* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
@ 2021-08-30 18:13 ` jakub at gcc dot gnu.org
2021-08-31 6:34 ` guojiufu at gcc dot gnu.org
` (11 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: jakub at gcc dot gnu.org @ 2021-08-30 18:13 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
CC| |jakub at gcc dot gnu.org
Version|unknown |12.0
Target Milestone|--- |12.0
Status|UNCONFIRMED |NEW
Last reconfirmed| |2021-08-30
Summary|wrong code at -O1 and above |[12 Regression] wrong code
|on x86_64-linux-gnu |at -O1 and above on
| |x86_64-linux-gnu
Priority|P3 |P1
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r12-3136-g3673dcf6d6baeb67bb70ff03d4cb3f92beed0075
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
2021-08-30 18:13 ` [Bug tree-optimization/102131] [12 Regression] " jakub at gcc dot gnu.org
@ 2021-08-31 6:34 ` guojiufu at gcc dot gnu.org
2021-08-31 9:51 ` guojiufu at gcc dot gnu.org
` (10 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: guojiufu at gcc dot gnu.org @ 2021-08-31 6:34 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #2 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
Thank you!
For this case, there are two exits, and through these two exits, different
niters(number of iterations) are calculated. It fails to handle this kind of
case well.
In ivcanon pass, the edge on the condition was removed incorrectly.
int a;
int main() {
unsigned b = 0;
int c = 1;
for (;b < 3; b++) {
if (c < b)
__builtin_abort ();
c+=3;
}
return 0;
}
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
2021-08-30 18:13 ` [Bug tree-optimization/102131] [12 Regression] " jakub at gcc dot gnu.org
2021-08-31 6:34 ` guojiufu at gcc dot gnu.org
@ 2021-08-31 9:51 ` guojiufu at gcc dot gnu.org
2021-08-31 10:34 ` amker at gcc dot gnu.org
` (9 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: guojiufu at gcc dot gnu.org @ 2021-08-31 9:51 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #3 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
The issue may come from 'iv0 cmp iv1' transform:
if (c<b) exit-loop
-->if (c>=b) in-loop
-->if (b<=c) in-loop
c: {4, +, 3}
b: {1, +, 1}
if ({1, +, 1} <= {4, +, 3})
==> if ({1,+,-2} <= {4,+,0}) //// here, error occur
==> if ({1,+,-2} < {5,+,0}) le-->lt
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (2 preceding siblings ...)
2021-08-31 9:51 ` guojiufu at gcc dot gnu.org
@ 2021-08-31 10:34 ` amker at gcc dot gnu.org
2021-09-01 9:23 ` guojiufu at gcc dot gnu.org
` (8 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: amker at gcc dot gnu.org @ 2021-08-31 10:34 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #4 from bin cheng <amker at gcc dot gnu.org> ---
(In reply to Jiu Fu Guo from comment #3)
> The issue may come from 'iv0 cmp iv1' transform:
>
> if (c<b) exit-loop
> -->if (c>=b) in-loop
> -->if (b<=c) in-loop
>
> c: {4, +, 3}
> b: {1, +, 1}
>
> if ({1, +, 1} <= {4, +, 3})
> ==> if ({1,+,-2} <= {4,+,0}) //// here, error occur
> ==> if ({1,+,-2} < {5,+,0}) le-->lt
So this duplicates to PR100740? Thanks
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (3 preceding siblings ...)
2021-08-31 10:34 ` amker at gcc dot gnu.org
@ 2021-09-01 9:23 ` guojiufu at gcc dot gnu.org
2021-09-02 7:14 ` guojiufu at gcc dot gnu.org
` (7 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: guojiufu at gcc dot gnu.org @ 2021-09-01 9:23 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #5 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
(In reply to bin cheng from comment #4)
> (In reply to Jiu Fu Guo from comment #3)
> > The issue may come from 'iv0 cmp iv1' transform:
> >
> > if (c<b) exit-loop
> > -->if (c>=b) in-loop
> > -->if (b<=c) in-loop
> >
> > c: {4, +, 3}
> > b: {1, +, 1}
> >
> > if ({1, +, 1} <= {4, +, 3})
> > ==> if ({1,+,-2} <= {4,+,0}) //// here, error occur
> > ==> if ({1,+,-2} < {5,+,0}) le-->lt
>
> So this duplicates to PR100740? Thanks
Yes, in theory, these PRs are related to the inaccurate
converting "{b1,s1} LT/LE {b2,s2}" to "{b1,s1-s2} LT/LE {b2,0}".
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (4 preceding siblings ...)
2021-09-01 9:23 ` guojiufu at gcc dot gnu.org
@ 2021-09-02 7:14 ` guojiufu at gcc dot gnu.org
2021-11-15 17:04 ` [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136 jakub at gcc dot gnu.org
` (6 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: guojiufu at gcc dot gnu.org @ 2021-09-02 7:14 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #6 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
Drafted a patch as below. With this patch, those cases can pass.
diff --git a/gcc/tree-ssa-loop-niter.c b/gcc/tree-ssa-loop-niter.c
index 7af92d1c893..a400c42919b 100644
--- a/gcc/tree-ssa-loop-niter.c
+++ b/gcc/tree-ssa-loop-niter.c
@@ -1866,6 +1866,24 @@ number_of_iterations_cond (class loop *loop,
|| !iv0->no_overflow || !iv1->no_overflow))
return false;
+ /* GT/GE has been transformed to LT/LE already.
+ cmp_code could be LT, LE or NE
+
+ For LE/LT transform
+ {iv0.base, iv0.step} LT/LE {iv1.base, iv1.step}
+ to
+ {iv0.base, iv0.step - iv1.step} LT/LE {iv1.base, 0}
+ Negative iv0.step - iv1.step means decreasing until wrap,
+ then the transform is not accurate.
+
+ For example:
+ {1, +, 1} <= {4, +, 3}
+ is not same with
+ {1, +, -2} <= {4, +, 0}
+ */
+ if ((code == LE_EXPR || code == LT_EXPR) && tree_int_cst_sign_bit
(step))
+ return false;
+
iv0->step = step;
if (!POINTER_TYPE_P (type))
iv0->no_overflow = false;
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (5 preceding siblings ...)
2021-09-02 7:14 ` guojiufu at gcc dot gnu.org
@ 2021-11-15 17:04 ` jakub at gcc dot gnu.org
2021-11-16 2:01 ` guojiufu at gcc dot gnu.org
` (5 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: jakub at gcc dot gnu.org @ 2021-11-15 17:04 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Any further progress on this?
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (6 preceding siblings ...)
2021-11-15 17:04 ` [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136 jakub at gcc dot gnu.org
@ 2021-11-16 2:01 ` guojiufu at gcc dot gnu.org
2021-12-15 10:18 ` zhendong.su at inf dot ethz.ch
` (4 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: guojiufu at gcc dot gnu.org @ 2021-11-16 2:01 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #8 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #7)
> Any further progress on this?
Thanks, Jabkub!
There is a patch that may cover more cases (PR102636/PR100740.. and other cases
where 'vi0.step - iv1.step > 0'), but it seems complex.
https://gcc.gnu.org/pipermail/gcc-patches/2021-October/582766.html
We may need a better fix.
BR,
Jiufu
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (7 preceding siblings ...)
2021-11-16 2:01 ` guojiufu at gcc dot gnu.org
@ 2021-12-15 10:18 ` zhendong.su at inf dot ethz.ch
2022-01-24 10:43 ` rguenth at gcc dot gnu.org
` (3 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: zhendong.su at inf dot ethz.ch @ 2021-12-15 10:18 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #9 from Zhendong Su <zhendong.su at inf dot ethz.ch> ---
Perhaps the following is due to the same root cause?
[558] % gcctk -O0 small.c; ./a.out
[559] % gcctk -O1 small.c; ./a.out
Aborted
[560] % cat small.c
int a;
int main() {
unsigned b = 0;
for (a = 2; a < 8; a += 2)
if (++b > a)
__builtin_abort();
return 0;
}
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (8 preceding siblings ...)
2021-12-15 10:18 ` zhendong.su at inf dot ethz.ch
@ 2022-01-24 10:43 ` rguenth at gcc dot gnu.org
2022-01-24 12:15 ` cvs-commit at gcc dot gnu.org
` (2 subsequent siblings)
12 siblings, 0 replies; 14+ messages in thread
From: rguenth at gcc dot gnu.org @ 2022-01-24 10:43 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
*** Bug 102636 has been marked as a duplicate of this bug. ***
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (9 preceding siblings ...)
2022-01-24 10:43 ` rguenth at gcc dot gnu.org
@ 2022-01-24 12:15 ` cvs-commit at gcc dot gnu.org
2022-01-24 12:48 ` rguenth at gcc dot gnu.org
2022-01-25 1:28 ` crazylht at gmail dot com
12 siblings, 0 replies; 14+ messages in thread
From: cvs-commit at gcc dot gnu.org @ 2022-01-24 12:15 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
--- Comment #11 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:
https://gcc.gnu.org/g:f1af8528d34418bc874ae9d993ee0dc3559972d2
commit r12-6844-gf1af8528d34418bc874ae9d993ee0dc3559972d2
Author: Richard Biener <rguenther@suse.de>
Date: Mon Jan 24 11:50:06 2022 +0100
tree-optimization/102131 - fix niter analysis wrt overflow
This fixes the overflow issues seen with analyzing
BASE0 + STEP0 cmp BASE1 + STEP1 as BASE0 + STEP0 - STEP1 cmp BASE1
by following the logic we have when simplifying comparisons.
2022-01-24 Richard Biener <rguenther@suse.de>
Jiufu Guo <guojiufu@linux.ibm.com>
PR tree-optimization/100740
PR tree-optimization/101508
PR tree-optimization/101972
PR tree-optimization/102131
* tree-ssa-loop-niter.cc (number_of_iterations_cond): Properly
constrain BASE0 + STEP0 cmp BASE1 + STEP1 to
BASE0 + STEP0 - STEP1 cmp BASE1 transform.
* gcc.dg/torture/pr100740.c: New testcase.
* gcc.dg/torture/pr101508.c: Likewise.
* gcc.dg/torture/pr101972.c: Likewise.
* gcc.dg/torture/pr102131-1.c: Likewise.
* gcc.dg/torture/pr102131-2.c: Likewise.
* gcc.dg/torture/pr102131-3.c: Likewise.
* gcc.dg/torture/pr102131-4.c: Likewise.
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (10 preceding siblings ...)
2022-01-24 12:15 ` cvs-commit at gcc dot gnu.org
@ 2022-01-24 12:48 ` rguenth at gcc dot gnu.org
2022-01-25 1:28 ` crazylht at gmail dot com
12 siblings, 0 replies; 14+ messages in thread
From: rguenth at gcc dot gnu.org @ 2022-01-24 12:48 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |FIXED
Known to work| |12.0
--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed. PR100740 is still open to track backporting for testcases miscompiled
in earlier releases.
^ permalink raw reply [flat|nested] 14+ messages in thread* [Bug tree-optimization/102131] [12 Regression] wrong code at -O1 and above on x86_64-linux-gnu since r12-3136
2021-08-30 17:12 [Bug tree-optimization/102131] New: wrong code at -O1 and above on x86_64-linux-gnu zhendong.su at inf dot ethz.ch
` (11 preceding siblings ...)
2022-01-24 12:48 ` rguenth at gcc dot gnu.org
@ 2022-01-25 1:28 ` crazylht at gmail dot com
12 siblings, 0 replies; 14+ messages in thread
From: crazylht at gmail dot com @ 2022-01-25 1:28 UTC (permalink / raw)
To: gcc-bugs
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102131
Hongtao.liu <crazylht at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |crazylht at gmail dot com
--- Comment #13 from Hongtao.liu <crazylht at gmail dot com> ---
(In reply to CVS Commits from comment #11)
> The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:
>
> https://gcc.gnu.org/g:f1af8528d34418bc874ae9d993ee0dc3559972d2
>
> commit r12-6844-gf1af8528d34418bc874ae9d993ee0dc3559972d2
> Author: Richard Biener <rguenther@suse.de>
> Date: Mon Jan 24 11:50:06 2022 +0100
>
> tree-optimization/102131 - fix niter analysis wrt overflow
>
> This fixes the overflow issues seen with analyzing
> BASE0 + STEP0 cmp BASE1 + STEP1 as BASE0 + STEP0 - STEP1 cmp BASE1
> by following the logic we have when simplifying comparisons.
>
> 2022-01-24 Richard Biener <rguenther@suse.de>
> Jiufu Guo <guojiufu@linux.ibm.com>
>
> PR tree-optimization/100740
> PR tree-optimization/101508
> PR tree-optimization/101972
> PR tree-optimization/102131
> * tree-ssa-loop-niter.cc (number_of_iterations_cond): Properly
> constrain BASE0 + STEP0 cmp BASE1 + STEP1 to
> BASE0 + STEP0 - STEP1 cmp BASE1 transform.
>
> * gcc.dg/torture/pr100740.c: New testcase.
> * gcc.dg/torture/pr101508.c: Likewise.
> * gcc.dg/torture/pr101972.c: Likewise.
> * gcc.dg/torture/pr102131-1.c: Likewise.
> * gcc.dg/torture/pr102131-2.c: Likewise.
> * gcc.dg/torture/pr102131-3.c: Likewise.
> * gcc.dg/torture/pr102131-4.c: Likewise.
It regresses
https://gcc.gnu.org/pipermail/gcc-regression/2022-January/076279.html
^ permalink raw reply [flat|nested] 14+ messages in thread