot org 2008-05-15 09:27 --- int q, x; is missing in the testcase. I suppose you tried on the 4.3 branch? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #15 from razya at il dot ibm dot com 2008-05-15 09:46 --- (In reply to comment #13) > I see this also even without the patch. Can you open a new PR for this? Sure. http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36244 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #11 from razya at il dot ibm dot com 2008-05-15 09:23 ---
(In reply to comment #10)
> Subject: Bug 34330
> Author: rguenth
> Date: Thu May 15 08:21:32 2008
> New Revision: 135329
> URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135329
> Log:
> 2008-05-15 Richard Guenther <[EMAIL PROTECTED]>
> PR tree-optimization/34330
> * tree-ssa-alias.c (get_smt_for): Only assert that accesses
> through the pointer will alias the SMT.
> * gcc.dg/torture/pr34330.c: New testcase.
> Added:
> trunk/gcc/testsuite/gcc.dg/torture/pr34330.c
> Modified:
> trunk/gcc/ChangeLog
> trunk/gcc/testsuite/ChangeLog
> trunk/gcc/tree-ssa-alias.c
/* { dg-do compile } */
/* { dg-options "-O3 -ftree-parallelize-loops=4 -c" } */
struct p7prior_s {
int mnum; /* number of mat emission Dirichlet mixtures */
float mq[200]; /* probabilities of mnum components */
float m[200][20]; /* match emission terms per mix component*/
};
struct p7prior_s *default_amino_prior(void);
struct p7prior_s *P7AllocPrior(void);
struct p7prior_s *
default_amino_prior(void)
{
struct p7prior_s *pri;
static float defmq[5] = {
0.178091, 0.056591, 0.0960191, 0.0781233, 0.0834977 };
static float defm[5][6] = {
{ 0.270671, 0.039848, 0.017576, 0.016415, 0.014268,
0.216147 },
{ 0.021465, 0.010300, 0.011741, 0.010883, 0.385651,
0.029156 },
{ 0.561459, 0.045448, 0.438366, 0.764167, 0.087364,
0.583402 },
{ 0.070143, 0.011140, 0.019479, 0.094657, 0.013162,
0.073732 },
{ 0.041103, 0.014794, 0.005610, 0.010216, 0.153602,
0.012049 }
};
pri = P7AllocPrior();
pri->mnum = 5;
for (q = 0; q < pri->mnum; q++)
{
pri->mq[q] = defmq[q];
for (x = 0; x < 6; x++)
pri->m[q][x] = defm[q][x];
}
return pri;
}
After applying Richard's patch, I'm still getting what seems to be a similar
problem.
prior.c:17: internal compiler error: in add_call_clobber_ops, at
tree-ssa-operands.c:1822
this was tested on power6, when both vectoriztion and autopar are enabled.
no failure when enabled seperately.
--
razya at il dot ibm dot com changed:
What|Removed |Added
CC||razya at il dot ibm dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #13 from rguenth at gcc dot gnu dot org 2008-05-15 09:29 --- I see this also even without the patch. Can you open a new PR for this? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #16 from rguenth at gcc dot gnu dot org 2008-05-20 21:04 --- Fixed. -- rguenth at gcc dot gnu dot org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED Target Milestone|4.4.0 |4.3.1 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
[Bug tree-optimization/34330] -ftree-parallelize-loops=4 ICE with the vectorizer also
--- Comment #17 from rguenth at gcc dot gnu dot org 2008-05-20 21:05 --- Subject: Bug 34330 Author: rguenth Date: Tue May 20 21:03:59 2008 New Revision: 135680 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=135680 Log: 2008-05-20 Richard Guenther <[EMAIL PROTECTED]> Backport from mainline: 2008-05-15 Richard Guenther <[EMAIL PROTECTED]> PR tree-optimization/34330 * tree-ssa-alias.c (get_smt_for): Only assert that accesses through the pointer will alias the SMT. * gcc.dg/torture/pr34330.c: New testcase. PR middle-end/36244 * tree-ssa-alias.c (new_type_alias): Do not set TREE_READONLY. * tree-flow-inline.h (unmodifiable_var_p): Memory tags never represent unmodifiable vars. * gcc.dg/torture/pr36244.c: New testcase. Added: branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pr34330.c - copied unchanged from r135329, trunk/gcc/testsuite/gcc.dg/torture/pr34330.c branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pr36244.c - copied unchanged from r135336, trunk/gcc/testsuite/gcc.dg/torture/pr36244.c Modified: branches/gcc-4_3-branch/gcc/ChangeLog branches/gcc-4_3-branch/gcc/testsuite/ChangeLog branches/gcc-4_3-branch/gcc/tree-flow-inline.h branches/gcc-4_3-branch/gcc/tree-ssa-alias.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34330
