[Bug tree-optimization/32964] New: union cause ineffient code inside loops

[Bug tree-optimization/32964] New: union cause ineffient code inside loops

[pinskia at gcc dot gnu dot org]

Testcase:
union A
{
 float a;
};
float t(float a)
{
  union A a1, a2, a3;
  a1.a = a;
  for(int i =0;i<100;i++)
  {
  a2 = a1;
  a2.a += a;
  a1 = a2;
  }
  a3 = a1;
  return a3.a;
}

--------------
We currently get on powerpc-linux-gnu in the inner loop:
.L2:
        stw 0,8(1)
        lfs 0,8(1)
        fadds 0,1,0
        stfs 0,8(1)
        lwz 0,8(1)
        bdnz .L2

Notice how we are storing and loading from the same address twice in the loop.
That is bad news for the Cell.
This also happens on x86:
.L2:
        flds    -4(%ebp)  <--- load a1.a
        addl    $1, %eax
        fadd    %st(1), %st
        cmpl    $100, %eax
        fstps   -8(%ebp)   <--- store a2.a
        movl    -8(%ebp), %edx  <--- load a2.a
        movl    %edx, -4(%ebp)  <--- store a1.a
        jne     .L2


-- 
           Summary: union cause ineffient code inside loops
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug tree-optimization/32964] [4.3 Regression] union cause ineffient code inside loops

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2007-08-01 23:16 -------
Hmm, the trunk is worse than the 4.2 branch for -msse2 -mfpmath=sse.
Trunk:
.L2:
        movss   -4(%ebp), %xmm1
        addl    $1, %eax
        cmpl    $100, %eax
        addss   %xmm0, %xmm1
        movss   %xmm1, -8(%ebp)
        movl    -8(%ebp), %edx
        movl    %edx, -4(%ebp)
        jne     .L2

4.2.0:
.L2:
        movaps  %xmm2, %xmm0
        addl    $1, %eax
        addss   %xmm1, %xmm0
        cmpl    $100, %eax
        movaps  %xmm0, %xmm1
        jne     .L2

So I guess I can mark this as a regression :).


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|union cause ineffient code  |[4.3 Regression] union cause
                   |inside loops                |ineffient code inside loops
   Target Milestone|---                         |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug tree-optimization/32964] [4.3 Regression] union cause ineffient code inside loops

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2007-08-01 23:18 -------
One should note the original testcase included a vector float and a struct
inside the union and used C++, this is just a reduced testcase of the issue.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug tree-optimization/32964] [4.3 Regression] union cause ineffient code inside loops

[mmitchel at gcc dot gnu dot org]

-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3 Regression] union cause ineffient code inside loops

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2007-11-03 18:13 -------
Ok, looking a little into this shows this is the same issue (or at least
related to) as PR 33989.  We have SImode in some places where we could use
SFmode.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2008-01-02 14:11 -------
On the tree level there is no visible regression, but we have V_MUST_DEFs and
a single SFT on the 4.2 branch while trunk does not create an extra virtual tag
for the single-field union:

<bb 2>:
  # a1_12 = VDEF <a1_11(D)>
  a1.a = a;
  i = 0;

<bb 3>:
  # VUSE <a1_23>
  # a2_14 = VDEF <a2_24>
  a2 = a1;
  # a2_15 = VDEF <a2_14>
  a2.a = a2.a + a;
  # VUSE <a2_15>
  # a1_16 = VDEF <a1_23>
  a1 = a2;
  i = i + 1;
  if (i != 100)
    goto <bb 3>;
  else
    goto <bb 4>;

but, confirmed, there is a code quality regression as noted compared to 4.2.

Also SRA could be extended to scalarize single-field unions.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
      Known to fail|                            |4.3.0
      Known to work|                            |4.2.2
   Last reconfirmed|0000-00-00 00:00:00         |2008-01-02 14:11:21
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2008-01-04 16:59 -------
The plain DECL does not work because

  1) we don't do the storetmp
  2) can_PRE_operation and can_value_number_operation return false (so we
     do not enter a into EXP_GEN
  3) create_value_expr_from cannot handle it

most of the rest of PRE looks like it may handle DECLs, but if you fix the
above, vn_lookup_with_vuses still not finds VH.10, because it is not ANTIC_IN.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2008-01-04 16:59 -------
Doh, wrong bug.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2008-03-14 16:57 -------
This could be fixed by MEM_REF (but we don't lower struct copies at the moment,
doh!), or by tweaking SCCVN/FRE to also look at struct copies.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.0                       |4.3.1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2008-06-06 14:57 -------
4.3.1 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.1                       |4.3.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[jsm28 at gcc dot gnu dot org]

------- Comment #9 from jsm28 at gcc dot gnu dot org  2008-08-27 22:02 -------
4.3.2 is released, changing milestones to 4.3.3.


-- 

jsm28 at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.2                       |4.3.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #10 from rguenth at gcc dot gnu dot org  2009-01-24 10:19 -------
GCC 4.3.3 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.3                       |4.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[bonzini at gnu dot org]

------- Comment #11 from bonzini at gnu dot org  2009-02-01 08:07 -------
still present.


-- 

bonzini at gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2008-03-25 03:46:33         |2009-02-01 08:07:23
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2009-02-01 10:40 -------
New SRA should fix this.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mjambor at suse dot cz


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[jamborm at gcc dot gnu dot org]

------- Comment #13 from jamborm at gcc dot gnu dot org  2009-02-06 21:45 -------
Indeed, this testcase is fully scalarized by the new SRA even as it is
today.  (If you don't know what new SRA I mean, I hope I'll post it to
the list as an RFC next week).


-- 

jamborm at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jamborm at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4/4.5 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #14 from rguenth at gcc dot gnu dot org  2009-05-22 21:06 -------
If not something for SRA then this needs PRE.  But then phi-translation needs
to be able to translate through struct copies as well.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4/4.5 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #15 from rguenth at gcc dot gnu dot org  2009-08-04 12:28 -------
GCC 4.3.4 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.4                       |4.3.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4/4.5 Regression] union cause inefficient code inside loops

[jamborm at gcc dot gnu dot org]

------- Comment #16 from jamborm at gcc dot gnu dot org  2009-08-06 11:55 -------
Subject: Bug 32964

Author: jamborm
Date: Thu Aug  6 11:55:30 2009
New Revision: 150523

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=150523
Log:
2009-08-06  Martin Jambor  <mjambor@suse.cz>

        PR middle-end/32964
        * testsuite/gcc.dg/tree-ssa/pr32964.c: New test.



Added:
    trunk/gcc/testsuite/gcc.dg/tree-ssa/pr32964.c
Modified:
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[jamborm at gcc dot gnu dot org]

------- Comment #17 from jamborm at gcc dot gnu dot org  2009-08-06 12:09 -------
New SRA scalarizes the unions in this testcase and so this is no
longer a 4.5 regression.  I have committed a testcase for reference
and am updating relevant bugzilla tags/summary now.

As I won't be backporting new SRA to any of the regressing versions, I
am no longer really interested in this bug and will remove myself from
the CC list too.  Feel free to re-add me if anything I should know
about arises.


-- 

jamborm at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|mjambor at suse dot cz,     |
                   |jamborm at gcc dot gnu dot  |
                   |org                         |
      Known to work|4.2.2                       |4.2.2 4.5.0
            Summary|[4.3/4.4/4.5 Regression]    |[4.3/4.4 Regression] union
                   |union cause inefficient code|cause inefficient code
                   |inside loops                |inside loops


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964

[Bug middle-end/32964] [4.3/4.4 Regression] union cause inefficient code inside loops

[rguenth at gcc dot gnu dot org]

------- Comment #18 from rguenth at gcc dot gnu dot org  2010-05-22 18:11 -------
GCC 4.3.5 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.5                       |4.3.6


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32964