[Bug middle-end/54850] New: [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[Bug middle-end/54850] New: [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

             Bug #: 54850
           Summary: [4.8 Regression] FAIL:
                    gcc.c-torture/execute/20041113-1.c execution,  -Os
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: danglin@gcc.gnu.org
              Host: hppa64-hp-hpux11.11
            Target: hppa64-hp-hpux11.11
             Build: hppa64-hp-hpux11.11


Created attachment 28380
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28380
rtl dump

Executing on host: /test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c 
-fno-diagnostics-show-caret  -w  -Os   -lm   -o
/test/gnu/gcc/objdir/gcc/testsuite/gcc/20041113-1.x5    (timeout = 300)
spawn /test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c
-fno-diagnostics-show-caret -w -Os -lm -o
/test/gnu/gcc/objdir/gcc/testsuite/gcc/20041113-1.x5
PASS: gcc.c-torture/execute/20041113-1.c compilation,  -Os 
Setting LD_LIBRARY_PATH to :/test/gnu/gcc/objdir/gcc::/test/gnu/gcc/objdir/gcc
spawn [open ...]
FAIL: gcc.c-torture/execute/20041113-1.c execution,  -Os 

Function test is miscompiled:

test:
        .PROC
        .CALLINFO FRAME=144,CALLS,SAVE_RP,ENTRY_GR=3
        .ENTRY
        std %r2,-16(%r30)
        ldo 144(%r30),%r30
        ldo -56(%r29),%r29
        std %r4,-128(%r30)
        ldw 4(%r29),%r28
        std %r26,-8(%r29)
        std %r25,0(%r29)
        std %r24,8(%r29)
        std %r23,16(%r29)
        std %r22,24(%r29)
        std %r21,32(%r29)
        std %r20,40(%r29)
        std %r19,48(%r29)
        cmpib,= 1,%r28,L$0002
        std %r29,-136(%r30)
        ...

The cmpib instruction fails because the "ldw 4(%r29),%r28" was incorrectly
scheduled before the "std %r25,0(%r29)" instruction.  This happens in the
20041113-1.c.223r.sched2 pass.  In the previous pass, we had

(note 5 1 50 2 [bb 2] NOTE_INSN_BASIC_BLOCK)
(insn/f 50 5 51 2 (set (mem:DI (plus:DI (reg/f:DI 30 %r30)
                (const_int -16 [0xfffffffffffffff0])) [0 S8 A64])
        (reg:DI 2 %r2))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:4 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 2 %r2)
        (nil)))
(insn/f 51 50 52 2 (set (reg/f:DI 30 %r30)
        (plus:DI (reg/f:DI 30 %r30)
            (const_int 144 [0x90])))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:4 161
{*pa.md:4945}
     (nil))
(insn/f 52 51 53 2 (set (mem:DI (plus:DI (reg/f:DI 30 %r30)
                (const_int -128 [0xffffffffffffff80])) [0 S8 A64])
        (reg:DI 4 %r4))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:4 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 4 %r4)
        (nil)))
(note 53 52 8 2 NOTE_INSN_PROLOGUE_END)
(insn 8 53 9 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -64 [0xffffffffffffffc0])) [0 S8 A64])
        (reg:DI 26 %r26))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 26 %r26)
        (nil)))
(insn 9 8 10 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -56 [0xffffffffffffffc8])) [0 S8 A64])
        (reg:DI 25 %r25))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 25 %r25)
        (nil)))
(insn 10 9 11 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -48 [0xffffffffffffffd0])) [0 S8 A64])
        (reg:DI 24 %r24))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 24 %r24)
        (nil)))
(insn 11 10 12 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -40 [0xffffffffffffffd8])) [0 S8 A64])
        (reg:DI 23 %r23))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 23 %r23)
        (nil)))
(insn 12 11 13 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -32 [0xffffffffffffffe0])) [0 S8 A64])
        (reg:DI 22 %r22))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 22 %r22)
        (nil)))
(insn 13 12 14 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -24 [0xffffffffffffffe8])) [0 S8 A64])
        (reg:DI 21 %r21))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 21 %r21)
        (nil)))
(insn 14 13 15 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -16 [0xfffffffffffffff0])) [0 S8 A64])
        (reg:DI 20 %r20))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 20 %r20)
        (nil)))
(insn 15 14 17 2 (set (mem:DI (plus:DI (reg/f:DI 29 %r29)
                (const_int -8 [0xfffffffffffffff8])) [0 S8 A64])
        (reg:DI 19 %r19))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (expr_list:REG_DEAD (reg:DI 19 %r19)
        (nil)))
(insn 17 15 4 2 (set (reg/f:DI 29 %r29 [76])
        (plus:DI (reg/f:DI 29 %r29)
            (const_int -56 [0xffffffffffffffc8])))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 161
{*pa.md:4945}
     (nil))
(note 4 17 19 2 NOTE_INSN_FUNCTION_BEG)
(insn 19 4 21 2 (set (mem/f/c:DI (plus:DI (reg/f:DI 30 %r30)
                (const_int -136 [0xffffffffffffff78])) [2 MEM[(void * *)&ap]+0
S8 A64])
        (reg/f:DI 29 %r29 [76]))
/test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:7 121
{*pa.md:4105}
     (nil))
(insn 21 19 22 2 (set (reg:DI 28 %r28)
        (zero_extend:DI (mem:SI (plus:DI (reg/f:DI 29 %r29 [76])
                    (const_int 4 [0x4])) [3 MEM[(int *)ap.0_3 + 4B]+0 S4
A32]))) /test/gnu/gcc/gcc/gcc/testsuite/gcc.c-torture/execute/20041113-1.c:8
134 {*pa.md:4637}
     (expr_list:REG_EQUIV (mem:SI (plus:DI (reg/f:DI 29 %r29 [76])
                (const_int 4 [0x4])) [3 MEM[(int *)ap.0_3 + 4B]+0 S4 A32])
        (nil)))

The regression was introduced between revisions 191481 and 191764.

Other similar fails are:

FAIL: gcc.c-torture/execute/920501-8.c execution,  -Os 
FAIL: gcc.c-torture/execute/980608-1.c execution,  -Os 
FAIL: gcc.c-torture/execute/stdarg-4.c execution,  -Os 
FAIL: gcc.c-torture/execute/va-arg-19.c execution,  -Os

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bernds at gcc dot gnu.org

--- Comment #1 from John David Anglin <danglin at gcc dot gnu.org> 2012-10-08 13:56:43 UTC ---
Appears to have been introduced in revision 191493.

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #2 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-10-08 14:18:58 UTC ---
Could you attach both dumps? (and use -fsched-verbose=5)
Did your test include r191838?

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #4 from dave.anglin at bell dot net 2012-10-08 14:45:51 UTC ---
Created attachment 28389
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28389
20041113-1.c.224r.sched2.txt

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #3 from dave.anglin at bell dot net 2012-10-08 14:45:48 UTC ---
On 8-Oct-12, at 10:18 AM, bernds at gcc dot gnu.org wrote:

>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850
>
> --- Comment #2 from Bernd Schmidt <bernds at gcc dot gnu.org>  
> 2012-10-08 14:18:58 UTC ---
> Could you attach both dumps? (and use -fsched-verbose=5)

Done.

> Did your test include r191838?


No but the bug is still present with r191838.  Dumps are from r191493.

--
John David Anglin    dave.anglin@bell.net

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #5 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-10-17 23:38:16 UTC ---
I'm seeing the same thing on cpu2006 benchmark 44.namd on PowerPC64. A load is
being moved above a store to the same location, starting with revision 191493.

Reduced testcase and compile command:

[pthaugen@igoo pr54850]$ cat junk.C
// g++ -S -m64 -O2 -mcpu=power7 junk.C

class Vector {
   public:
     double x,y,z;

     //  v1 += v2;
     inline void operator+=(const Vector &v2) {
       x += v2.x;
       y += v2.y;
       z += v2.z;
     }
};


Vector* foo(Vector *p, Vector *shift, Vector *p1, int n) {
  int i;

  for(i = 0; i < n; i++) {
    p[i] = p1[i];
    p[i] += *shift;
  }
  return p;
}

[pthaugen@igoo pr54850]$ /home/pthaugen/install/gcc/temp/bin/g++ -S -m64 -O2
-mcpu=power7 junk.C


The sched1 dump correctly lists forward dependencies of the initial store(s) to
p1[i] to the subsequent loads of p1[i], but those dependencies are gone in the
.sched2 pass.

>From .sched1  (stores are insn 75,77,79, and dependent loads are insn
80,84,88):

;;   ======================================================
;;   -- basic block 4 from 74 to 96 -- before reload
;;   ======================================================

;;   --------------- forward dependences: ------------

;;   --- Region Dependences --- b 4 bb 0
;;      insn  code    bb   dep  prio  cost   reservation
;;      ----  ----    --   ---  ----  ----   -----------
;;       74   379     4     0    47     2   DU_power7,LSU_power7        : 96 92
91n 87n 83n 79n 77n 75nm
;;       76   379     4     0    45     2   DU_power7,LSU_power7        : 96 92
91n 87n 83n 79n 77nm 75n
;;       78   379     4     0    45     2   DU_power7,LSU_power7        : 96 92
91n 87n 83n 79nm 77n 75n
;;       75   379     4     3    45     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 93 89n 85n 83n 81n 80n
;;       77   379     4     3    31     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 93 89n 87n 85n 84n
;;       79   379     4     3    17     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 93 91n 89n 88n
;;       80   366     4     1    39     2   DU_power7,LSU_power7        : 96 93
83n 82
;;       81   366     4     1    39     2   DU_power7,LSU_power7        : 96
91n 87n 83n 82
;;       82   740     4     2    37     6   DU_power7,VSU_power7        : 96 83
;;       83   366     4     7    31     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 93 89n 85n
;;       84   366     4     1    25     2   DU_power7,LSU_power7        : 96 93
87n 86
;;       85   366     4     3    25     2   DU_power7,LSU_power7        : 96
91n 87n 86
;;       86   740     4     2    23     6   DU_power7,VSU_power7        : 96 87
;;       87   366     4     8    17     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 93 89n
;;       88   366     4     1    11     2   DU_power7,LSU_power7        : 96 93
91n 90



>From .sched2 pass:

;;   ======================================================
;;   -- basic block 4 from 74 to 96 -- after reload
;;   ======================================================

;;   --------------- forward dependences: ------------

;;   --- Region Dependences --- b 4 bb 0
;;      insn  code    bb   dep  prio  cost   reservation
;;      ----  ----    --   ---  ----  ----   -----------
;;       74   379     4     0    50     2   DU_power7,LSU_power7        : 96
91n 87n 83n 92 79n 77n 75nm
;;       76   379     4     0    48     2   DU_power7,LSU_power7        : 96
91n 87n 83n 92 79n 77nm 75n
;;       78   379     4     0    48     2   DU_power7,LSU_power7        : 96
91n 87n 83n 92 79nm 77n 75n
;;       75   379     4     3    48     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 89n 85n 81n 93
;;       77   379     4     3    43     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 89n 85n 93
;;       79   379     4     3    43     6   DU_power7,(LSU_power7+FXU_power7)  
: 96m 89n 93
;;       92    80     4     3     7     1   DU_power7,FXU_power7        : 96 95
;;       95   545     4     1     6     1   DU_power7,FXU_power7        : 96m
;;       93    80     4     3    43     1   DU_power7,FXU_power7        : 96 91
87 83 88 84 80
;;       81   366     4     1    42     3   DU_power7,LSU_power7        : 96
91n 87n 83n 82m
;;       80   366     4     1    42     3   DU_power7,LSU_power7        : 96
83n 88 82
;;       84   366     4     1    27     3   DU_power7,LSU_power7        : 96
87n 86
;;       82   740     4     2    39     6   DU_power7,VSU_power7        : 96 85
83 88
;;       88   366     4     3    12     3   DU_power7,LSU_power7        : 96
91n 90

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #6 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-10-18 01:44:35 UTC ---
> The sched1 dump correctly lists forward dependencies of the initial store(s) to
> p1[i] to the subsequent loads of p1[i], but those dependencies are gone in the
> .sched2 pass.
> 

Sorry, I obviously meant stores/loads of p[i].

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #7 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-10-18 01:50:00 UTC ---
Created attachment 28473
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28473
sched1 dump

[Bug middle-end/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #8 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-10-18 01:51:03 UTC ---
Created attachment 28475
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28475
sched2 dump

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #9 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-10-18 11:54:34 UTC ---
Created attachment 28482
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28482
Candidate patch.

Could you both please test this patch?

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #10 from dave.anglin at bell dot net 2012-10-19 18:33:27 UTC ---
I'm not seeing the fail with the candidate patch.

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #11 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-10-22 15:50:06 UTC ---
(In reply to comment #9)
> Created attachment 28482 [details]
> Candidate patch.
> 
> Could you both please test this patch?

The patch fixes the issue for me too, on both the reduced testcase and the full
benchmark.

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[dje at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

David Edelsohn <dje at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-10-29
                 CC|                            |dje at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #12 from David Edelsohn <dje at gcc dot gnu.org> 2012-10-29 16:59:08 UTC ---
Confirmed.  What's the status of reviewing and committing the patch?

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #13 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-11-08 17:10:34 UTC ---
Author: bernds
Date: Thu Nov  8 17:10:26 2012
New Revision: 193332

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193332
Log:
    PR rtl-optimization/54850
    * sched-deps.c (find_inc): Add all dependencies from the inc_insn
    to the mem_insn.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/sched-deps.c

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[pthaugen at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

--- Comment #14 from Pat Haugen <pthaugen at gcc dot gnu.org> 2012-11-16 00:11:40 UTC ---
Can this be closed? Working fine on powerpc64.

[Bug rtl-optimization/54850] [4.8 Regression] FAIL: gcc.c-torture/execute/20041113-1.c execution, -Os

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54850

Bernd Schmidt <bernds at gcc dot gnu.org> changed:

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

--- Comment #15 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-11-23 12:06:37 UTC ---
Fixed.