[Bug testsuite/108776] New: new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[Bug testsuite/108776] New: new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

            Bug ID: 108776
           Summary: new test case c-c++-common/rotate-11.c from
                    r12-9158-ga015ebe382cd6d fails
           Product: gcc
           Version: 12.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: testsuite
          Assignee: unassigned at gcc dot gnu.org
          Reporter: seurer at gcc dot gnu.org
  Target Milestone: ---

g:a015ebe382cd6d0beab9db4ad33fbd252b7e2339, r12-9158-ga015ebe382cd6d

make  -k check-gcc RUNTESTFLAGS="dg.exp=c-c++-common/rotate-11.c"
FAIL: c-c++-common/rotate-11.c  -Wc++-compat   scan-tree-dump-times optimized "
r<< " 5
FAIL: c-c++-common/rotate-11.c  -std=gnu++98  scan-tree-dump-times optimized "
r<< " 5
FAIL: c-c++-common/rotate-11.c  -std=gnu++14  scan-tree-dump-times optimized "
r<< " 5
FAIL: c-c++-common/rotate-11.c  -std=gnu++17  scan-tree-dump-times optimized "
r<< " 5
FAIL: c-c++-common/rotate-11.c  -std=gnu++20  scan-tree-dump-times optimized "
r<< " 5
# of expected passes            2
# of expected passes            8
# of unexpected failures        1
# of unexpected failures        4


commit a015ebe382cd6d0beab9db4ad33fbd252b7e2339 (HEAD, refs/bisect/bad)
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Jan 19 10:00:51 2023 +0100

    forwprop: Further fixes for simplify_rotate [PR108440]

[Bug testsuite/108776] new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|powerpc64le-linux-gnu       |powerpc64le-linux-gnu
                   |                            |s390x-linux-gnu
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
                 CC|                            |rguenth at gcc dot gnu.org
   Last reconfirmed|                            |2023-05-11

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Also fails on s390x.

On ppc64le we only have four r<<, baz() and qux() are not handled (why
do we have 6 functions but expect 5 rotates only without checking exactly
which function we expect not to transform?).  On trunk we handle
baz() fine but still do not handle qux().

The IL into forwprop1 is the same so it's likely differences in ranger
behavior triggering here.

branch:

;; Function baz (baz, funcdef_no=2, decl_uid=3331, cgraph_uid=3,
symbol_order=2)

326      range_of_expr(y_11(D)) at stmt _2 = _1 << y_11(D);
327        range_on_entry (y_11(D)) to BB 5
328          range_of_stmt (y_11(D)) at stmt GIMPLE_NOP
   GLOBAL : UPDATE cache for y_11(D) in BB 0 : successors :   : No updates!
             TRUE : (328) range_of_stmt (y_11(D)) unsigned int VARYING

y_11(D) : CACHE: BB 5 DOM query, found unsigned int VARYING at BB0
CACHE: Range for DOM returns : unsigned int VARYING
Filled from dominator! :  unsigned int VARYING
           TRUE : (327) range_on_entry (y_11(D)) unsigned int VARYING
         TRUE : (326) range_of_expr (y_11(D)) unsigned int VARYING
unsigned char baz (unsigned char x, unsigned int y)

trunk:

;; Function baz (baz, funcdef_no=2, decl_uid=3954, cgraph_uid=3,
symbol_order=2)

299      range_of_expr(y_11(D)) at stmt _2 = _1 << y_11(D);
300        range_on_entry (y_11(D)) to BB 5
301          range_of_stmt (y_11(D)) at stmt GIMPLE_NOP
   GLOBAL : UPDATE cache for y_11(D) in BB 0 : successors :   : No updates!
             TRUE : (301) range_of_stmt (y_11(D)) [irange] unsigned int VARYING

y_11(D) : CACHE: BB 5 DOM query for y_11(D), found [irange] unsigned int
VARYING at BB0
302     GORI  outgoing_edge for y_11(D) on edge 2->5
303     GORI    compute op 1 (y_11(D)) at if (y_11(D) > 8)
        GORI      LHS =[irange] _Bool [0, 0] NONZERO 0x0
        GORI      Computes y_11(D) = [irange] unsigned int [0, 8] NONZERO 0xf
intersect Known range : [irange] unsigned int VARYING
        GORI    TRUE : (303) produces  (y_11(D)) [irange] unsigned int [0, 8]
NONZERO 0xf
        GORI  TRUE : (302) outgoing_edge (y_11(D)) [irange] unsigned int [0, 8]
NONZERO 0xf
CACHE: BB 3 DOM query for y_11(D), found [irange] unsigned int VARYING at BB2
304     GORI  outgoing_edge for y_11(D) on edge 2->3
305     GORI    compute op 1 (y_11(D)) at if (y_11(D) > 8)
        GORI      LHS =[irange] _Bool [1, 1]
        GORI      Computes y_11(D) = [irange] unsigned int [9, +INF] intersect
Known range : [irange] unsigned int VARYING
        GORI    TRUE : (305) produces  (y_11(D)) [irange] unsigned int [9,
+INF]
        GORI  TRUE : (304) outgoing_edge (y_11(D)) [irange] unsigned int [9,
+INF]
CACHE: Adjusted edge range for 2->3 : [irange] unsigned int [9, +INF]
CACHE: Range for DOM returns : [irange] unsigned int [9, +INF]
306     GORI  outgoing_edge for y_11(D) on edge 3->5
307     GORI    compute op 1 (y_11(D)) at if (y_11(D) != 16)
        GORI      LHS =[irange] _Bool [0, 0] NONZERO 0x0
        GORI      Computes y_11(D) = [irange] unsigned int [16, 16] NONZERO
0x10 intersect Known range : [irange] unsigned int VARYING
        GORI    TRUE : (307) produces  (y_11(D)) [irange] unsigned int [16, 16]
NONZERO 0x10
        GORI  TRUE : (306) outgoing_edge (y_11(D)) [irange] unsigned int [16,
16] NONZERO 0x10
CACHE: Range for DOM returns : [irange] unsigned int [0, 8][16, 16] NONZERO
0x1f
Filled from dominator! :  [irange] unsigned int [0, 8][16, 16] NONZERO 0x1f
           TRUE : (300) range_on_entry (y_11(D)) [irange] unsigned int [0,
8][16, 16] NONZERO 0x1f
         TRUE : (299) range_of_expr (y_11(D)) [irange] unsigned int [0, 8][16,
16] NONZERO 0x1f
gimple_simplified to _14 = _16;
Removing dead stmt _14 = _16;

unsigned char baz (unsigned char x, unsigned int y)


On x86_64 the IL into forwprop1 is different (logical-op-non-short-circuit):

ppc64le:
  <bb 2> :
  if (y_11(D) > 8)
    goto <bb 3>; [INV]
  else
    goto <bb 5>; [INV]

  <bb 3> :
  if (y_11(D) != 16)
    goto <bb 4>; [INV]
  else
    goto <bb 5>; [INV]

x86_64:
  <bb 2> :
  _1 = y_14(D) > 8;
  _2 = y_14(D) != 16;
  _3 = _1 & _2;
  if (_3 != 0)
    goto <bb 3>; [INV]
  else
    goto <bb 4>; [INV]

Setting --param logical-op-non-short-circuit=1 fixes the testcase for ppc64le
on the branch, I suspect it will be the same for s390x.

[Bug testsuite/108776] new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #1)
> Setting --param logical-op-non-short-circuit=1 fixes the testcase for ppc64le
> on the branch, I suspect it will be the same for s390x.

Then let's go with that.
The PR106523/PR108440 fixes are heavily dependent on the ranger so that while
fixing the wrong-code it doesn't regress much in rotate recognition.  And the
ranger is increasingly less capable in older branches, that is why I chose not
to backport those
fixes at all to 10/11 branches.

[Bug testsuite/108776] new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

[Bug testsuite/108776] new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

--- Comment #3 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:9410a52061ffe4a9e6ef8fd82f9c21d1874ff589

commit r12-9538-g9410a52061ffe4a9e6ef8fd82f9c21d1874ff589
Author: Richard Biener <rguenther@suse.de>
Date:   Thu May 11 09:13:31 2023 +0200

    testsuite/108776 - avoid c-c++-common/rotate-11.c FAIL

    On the branch ranger isn't powerful enough to handle some cases
    appearing with logical-op-non-short-circuit evaluating to false
    causing FAILs of the testcase for ppc64le and s390x.  The following
    foces logical-op-non-short-circuit to true for this testcase
    on the branch.

            PR testsuite/108776
            * c-c++-common/rotate-11.c: Add --param
logical-op-non-short-circuit=1.

[Bug testsuite/108776] new test case c-c++-common/rotate-11.c from r12-9158-ga015ebe382cd6d fails

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108776

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.