[Bug middle-end/43958] New: FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[Bug middle-end/43958] New: FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[danglin at gcc dot gnu dot org]

Executing on host: /test/gnu/gcc/objdir/gcc/xgcc -B/test/gnu/gcc/objdir/gcc/
/te
st/gnu/gcc/gcc/gcc/testsuite/gcc.dg/ipa/ipa-pta-10.c   -O2 -fipa-pta
-fdump-ipa-
pta-details  -lm   -o ./ipa-pta-10.exe    (timeout = 300)
PASS: gcc.dg/ipa/ipa-pta-10.c (test for excess errors)
Setting LD_LIBRARY_PATH to :/test/gnu/gcc/objdir/gcc::/test/gnu/gcc/objdir/gcc
PASS: gcc.dg/ipa/ipa-pta-10.c execution test
FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

Generating constraints for global initializers

ANYTHING = &ANYTHING
READONLY = &READONLY
ESCAPED = *ESCAPED
ESCAPED = ESCAPED + UNKNOWN
*ESCAPED = NONLOCAL
NONLOCAL = &NONLOCAL
NONLOCAL = &ESCAPED
INTEGER = &ANYTHING

Generating constraints for main

foo.varargs = &i
main.clobber = foo.clobber
main.use = foo.use
main.use = ESCAPED

Generating constraints for foo

ap = &ANYTHING
CALLCLOBBERED = &ap
ap.1_1 = ap
D.1970_2 = ap.1_1 + -32
ESCAPED = D.1970_2
ap.2_5 = &ANYTHING
ap = ap.2_5
ap.1_6 = ap
ap.3_7 = ap.1_6
p_8 = *ap.3_7
foo.use = ap.3_7
foo.clobber = p_8


Collapsing static cycles and doing variable substitution
Building predecessor graph
Detecting pointer and location equivalences
Found location equivalence for node NONLOCAL
Equivalence classes for Indirect node node id 0:NULL are pointer: 1, location:0
Equivalence classes for Indirect node node id 1:ANYTHING are pointer: 2,
locatio
n:1
Equivalence classes for Indirect node node id 2:READONLY are pointer: 3,
locatio
n:2
Equivalence classes for Indirect node node id 3:ESCAPED are pointer: 7,
location
:3
Equivalence classes for Indirect node node id 4:NONLOCAL are pointer: 5,
locatio
n:3
Equivalence classes for Direct node node id 5:STOREDANYTHING are pointer: 0,
loc
ation:0
Equivalence classes for Indirect node node id 6:INTEGER are pointer: 8,
location
:0
Equivalence classes for Direct node node id 7:main are pointer: 0, location:0
Equivalence classes for Direct node node id 8:main.clobber are pointer: 9,
locat
ion:0
Equivalence classes for Direct node node id 9:main.use are pointer: 12,
location
:0
Equivalence classes for Direct node node id 10:main.result are pointer: 0,
locat
ion:0
Equivalence classes for Direct node node id 11:main.varargs are pointer: 0,
loca
tion:0
Equivalence classes for Direct node node id 12:foo are pointer: 0, location:0
Equivalence classes for Direct node node id 13:foo.clobber are pointer: 9,
locat
ion:0
Equivalence classes for Direct node node id 14:foo.use are pointer: 11,
location
:0
Equivalence classes for Direct node node id 15:foo.result are pointer: 0,
locati
on:0
Equivalence classes for Direct node node id 16:foo.arg0 are pointer: 0,
location
:0
Equivalence classes for Direct node node id 17:foo.varargs are pointer: 13,
loca
tion:0
Equivalence classes for Indirect node node id 18:i are pointer: 14, location:4
Equivalence classes for Direct node node id 19:abort are pointer: 0, location:0
Equivalence classes for Indirect node node id 20:ap are pointer: 11, location:5
Equivalence classes for Direct node node id 21:CALLUSED are pointer: 0,
location
:0
Equivalence classes for Direct node node id 22:CALLCLOBBERED are pointer: 15,
lo
cation:0
Equivalence classes for Direct node node id 23:ap.1_1 are pointer: 11,
location:
0
Equivalence classes for Indirect node node id 24:D.1970_2 are pointer: 4,
locati
on:0
Equivalence classes for Direct node node id 25:ap.2_5 are pointer: 10,
location:
0
Equivalence classes for Direct node node id 26:ap.1_6 are pointer: 11,
location:
0
Equivalence classes for Direct node node id 27:ap.3_7 are pointer: 11,
location:
0
Equivalence classes for Direct node node id 28:p_8 are pointer: 9, location:0
STOREDANYTHING is a non-pointer variable, eliminating edges.
main is a non-pointer variable, eliminating edges.
main.result is a non-pointer variable, eliminating edges.
main.varargs is a non-pointer variable, eliminating edges.
foo is a non-pointer variable, eliminating edges.
foo.result is a non-pointer variable, eliminating edges.
foo.arg0 is a non-pointer variable, eliminating edges.
abort is a non-pointer variable, eliminating edges.
CALLUSED is a non-pointer variable, eliminating edges.
Rewriting constraints and unifying variables
Unifying foo.clobber to main.clobber
Unifying ap.1_1 to foo.use
Unifying ap.1_6 to foo.use
Unifying ap.3_7 to foo.use
Unifying p_8 to main.clobber
Uniting pointer but not location equivalent variables
Unifying ap to foo.use
Finding indirect cycles
Solving graph

Points-to sets

ANYTHING = { ANYTHING }
READONLY = { READONLY }
ESCAPED = { ANYTHING ESCAPED NONLOCAL }
NONLOCAL = { ESCAPED NONLOCAL }
STOREDANYTHING = { ESCAPED NONLOCAL }
INTEGER = { ANYTHING }
main.clobber = { ANYTHING ESCAPED NONLOCAL }
main.use = { ANYTHING ESCAPED NONLOCAL }
main.varargs = { }
foo.clobber = { ANYTHING ESCAPED NONLOCAL } same as main.clobber
foo.use = { ANYTHING ESCAPED NONLOCAL }
foo.varargs = { i }
ap = { ANYTHING ESCAPED NONLOCAL } same as foo.use
CALLUSED = { }
CALLCLOBBERED = { ap }
ap.1_1 = { ANYTHING ESCAPED NONLOCAL } same as foo.use
D.1970_2 = { ANYTHING ESCAPED NONLOCAL }
ap.2_5 = { ANYTHING }
ap.1_6 = { ANYTHING ESCAPED NONLOCAL } same as foo.use
ap.3_7 = { ANYTHING ESCAPED NONLOCAL } same as foo.use
p_8 = { ANYTHING ESCAPED NONLOCAL } same as main.clobber


-- 
           Summary: FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED
                    = { }"
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: danglin at gcc dot gnu dot org
 GCC build triplet: hppa2.0w-hp-hpux11.11
  GCC host triplet: hppa2.0w-hp-hpux11.11
GCC target triplet: hppa2.0w-hp-hpux11.11


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug middle-end/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2010-05-02 15:28 -------
Can you attach preprocessed source so i can use a cross for examination?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug middle-end/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2010-05-02 15:28 -------
Hm.  I guess hppa-hpux has some strange varargs representation (like a type
with no pointers?)


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rguenth at gcc dot gnu dot
                   |dot org                     |org
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-05-02 15:28:17
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug middle-end/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #3 from dave at hiauly1 dot hia dot nrc dot ca  2010-05-04 00:27 -------
Subject: Re:  FAIL: gcc.dg/ipa/ipa-pta-10.c
        scan-ipa-dump pta "ESCAPED = { }"

Attached preprocessed source.

Dave


------- Comment #4 from dave at hiauly1 dot hia dot nrc dot ca  2010-05-04 00:27 -------
Created an attachment (id=20553)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20553&action=view)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug middle-end/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2010-05-05 14:24 -------
Ah, we have

  __builtin_va_start (&ap, 0);
  ap.1_1 = ap;
  D.2000_2 = ap.1_1 + -4;
  D.2001_3 = (long unsigned int) D.2000_2;
  D.2002_4 = D.2001_3 & -4;
  ap.2_5 = (void *) D.2002_4;

From

      /* Copied from va-pa.h, but we probably don't need to align to
         word size, since we generate and preserve that invariant.  */
      u = size_int (size > 4 ? -8 : -4);
      t = fold_convert (sizetype, t);
      t = build2 (BIT_AND_EXPR, sizetype, t, u);
      t = fold_convert (valist_type, t);

and the conversion to integer makes PTA drop out and consider
ap.2_5 point to anything and thus we _do_ let stuff escape through it.
In fact this causes everything to escape anywhere.

Thus it's a missed optimization for targets doing this kind of
re-alignment.

But maybe it's not necessary anyway as the comment in pa.c suggests?


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2010-05-21 10:31 -------
target or testsuite issue.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org
         AssignedTo|rguenth at gcc dot gnu dot  |unassigned at gcc dot gnu
                   |org                         |dot org
             Status|ASSIGNED                    |NEW
          Component|middle-end                  |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[danglin at gcc dot gnu dot org]

------- Comment #7 from danglin at gcc dot gnu dot org  2010-06-30 00:24 -------
The re-alignment is necessary, so this is a testsuite issue.  The comment
needs adjustment. 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2010-06-30 08:27 -------
Can you, instead of

      /* Copied from va-pa.h, but we probably don't need to align to
         word size, since we generate and preserve that invariant.  */
      u = size_int (size > 4 ? -8 : -4);
      t = fold_convert (sizetype, t);
      t = build2 (BIT_AND_EXPR, sizetype, t, u);
      t = fold_convert (valist_type, t);

use

      u = build_int_cst (TREE_TYPE (t), (HOST_WIDE_INT)(size > 4 ? -8 : -4));
      t = build2 (BIT_AND_EXPR, TREE_TYPE (t), t, u);
      t = fold_convert (valist_type, t);

thus, perform the AND in pointer types?  This is what I am going to do
to replace REALIGN_INDIRECT_REF with and teach PTA about this (so to not
pessimize hppa here).

Mine for now - I'll ping you when infrastructure is ready to deal with
pointer BIT_AND_EXPRs.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rguenth at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2010-05-02 15:28:17         |2010-06-30 08:27:49
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #9 from dave at hiauly1 dot hia dot nrc dot ca  2010-06-30 14:06 -------
Subject: Re:  FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

> Can you, instead of
> 
>       /* Copied from va-pa.h, but we probably don't need to align to
>          word size, since we generate and preserve that invariant.  */
>       u = size_int (size > 4 ? -8 : -4);
>       t = fold_convert (sizetype, t);
>       t = build2 (BIT_AND_EXPR, sizetype, t, u);
>       t = fold_convert (valist_type, t);
> 
> use
> 
>       u = build_int_cst (TREE_TYPE (t), (HOST_WIDE_INT)(size > 4 ? -8 : -4));
>       t = build2 (BIT_AND_EXPR, TREE_TYPE (t), t, u);
>       t = fold_convert (valist_type, t);
> 
> thus, perform the AND in pointer types?  This is what I am going to do
> to replace REALIGN_INDIRECT_REF with and teach PTA about this (so to not
> pessimize hppa here).

I'll test this change in my next build tonight.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #10 from rguenth at gcc dot gnu dot org  2010-06-30 14:13 -------
Created an attachment (id=21043)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21043&action=view)
additional patch

You need the following additional patch.  With that the testcase should pass.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #11 from dave at hiauly1 dot hia dot nrc dot ca  2010-07-02 15:50 -------
Subject: Re:  FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump
        pta "ESCAPED = { }"

On Wed, 30 Jun 2010, rguenth at gcc dot gnu dot org wrote:

> You need the following additional patch.  With that the testcase should pass.

The testsuite now passes.  Tested on hppa2.0w-hp-hpux11.11.  I've attached
the change as tested.

Dave


------- Comment #12 from dave at hiauly1 dot hia dot nrc dot ca  2010-07-02 15:50 -------
Created an attachment (id=21069)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21069&action=view)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[rguenth at gcc dot gnu dot org]

------- Comment #13 from rguenth at gcc dot gnu dot org  2010-07-02 15:56 -------
I have committed the middle-end pieces already today, so if you can approve and
commit the pa parts that would be nice.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|rguenth at gcc dot gnu dot  |danglin at gcc dot gnu dot
                   |org                         |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[danglin at gcc dot gnu dot org]

------- Comment #14 from danglin at gcc dot gnu dot org  2010-07-02 20:52 -------
Subject: Bug 43958

Author: danglin
Date: Fri Jul  2 20:51:58 2010
New Revision: 161746

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=161746
Log:
        PR target/43958
        * config/pa/pa.c (hppa_gimplify_va_arg_expr): Use pointer arithmetic
        for argument alignment.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/pa/pa.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958

[Bug target/43958] FAIL: gcc.dg/ipa/ipa-pta-10.c scan-ipa-dump pta "ESCAPED = { }"

[danglin at gcc dot gnu dot org]

------- Comment #15 from danglin at gcc dot gnu dot org  2010-07-02 20:58 -------
Fixed.


-- 

danglin at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43958