[Bug other/97705] New: [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[Bug other/97705] New: [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

            Bug ID: 97705
           Summary: [11 regression]
                    cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after
                    r11-4637
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
          Assignee: unassigned at gcc dot gnu.org
          Reporter: seurer at gcc dot gnu.org
  Target Milestone: ---

g:f5e18dd9c7dacc9671044fc669bd5c1b26b6bdba, r11-4637

make  -k check-gcc
RUNTESTFLAGS="unsorted.exp=gcc.c-torture/unsorted/dump-noaddr.c"

FAIL: gcc.c-torture/unsorted/dump-noaddr.c.*r.ira,  -O3 -g  comparison

# of expected passes            1467
# of unexpected failures        1

There are many (thousands?) of lines of miscompare for the failing test.

commit f5e18dd9c7dacc9671044fc669bd5c1b26b6bdba
Author: Kewen Lin <linkw@gcc.gnu.org>
Date:   Tue Nov 3 02:51:47 2020 +0000

[Bug other/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

--- Comment #1 from seurer at gcc dot gnu.org ---
I tried this on multiple systems and it looks like it may only fail on a BE
system.

[Bug other/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |linkw at gcc dot gnu.org
   Last reconfirmed|                            |2020-11-04

--- Comment #2 from Kewen Lin <linkw at gcc dot gnu.org> ---
Thanks for reporting and sorry for the failure. I did run the regression
testing on P8 LE, but thought it's endianness irrelevant and didn't run it on
BE.

[Bug testsuite/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.0
          Component|other                       |testsuite

[Bug testsuite/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

--- Comment #3 from Kewen Lin <linkw at gcc dot gnu.org> ---
The "-DMASK=2" dumping has more lines for register 282, which is introduced in
ira. Something weird causes ira to dump more contexts.

$ diff dump1/dump-noaddr.c.289r.ira dump2/dump-noaddr.c.289r.ira
107a108
>   r282 costs: BASE_REGS:0 GENERAL_REGS:0 FLOAT_REGS:312 ALTIVEC_REGS:312 VSX_REGS:312 GEN_OR_FLOAT_REGS:312 GEN_OR_VSX_REGS:312 LINK_REGS:468 CTR_REGS:468 LINK_OR_CTR_REGS:468 SPEC_OR_GEN_REGS:468 ALL_REGS:1872 MEM:312
373a375
>   r282 costs: GENERAL_REGS:0 FLOAT_REGS:312 ALTIVEC_REGS:312 VSX_REGS:312 GEN_OR_FLOAT_REGS:312 GEN_OR_VSX_REGS:312 LINK_REGS:468 CTR_REGS:468 LINK_OR_CTR_REGS:468 SPEC_OR_GEN_REGS:468 ALL_REGS:1872 MEM:312

[Bug testsuite/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

--- Comment #4 from Kewen Lin <linkw at gcc dot gnu.org> ---
I think my commit just exposed one bug in ira. The newly introduced function
remove_scratches can bump the max_regno, then the data structures
regstat_n_sets_and_refs and reg_info_p which are allocated according to
max_regno become stale. When we call print_pseudo_costs to dump some register
information, it use the latest regno with max_reg_num (), it leads we can
access some regno which don't have relevant data structures which have been
allocated, the read values can be random (out of array bound access).

The fix can be to free/re-init/re-compute the relevant data structures when we
know the max_regno already changes like remove_scratches succeeds.

[Bug rtl-optimization/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Kewen Lin <linkw@gcc.gnu.org>:

https://gcc.gnu.org/g:ce4ae1f4893e322495c5d24b2f0e807a7f7cf92f

commit r11-4827-gce4ae1f4893e322495c5d24b2f0e807a7f7cf92f
Author: Kewen Lin <linkw@linux.ibm.com>
Date:   Sun Nov 8 20:35:21 2020 -0600

    ira: Recompute regstat as max_regno changes [PR97705]

    As PR97705 shows, the commit r11-4637 caused some dumping
    comparison difference error on pass ira.  It exposed one
    issue about the newly introduced function remove_scratches,
    which can increase the largest pseudo reg number if it
    succeeds, later some function will use the max_reg_num()
    to get the latest max_regno, when iterating the numbers
    we can access some data structures which are allocated as
    the previous max_regno, some out of array bound accesses
    can occur, the failure can be random since the values
    beyond the array could be random.

    This patch is to free/reinit/recompute the relevant data
    structures that is regstat_n_sets_and_refs and reg_info_p
    to ensure we won't access beyond some array bounds.

    Bootstrapped/regtested on powerpc64le-linux-gnu P9 and
    powerpc64-linux-gnu P8.

    gcc/ChangeLog:

            PR rtl-optimization/97705
            * ira.c (ira): Refactor some regstat free/init/compute invocation
            into lambda function regstat_recompute_for_max_regno, and call it
            when max_regno increases as remove_scratches succeeds.

[Bug rtl-optimization/97705] [11 regression] cc.c-torture/unsorted/dump-noaddr.c.*r.ira fails after r11-4637

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97705

Kewen Lin <linkw at gcc dot gnu.org> changed:

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

--- Comment #6 from Kewen Lin <linkw at gcc dot gnu.org> ---
Should be fixed with latest trunk r11-4827.