[Bug middle-end/24627] New: [4.1 Regression] xntp miscompiled

[Bug middle-end/24627] New: [4.1 Regression] xntp miscompiled

[mueller at kde dot org]

xntp is miscompiled since SVN revision 101841.

will attach testcase in a minute


-- 
           Summary: [4.1 Regression] xntp miscompiled
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: critical
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: mueller at kde dot org
  GCC host triplet: i686-gnu-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[mueller at kde dot org]

------- Comment #1 from mueller at kde dot org  2005-11-01 23:06 -------
Created an attachment (id=10110)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=10110&action=view)
testcase

-O1 or higher exposes the bug. 

-fno-tree-dce or undoing commit 101841 fixes it. 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[steven at gcc dot gnu dot org]

------- Comment #2 from steven at gcc dot gnu dot org  2005-11-01 23:08 -------
-fno-tree-loop-im fixes it too, fwiw.


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-11-01 23:08:47
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[steven at gcc dot gnu dot org]

------- Comment #3 from steven at gcc dot gnu dot org  2005-11-01 23:09 -------
Most likely aliasing related.


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |alias


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[steven at gcc dot gnu dot org]

------- Comment #4 from steven at gcc dot gnu dot org  2005-11-01 23:09 -------
Mark, this is a new wrong-code bug.
Could you look at it and set a priority please.


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mark at codesourcery dot com
           Keywords|                            |wrong-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   GCC host triplet|i686-gnu-linux              |
 GCC target triplet|                            |i686-gnu-linux
   Target Milestone|---                         |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2005-11-01 23:12 -------
Note this is union related as if I remove the unions, it works.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #6 from pinskia at gcc dot gnu dot org  2005-11-01 23:22 -------
Here is a little cleaned up testcase without any do {} while(0);:
extern void abort (void) __attribute__ ((noreturn));

typedef unsigned int u_int32;
typedef struct {
  union {u_int32 Xl_ui;} Ul_i;
  union {u_int32 Xl_uf;} Ul_f;
} l_fp;
void dolfptoa(short ndec)
{
  l_fp work, ftmp;
  work.Ul_i.Xl_ui = 0;
  work.Ul_f.Xl_uf = 0x535f3d8;
  while (ndec > 0) {

    ndec--;
    work.Ul_i.Xl_ui = 0;
    work.Ul_i.Xl_ui <<= 1;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1; (work.Ul_f.Xl_uf) <<= 1;
  //  printf("%d\t", work.Ul_i.Xl_ui);

    ftmp = work;
    (work.Ul_i.Xl_ui) <<= 1;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1; (work.Ul_f.Xl_uf) <<= 1;

    (work.Ul_i.Xl_ui) <<= 1;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1; (work.Ul_f.Xl_uf) <<= 1;
  //  printf("%d\t", work.Ul_i.Xl_ui);
    u_int32 lo_tmp;
    u_int32 hi_tmp;
    lo_tmp = ((work.Ul_f.Xl_uf) & 0xffff) + ((ftmp.Ul_f.Xl_uf) & 0xffff);
    hi_tmp = (((work.Ul_f.Xl_uf) >> 16) & 0xffff) + (((ftmp.Ul_f.Xl_uf) >> 16)
& 0xffff);

    if (lo_tmp & 0x10000)
      hi_tmp++;

    (work.Ul_f.Xl_uf) = ((hi_tmp & 0xffff) << 16) | (lo_tmp & 0xffff);
    (work.Ul_i.Xl_ui) += (ftmp.Ul_i.Xl_ui);

    if (hi_tmp & 0x10000)
     (work.Ul_i.Xl_ui)++;
   // printf("%d\n", work.Ul_i.Xl_ui);
    if (!(work.Ul_i.Xl_ui < 10))
      abort ();
  }
}
int main()
{
  dolfptoa(6);
  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #7 from pinskia at gcc dot gnu dot org  2005-11-01 23:32 -------
wtf:
-  #   SFT.1_22 = V_MUST_DEF <SFT.1_19>;
+  #   SFT.1_22 = V_MUST_DEF <SFT.1_11>;
   work.Ul_i.Xl_ui = D.1298_21;
...
-  # SFT.1_10 = PHI <SFT.1_12(0), SFT.1_9(11)>;

That is just wrong.  DCE is doing something wrong, very wrong.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #8 from pinskia at gcc dot gnu dot org  2005-11-01 23:34 -------
Eliminating unnecessary statements:
Deleting : SFT.1_10 = PHI <SFT.1_12(0), SFT.1_9(11)>;

Deleting : work.Ul_i.Xl_ui = 0;

Deleting : work.Ul_i.Xl_ui = 0;

Deleting : work.Ul_f.Xl_uf = D.1302_36;

Deleting : work.Ul_f.Xl_uf = D.1302_44;


No SFT.1_10 = PHI <SFT.1_12(0), SFT.1_9(11)>; is not useless


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[mmitchel at gcc dot gnu dot org]

------- Comment #9 from mmitchel at gcc dot gnu dot org  2005-11-02 00:55 -------
Miscompilation of a popular package on a major architecture; showstopper.


-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[dnovillo at gcc dot gnu dot org]

-- 

dnovillo at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|dnovillo at redhat dot com  |
         AssignedTo|unassigned at gcc dot gnu   |dnovillo at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2005-11-02 15:47 -------
Here is a further reduced testase:
typedef struct {
  union {unsigned Xl_ui;} Ul_i;
  union {unsigned Xl_uf;} Ul_f;
} l_fp;
void dolfptoa(short ndec)
{
  l_fp work;
  work.Ul_f.Xl_uf = 0x535f3d8;
  while (ndec > 0) {

    ndec--;
    work.Ul_i.Xl_ui = 0;
    work.Ul_i.Xl_ui = 0;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1;
    (work.Ul_f.Xl_uf) <<= 1;

    (work.Ul_i.Xl_ui) <<= 1;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1;
    (work.Ul_f.Xl_uf) <<= 1;

    (work.Ul_i.Xl_ui) <<= 1;

    if (!(work.Ul_i.Xl_ui < 10))
      abort ();
  }
}
int main()
{
  dolfptoa(6);
  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #11 from pinskia at gcc dot gnu dot org  2005-11-02 15:53 -------
The following fails with -O1 -fno-tree-sra.
typedef struct {
  struct {unsigned Xl_ui;} Ul_i;
  struct {unsigned Xl_uf;} Ul_f;
} l_fp;
void dolfptoa(short ndec)
{
  l_fp work;
  work.Ul_f.Xl_uf = 0x535f3d8;
  while (ndec > 0) {

    ndec--;
    work.Ul_i.Xl_ui = 0;
    work.Ul_i.Xl_ui = 0;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1;
    (work.Ul_f.Xl_uf) <<= 1;

    (work.Ul_i.Xl_ui) <<= 1;
    if ((work.Ul_f.Xl_uf) & 0x80000000)
      (work.Ul_i.Xl_ui) |= 0x1;
    (work.Ul_f.Xl_uf) <<= 1;

    (work.Ul_i.Xl_ui) <<= 1;

    if (!(work.Ul_i.Xl_ui < 10))
      abort ();
  }
}
int main()
{
  dolfptoa(6);
  return 0;
}
----

The -fno-tree-sra is need as otherwise SRA actually does it work and renames
the PHIs correctly.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #12 from pinskia at gcc dot gnu dot org  2005-11-02 16:03 -------
Here is another more reduced testcase (still at -O1 -fno-tree-sra):
typedef struct {
  unsigned a;
} l_fp;
void dolfptoa(short ndec)
{
  l_fp work;
  unsigned workUl_fXl_uf;
  work.a = 0x535f3d8;
  while (ndec > 0) {

    ndec--;
    work.a = 0;
    work.a = 0;
    if (workUl_fXl_uf & 0x80000000)
      work.a |= 0x1;
    workUl_fXl_uf <<= 1;

    work.a <<= 1;
    if (workUl_fXl_uf & 0x80000000)
      work.a |= 0x1;
    workUl_fXl_uf <<= 1;

    work.a <<= 1;

    if (!(work.a < 10))
      abort ();
  }
}
int main()
{
  dolfptoa(6);
  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #13 from pinskia at gcc dot gnu dot org  2005-11-02 16:12 -------
(In reply to comment #12)
> Here is another more reduced testcase (still at -O1 -fno-tree-sra):
That testcase is invalid, the one which is valid:
ypedef struct {
  unsigned a;
} l_fp;
void dolfptoa(short ndec)
{
  l_fp work;
  unsigned workUl_fXl_uf;
  workUl_fXl_uf = 0x535f3d8;
  while (ndec > 0) {

    ndec--;
    work.a = 0;
    work.a = 0;
    if (workUl_fXl_uf & 0x80000000)
      work.a |= 0x1;
    workUl_fXl_uf <<= 1;

    work.a <<= 1;
    if (workUl_fXl_uf & 0x80000000)
      work.a |= 0x1;
    workUl_fXl_uf <<= 1;

    work.a <<= 1;

    if (!(work.a < 10))
      abort ();
  }
}
int main()
{
  dolfptoa(8);
  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #14 from pinskia at gcc dot gnu dot org  2005-11-02 16:39 -------
This is as far as I can reduce it, -O1:
typedef union {
  unsigned a;
} l_fp;
int main(void)
{
  l_fp work;
  unsigned workUl_fXl_uf = 0xAAAAAAAA;
  int ndec = 10;
  do {
    ndec--;
    work.a = 0;
    work.a = 0;

    if (workUl_fXl_uf & 1)
      work.a ++;

    workUl_fXl_uf <<= 1;

    work.a ++;
    if (workUl_fXl_uf & 1)
      work.a ++;

    work.a ++;

    if (work.a > 4)
      abort ();
  }while(ndec > 0);
  return 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[pinskia at gcc dot gnu dot org]

------- Comment #15 from pinskia at gcc dot gnu dot org  2005-11-02 17:02 -------
A link to the changes (so I can be a little lazy):
http://gcc.gnu.org/viewcvs?view=rev&rev=101841


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug middle-end/24627] [4.1 Regression] xntp miscompiled

[dnovillo at gcc dot gnu dot org]

------- Comment #16 from dnovillo at gcc dot gnu dot org  2005-11-02 19:40 -------

testing patch


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug tree-optimization/24627] [4.1 Regression] xntp miscompiled

[dnovillo at gcc dot gnu dot org]

------- Comment #17 from dnovillo at gcc dot gnu dot org  2005-11-04 19:56 -------
Subject: Bug 24627

Author: dnovillo
Date: Fri Nov  4 19:56:28 2005
New Revision: 106502

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=106502
Log:

        PR 24627
        * tree-ssa-dce.c (pass_dce, pass_dce_loop, pass_cd_dce): Use
        TODO_update_ssa instead of TODO_update_ssa_no_phi.

testsuite/

        PR 24627
        * gcc.dg/tree-ssa/pr24627.c: New test.


Added:
    trunk/gcc/testsuite/gcc.dg/tree-ssa/pr24627.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-dce.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627

[Bug tree-optimization/24627] [4.1 Regression] xntp miscompiled

[dnovillo at gcc dot gnu dot org]

------- Comment #18 from dnovillo at gcc dot gnu dot org  2005-11-04 19:57 -------
Fixed. http://gcc.gnu.org/ml/gcc-patches/2005-11/msg00239.html


-- 

dnovillo at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24627