[Bug rtl-optimization/47866] New: gcc.dg/torture/vector-2.c fails on IA64

[Bug rtl-optimization/47866] New: gcc.dg/torture/vector-2.c fails on IA64

[sje at cup dot hp.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

           Summary: gcc.dg/torture/vector-2.c fails on IA64
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: sje@cup.hp.com


gcc.dg/torture/vector-2.c is failing on Linux and HP-UX IA64 platforms since
r165240 which is Richard Henderson's fix for PR rtl-opt/33721.  This may be a
target specific bug but the change that caused the failure to show up is not
target specific.

gcc.dg/torture/vector-2.c fails at -O2, but passes at -O0 or -O1.  With the
latest sources I can use this cutdown test case to see the difference:

#define vector __attribute__((vector_size(16) ))

vector int f0(vector int t, int a)
{
 ((int*)&t)[0] = a;
 return t;
}
vector int f1(vector int t, int a)
{
 ((int*)&t)[1] = a;
 return t;
}
int main(void)
{
  vector int a = {0, 0, 0, 0};
  vector int a0;
  a0 = f0(a, 1);
  printf("%d %d %d %d\n", a0[0], a0[1], a0[2], a0[3]);
  a0 = f1(a, 1);
  printf("%d %d %d %d\n", a0[0], a0[1], a0[2], a0[3]);
  return 0;
}


At -O0 or -O1 it prints:

1 0 0 0
0 1 0 0

At -O2 it prints:

1 0 0 0
0 0 0 0

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.03.06 19:07:33
   Target Milestone|---                         |4.5.3
            Summary|gcc.dg/torture/vector-2.c   |[4.5/4.6 Regression]
                   |fails on IA64               |gcc.dg/torture/vector-2.c
                   |                            |fails on IA64
     Ever Confirmed|0                           |1

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2011-03-06 19:07:33 UTC ---
Confirmed, a regression due to the failure in the testsuite.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Target|                            |ia64-*-*
           Priority|P3                          |P2

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-08 17:05:07 UTC ---
I think the bug here is during expansion.
We are expanding:
  vector(4) int t;
  vector(4) int t;
  vector(4) int a0;
  int D.2001;
  int D.2000;
  int D.1999;
  int D.1998;

<bb 2>:
  t = { 0, 0, 0, 0 };
  BIT_FIELD_REF <t, 32, 0> = 1;
  a0_18 = t;
  D.1998_3 = BIT_FIELD_REF <a0_18, 32, 0>;
  D.1999_4 = BIT_FIELD_REF <a0_18, 32, 32>;
  D.2000_5 = BIT_FIELD_REF <a0_18, 32, 64>;
  D.2001_6 = BIT_FIELD_REF <a0_18, 32, 96>;
  printf ("%d %d %d %d\n", D.1998_3, D.1999_4, D.2000_5, D.2001_6);
  t = { 0, 0, 0, 0 };
  BIT_FIELD_REF <t, 32, 32> = 1;
  a0_19 = t;
  D.1998_8 = BIT_FIELD_REF <a0_19, 32, 0>;
  D.1999_9 = BIT_FIELD_REF <a0_19, 32, 32>;
  D.2000_10 = BIT_FIELD_REF <a0_19, 32, 64>;
  D.2001_11 = BIT_FIELD_REF <a0_19, 32, 96>;
  printf ("%d %d %d %d\n", D.1998_8, D.1999_9, D.2000_10, D.2001_11);

All the memory references are MEM_SCALAR_P, except that the stores
BIT_FIELD_REF <t, 32, 0> = 1 and BIT_FIELD_REF <t, 32, 32> = 1 use
MEM_IN_STRUCT_P because store_field decides to override it (set_mem_attributes
doesn't override when it already has one or another one, but store_field does).
This, together with the fact that BIT_FIELD_REF <t, 32, 32> = 1 store
happens to use an address from post_modify and aliasing doesn't figure out it
is sfp based and what is the offset results in
fixed_scalar_and_varying_struct_p
returning true (where fixed scalar is e.g. the BIT_FIELD_REF <a0_19, 64, 0>
or BIT_FIELD_REF <a0_19, 64, 64> access from which BIT_FIELD_REF <a0_19, 32,
XX>
is derived, and "varying" "struct" is the BIT_FIELD_REF <t, 32, 32> = 1 store.
Thus sched1 reorders the reads after the store.

--- expr.c      (revision 170779)
+++ expr.c      (working copy)
@@ -5924,7 +5924,8 @@ store_field (rtx target, HOST_WIDE_INT b
       if (to_rtx == target)
        to_rtx = copy_rtx (to_rtx);

-      MEM_SET_IN_STRUCT_P (to_rtx, 1);
+      if (!MEM_SCALAR_P (to_rtx))
+       MEM_SET_IN_STRUCT_P (to_rtx, 1);
       if (!MEM_KEEP_ALIAS_SET_P (to_rtx) && MEM_ALIAS_SET (to_rtx) != 0)
        set_mem_alias_set (to_rtx, alias_set);

fixes this, though the generated code is still very much suboptimal (uses ld8.a
before the store (though, in the same insn group), thus I think it will always
find out that it changed and branch to a slow path doing the load again.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-08 17:40:53 UTC ---
The reason why aliasing believes reg 356 is varying is that sched1 doesn't use
cselib.  sched2 should be to figure out.  Why it generates ld8.a is unclear to
me, but at least that is not a wrong-code issue.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu.org,
                   |                            |mark at gcc dot gnu.org

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-08 18:41:04 UTC ---
Seems that MEM_SET_IN_STRUCT_P use in there has been introduced by Mark in
http://gcc.gnu.org/ml/gcc-patches/1999-01/msg00234.html
Any idea why you think it should be unconditional in that case?  BIT_FIELD_REF
on a scalar doesn't make it a struct IMHO.
Though, perhaps if (!MEM_SCALAR_P (to_rtx)) MEM_SET_IN_STRUCT_P (to_rtx, 1);
is better written as if (!MEM_SCALAR_P (to_rtx)) MEM_IN_STRUCT_P (to_rtx) = 1;
CCing Mark who wrote that patch and Jeff who reviewed it.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[mark at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

Mark Wielaard <mark at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|mark at gcc dot gnu.org     |mark at codesourcery dot
                   |                            |com

--- Comment #5 from Mark Wielaard <mark at gcc dot gnu.org> 2011-03-08 23:29:42 UTC ---
(In reply to comment #4)
> Seems that MEM_SET_IN_STRUCT_P use in there has been introduced by Mark in
> http://gcc.gnu.org/ml/gcc-patches/1999-01/msg00234.html

Different mark. Changed CC.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[sje at cup dot hp.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

--- Comment #6 from Steve Ellcey <sje at cup dot hp.com> 2011-03-09 17:26:44 UTC ---
I tried the patch in comment #2 and verified that it fixes vector-2.c on IA64
HP-UX and Linux and caused no regressions.

[Bug rtl-optimization/47866] [4.5/4.6 Regression] gcc.dg/torture/vector-2.c fails on IA64

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-09 20:49:03 UTC ---
Author: jakub
Date: Wed Mar  9 20:49:00 2011
New Revision: 170834

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=170834
Log:
    PR rtl-optimization/47866
    * expr.c (store_field): If MEM_SCALAR_P (target), don't use
    MEM_SET_IN_STRUCT_P (to_rtx, 1), just set MEM_IN_STRUCT_P (to_rtx)
    if target wasn't scalar.
    * function.c (assign_stack_temp_for_type): Assert that neither
    MEM_SCALAR_P nor MEM_IN_STRUCT_P is set previously, set either
    MEM_IN_STRUCT_P or MEM_SCALAR_P instead of using MEM_SET_IN_STRUCT_P
    macro.
    * rtl.h (MEM_SET_IN_STRUCT_P): Removed.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expr.c
    trunk/gcc/function.c
    trunk/gcc/rtl.h

[Bug rtl-optimization/47866] [4.5 Regression] gcc.dg/torture/vector-2.c fails on IA64

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.5/4.6 Regression]        |[4.5 Regression]
                   |gcc.dg/torture/vector-2.c   |gcc.dg/torture/vector-2.c
                   |fails on IA64               |fails on IA64

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-09 20:51:45 UTC ---
Should be fixed on the trunk, ia64-linux bootstrap/regtest just finished.

[Bug rtl-optimization/47866] [4.5 Regression] gcc.dg/torture/vector-2.c fails on IA64

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.5.3                       |4.5.4

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-28 14:51:41 UTC ---
GCC 4.5.3 is being released, adjusting target milestone.

[Bug rtl-optimization/47866] [4.5 Regression] gcc.dg/torture/vector-2.c fails on IA64

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47866

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.5.4                       |4.6.0

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-07-02 10:29:38 UTC ---
Fixed in 4.6.0, the 4.5 branch is being closed.