[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #14 from steven at gcc dot gnu dot org 2006-05-15 19:55 --- No need for investigation anymore :-) -- steven at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|steven 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=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #13 from patchapp at dberlin dot org 2006-05-15 19:45 --- Subject: Bug number PR27603 A patch for this bug has been added to the patch tracker. The mailing list url for the patch is http://gcc.gnu.org/ml/gcc-patches/2006-05/msg00655.html -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #12 from rguenth at gcc dot gnu dot org 2006-05-15 17:36 --- Subject: Bug 27603 Author: rguenth Date: Mon May 15 17:35:48 2006 New Revision: 113797 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=113797 Log: 2006-05-15 Richard Guenther <[EMAIL PROTECTED]> PR tree-optimization/27603 * tree-ssa-loop-niter.c (infer_loop_bounds_from_undefined): Do computations in original type. * gcc.dg/torture/pr27603.c: New testcase. Added: trunk/gcc/testsuite/gcc.dg/torture/pr27603.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-ssa-loop-niter.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #11 from rguenth at gcc dot gnu dot org 2006-05-15 13:46 --- I have a patch in testing. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #10 from pinskia at gcc dot gnu dot org 2006-05-15 06:39
---
After fixing a type mismatch in infer_loop_bounds_from_undefined, I get:
(analyze_scalar_evolution
(loop_nb = 1)
(scalar = D.1523_6)
(get_scalar_evolution
(scalar = D.1523_6)
(scalar_evolution = {4294967293, +, 1}_1))
(set_scalar_evolution
(scalar = D.1523_6)
(scalar_evolution = {4294967293, +, 1}_1))
)
(instantiate_parameters
(loop_nb = 1)
(chrec = {4294967293, +, 1}_1)
(res = {4294967293, +, 1}_1))
Found new range for D.1523_6: [4294967293, +INF]
I don't see anything wrong with the scev really. In fact it is correct. maybe
what is wrong is VRP not asking the correct question.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #9 from pinskia at gcc dot gnu dot org 2006-05-15 05:44 --- (In reply to comment #8) integer_nonzerop should be used instead of !integer_zerop. See http://gcc.gnu.org/ml/gcc-patches/2006-05/msg00587.html. -- Pinski -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #8 from ian at airs dot com 2006-05-15 05:41 ---
Steven: what's your patch? It seems to me that this fixes the problem in
mainline:
Index: tree-ssa-loop-niter.c
===
--- tree-ssa-loop-niter.c (revision 113773)
+++ tree-ssa-loop-niter.c (working copy)
@@ -1612,7 +1612,7 @@ infer_loop_bounds_from_undefined (struct
diff = fold_build2 (MINUS_EXPR, utype,
TYPE_MAX_VALUE (type), init);
- if (!integer_zerop (step))
+ if (!integer_zerop (step) && !TREE_OVERFLOW (diff))
{
estimation = fold_build2 (CEIL_DIV_EXPR, utype, diff,
step);
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP (-O2)
--- Comment #7 from steven at gcc dot gnu dot org 2006-05-15 04:36 --- Investigating a fix. -- steven at gcc dot gnu dot org changed: What|Removed |Added AssignedTo|unassigned at gcc dot gnu |steven at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2006-05-14 17:44:48 |2006-05-15 04:36:57 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
--- Comment #6 from mmitchel at gcc dot gnu dot org 2006-05-14 22:43 --- My mistake; I had thought it was only generating wrong-code with non-default options. I have adjusted this to P1. -- mmitchel at gcc dot gnu dot org changed: What|Removed |Added Priority|P2 |P1 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
--- Comment #5 from pinskia at gcc dot gnu dot org 2006-05-14 22:42 --- Is there a reason why this is not a P1? This is wrong code. -- pinskia at gcc dot gnu dot org changed: What|Removed |Added CC||mmitchel at gcc dot gnu dot ||org GCC build triplet|i686-pc-linux-gnu | GCC host triplet|i686-pc-linux-gnu | GCC target triplet|i686-pc-linux-gnu | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
-- mmitchel at gcc dot gnu dot org changed: What|Removed |Added Priority|P3 |P2 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
-- pinskia at gcc dot gnu dot org changed: What|Removed |Added CC||pinskia at gcc dot gnu dot ||org Severity|major |critical http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
--- Comment #4 from rguenth at gcc dot gnu dot org 2006-05-14 18:00 ---
(set_scalar_evolution
(scalar = j_3)
(scalar_evolution = {0, +, 1}_1))
)
(instantiate_parameters
(loop_nb = 1)
(chrec = {0, +, 1}_1)
(res = {0, +, 1}_1))
Found new range for j_3: [0, 0]
huh!? (this is from 4.1 dumps)
:;
j_3 = ASSERT_EXPR ;
j.0_5 = (unsigned int) j_3;
D.1285_6 = j.0_5 - 3;
if (D.1285_6 <= 1) goto ; else goto ;
:;
exit (0);
:;
D.1286_7 = 6 - j_3;
D.1287_8 = D.1286_7 * 1000;
a = D.1287_8;
j_10 = j_3 + 1;
# j_2 = PHI <0(0), j_10(3)>;
:;
if (j_2 <= 5) goto ; else goto ;
:;
abort ();
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
--- Comment #3 from rguenth at gcc dot gnu dot org 2006-05-14 17:56 ---
We fold (j == 3 || j == 4) to (unsigned)j - 3 <= 1, which VRP considers to be
always false.
void exit (int);
void abort (void);
int a;
int main()
{
int j;
for (j = 0; j < 6; j++)
{
if ((unsigned)j - 3 <= 1)
exit (0);
a = 1000 * (6 - j);
}
abort ();
}
somehow the false information is extracted from the multiplication expression.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
--- Comment #2 from rguenth at gcc dot gnu dot org 2006-05-14 17:44 ---
Confirmed. Reduced testcase:
void abort(void);
int g1[6] = { 0, 0, 0, 0, 0, 0 };
int g4 = 0;
void Foo(void)
{
int j;
for (j = 0; j < 6; j++) {
if (j == 3 || j == 4)
g1[j] = 100;
g4 += 1000 * (6 - j);
}
}
int main() {
Foo();
if (g1[3] != 100 || g1[4] != 100)
abort ();
return 0;
}
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Ever Confirmed|0 |1
Last reconfirmed|-00-00 00:00:00 |2006-05-14 17:44:48
date||
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
[Bug tree-optimization/27603] [4.1/4.2 Regression] wrong code, apparently due to bad VRP
-- rguenth at gcc dot gnu dot org changed: What|Removed |Added CC||rguenth at gcc dot gnu dot ||org Component|c++ |tree-optimization Keywords||wrong-code Known to fail|4.1.0 |4.1.0 4.2.0 Summary|wrong code, apparently due |[4.1/4.2 Regression] wrong |to bad VRP |code, apparently due to bad ||VRP Target Milestone|--- |4.1.1 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27603
