[Bug c/38533] New: [4.2/4.3/4.4 regression] Inefficient stack usage

[Bug c/38533] New: [4.2/4.3/4.4 regression] Inefficient stack usage

[h-shimamoto at ct dot jp dot nec dot com]

The stack usage of the code gcc-4.x generated looks inefficient on x86 and
x86_64. A simple test case is below;
========
#define copy_from_asm(x, addr, err)     \
asm volatile(                           \
        "1:\tmovl %2, %1\n"             \
        "2:\n"                          \
        ".section .fixup,\"ax\"\n"      \
        "\txor %1,%1\n"                 \
        "\tmov $1,%0\n"                 \
        "\tjmp 2b\n"                    \
        ".previous\n"                   \
        : "=r" (err), "=r" (x)          \
        : "m" (*(int*)(addr)), "0" (err))

#define copy_from(x, addr, err) do {            \
        (err) = 0;                              \
        copy_from_asm((x), (addr), (err));      \
} while (0)

#define copy(x, addr)   ({              \
        int __err;                      \
        copy_from((x), (addr), __err);  \
        __err;                          \
})

int src[32];
int dst[32];

#define my_copy(x)      do { err |= copy(dst[x], &src[x]); } while (0)

int test(void)
{
        int err = 0;

        my_copy( 0); my_copy( 1); my_copy( 2); my_copy( 3);
        my_copy( 4); my_copy( 5); my_copy( 6); my_copy( 7);
        my_copy( 8); my_copy( 9); my_copy(10); my_copy(11);
        my_copy(12); my_copy(13); my_copy(14); my_copy(15);
        my_copy(16); my_copy(17); my_copy(18); my_copy(19);
        my_copy(20); my_copy(21); my_copy(22); my_copy(23);
        my_copy(24); my_copy(25); my_copy(26); my_copy(27);
        my_copy(28); my_copy(29); my_copy(30); my_copy(31);

        return err;
}
======

I compiled this test case with gcc-3.4.6, 4.2.4, 4.3.2 and 4.4-20081205,
and the compile option is "-g -Os -mno-red-zone".
The code size of objects are below;
$ size test.o.*
   text    data     bss     dec     hex filename
    945       0       0     945     3b1 test.o.34
   1157       0       0    1157     485 test.o.42
   1133       0       0    1133     46d test.o.43
   1201       0       0    1201     4b1 test.o.44

gcc-3.4.6 generates;
0000000000000000 <test>:
   0:   31 c9                   xor    %ecx,%ecx
   2:   8b 05 00 00 00 00       mov    0x0(%rip),%eax        # 8 <test+0x8>
   8:   89 05 00 00 00 00       mov    %eax,0x0(%rip)        # e <test+0xe>
   e:   31 c0                   xor    %eax,%eax
  10:   8b 15 00 00 00 00       mov    0x0(%rip),%edx        # 16 <test+0x16>
  16:   09 c1                   or     %eax,%ecx
  18:   89 15 00 00 00 00       mov    %edx,0x0(%rip)        # 1e <test+0x1e>
  1e:   31 c0                   xor    %eax,%eax
  20:   8b 15 00 00 00 00       mov    0x0(%rip),%edx        # 26 <test+0x26>
  26:   09 c1                   or     %eax,%ecx

gcc-4.4 generates;
0000000000000000 <test>:
   0:   41 57                   push   %r15
   2:   31 c0                   xor    %eax,%eax
   4:   41 56                   push   %r14
   6:   41 55                   push   %r13
   8:   41 89 c5                mov    %eax,%r13d
   b:   41 54                   push   %r12
   d:   55                      push   %rbp
   e:   53                      push   %rbx
   f:   48 83 ec 58             sub    $0x58,%rsp
  13:   8b 15 00 00 00 00       mov    0x0(%rip),%edx        # 19 <test+0x19>
  19:   89 15 00 00 00 00       mov    %edx,0x0(%rip)        # 1f <test+0x1f>
  1f:   41 89 c6                mov    %eax,%r14d
  22:   8b 15 00 00 00 00       mov    0x0(%rip),%edx        # 28 <test+0x28>
  28:   89 15 00 00 00 00       mov    %edx,0x0(%rip)        # 2e <test+0x2e>
  2e:   41 89 c4                mov    %eax,%r12d
...
  bf:   31 d2                   xor    %edx,%edx
  c1:   44 8b 3d 00 00 00 00    mov    0x0(%rip),%r15d        # c8 <test+0xc8>
  c8:   89 54 24 04             mov    %edx,0x4(%rsp)
  cc:   44 89 3d 00 00 00 00    mov    %r15d,0x0(%rip)        # d3 <test+0xd3>
  d3:   45 31 ff                xor    %r15d,%r15d
  d6:   8b 15 00 00 00 00       mov    0x0(%rip),%edx        # dc <test+0xdc>
  dc:   44 89 7c 24 54          mov    %r15d,0x54(%rsp)
...
 26d:   0b 54 24 54             or     0x54(%rsp),%edx
 271:   0b 54 24 50             or     0x50(%rsp),%edx
 275:   0b 54 24 4c             or     0x4c(%rsp),%edx
 279:   0b 54 24 48             or     0x48(%rsp),%edx
...

On gcc-4.x, error values temporally stored to stack and at the last "or" all
stored data. This stack usage seems inefficient.


-- 
           Summary: [4.2/4.3/4.4 regression] Inefficient stack usage
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: h-shimamoto at ct dot jp dot nec dot com
  GCC host triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] Inefficient stack usage

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
          Component|c                           |middle-end
   GCC host triplet|x86_64-unknown-linux-gnu    |
 GCC target triplet|                            |x86_64-unknown-linux-gnu
   Target Milestone|---                         |4.2.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #1 from jakub at gcc dot gnu dot org  2008-12-16 13:49 -------
The culprit is tree-ssa-reassoc.c, with -fno-tree-ssa-reassoc the generated
code is comparable to 3.4.
For some reason it decides:
Transforming err_10 = __err_3 | __err_8;
 into err_10 = __err_8 | __err_3;
(no idea what advantages that has, but anyway) but instead of reinserting
the transformed stmt into the location where err_10 used to be set before,
it moves all the 30 err_A = err_B | __err_C stmts (where each stmt depends on
the preceeding one) after the last asm volatile.  This means that register
pressure jumps through the roof.  In this case (if the transformation actually
makes sense) certainly the first transformed stmt can be inserted into the old
definition spot and all the following stmts can be left untouched.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dberlin at gcc dot gnu dot
                   |                            |org
           Keywords|                            |missed-optimization
            Summary|[4.2/4.3/4.4 regression]    |[4.2/4.3/4.4 regression]
                   |Inefficient stack usage     |tree-ssa-reassoc.c increases
                   |                            |register pressure several
                   |                            |times


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[hjl dot tools at gmail dot com]

------- Comment #2 from hjl dot tools at gmail dot com  2008-12-16 14:58 -------
Sounds similar to PR 32183.


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl dot tools at gmail dot
                   |                            |com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #3 from jakub at gcc dot gnu dot org  2008-12-16 16:57 -------
In particular, the problem is in linearize_expr_tree.  The |s are already
perfectly linearized (in all but the innermost recursive linearize_expr_tree
binlhsisreassoc is 1 and binrhsisreassoc is 0, in the innermost both are 0),
yet:
1645  gsinow = gsi_for_stmt (stmt);
1646  gsilhs = gsi_for_stmt (SSA_NAME_DEF_STMT (binlhs));
1647  gsi_move_before (&gsilhs, &gsinow);
moves all of them to one spot.  Couldn't that moving be deferred till we
actually know it is worthwhile to reassociate something, not blindly just in
case something will be needed?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[dberlin at dberlin dot org]

------- Comment #4 from dberlin at gcc dot gnu dot org  2008-12-16 17:00 -------
Subject: Re:  [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register
pressure several times

Yes, that looks like a bug.
There are also numerous ways in which the placement can be improved.
A few people had talked about rewriting it to be a balanced tree
placement, etc, so i've never bothered to do even the small cleanups.
If you want to change placement, go for it.
:)

On Tue, Dec 16, 2008 at 11:57 AM, jakub at gcc dot gnu dot org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
>
> ------- Comment #3 from jakub at gcc dot gnu dot org  2008-12-16 16:57 -------
> In particular, the problem is in linearize_expr_tree.  The |s are already
> perfectly linearized (in all but the innermost recursive linearize_expr_tree
> binlhsisreassoc is 1 and binrhsisreassoc is 0, in the innermost both are 0),
> yet:
> 1645  gsinow = gsi_for_stmt (stmt);
> 1646  gsilhs = gsi_for_stmt (SSA_NAME_DEF_STMT (binlhs));
> 1647  gsi_move_before (&gsilhs, &gsinow);
> moves all of them to one spot.  Couldn't that moving be deferred till we
> actually know it is worthwhile to reassociate something, not blindly just in
> case something will be needed?
>
>
> --
>
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533
>
> ------- You are receiving this mail because: -------
> You are on the CC list for the bug, or are watching someone who is.
>


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-12-16 23:25:08
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #5 from jakub at gcc dot gnu dot org  2008-12-17 13:06 -------
Created an attachment (id=16915)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16915&action=view)
gcc44-pr38533.patch

Patch that cures this.  Bootstrapped/regtested on x86_64-linux, except for
gfortran.dg/reassoc_4.f failure where the undistribution doesn't happen any
more for some reason.  Looking into it.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #6 from jakub at gcc dot gnu dot org  2008-12-17 16:04 -------
The problem with the patch is that useless insns (visited with has_zero_uses on
lhs) are now scattered in the bbs and so aren't removed immediately.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3/4.4 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #7 from jakub at gcc dot gnu dot org  2008-12-18 07:56 -------
Subject: Bug 38533

Author: jakub
Date: Thu Dec 18 07:54:43 2008
New Revision: 142807

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142807
Log:
        PR middle-end/38533
        * tree-ssa-reassoc.c (remove_visited_stmt_chain): New function.
        (rewrite_expr_tree): Add moved argument, move stmts together if
        needed.  Call remove_visited_stmt_chain.
        (linearize_expr_tree): Don't move stmts here.
        (reassociate_bb): Call remove_visited_stmt_chain if num ops is 1.
        Adjust rewrite_expr_tree caller.

        * gcc.dg/tree-ssa/pr38533.c: New test.
        * gcc.c-torture/execute/pr38533.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr38533.c
    trunk/gcc/testsuite/gcc.dg/tree-ssa/pr38533.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-reassoc.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3 regression] tree-ssa-reassoc.c increases register pressure several times

[jakub at gcc dot gnu dot org]

------- Comment #8 from jakub at gcc dot gnu dot org  2008-12-18 07:56 -------
Fixed on the trunk.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.2/4.3/4.4 regression]    |[4.2/4.3 regression] tree-
                   |tree-ssa-reassoc.c increases|ssa-reassoc.c increases
                   |register pressure several   |register pressure several
                   |times                       |times


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3 regression] tree-ssa-reassoc.c increases register pressure several times

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.2/4.3 regression] tree-ssa-reassoc.c increases register pressure several times

[bonzini at gnu dot org]

------- Comment #9 from bonzini at gnu dot org  2009-02-04 07:44 -------
Did the patch cure PR28481 too?


-- 

bonzini at gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bonzini at gnu dot org
OtherBugsDependingO|                            |28481
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.3 regression] tree-ssa-reassoc.c increases register pressure several times

[jsm28 at gcc dot gnu dot org]

------- Comment #10 from jsm28 at gcc dot gnu dot org  2009-03-31 21:05 -------
Closing 4.2 branch.


-- 

jsm28 at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.2/4.3 regression] tree-  |[4.3 regression] tree-ssa-
                   |ssa-reassoc.c increases     |reassoc.c increases register
                   |register pressure several   |pressure several times
                   |times                       |
   Target Milestone|4.2.5                       |4.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533

[Bug middle-end/38533] [4.3 regression] tree-ssa-reassoc.c increases register pressure several times

[rguenth at gcc dot gnu dot org]

------- Comment #11 from rguenth at gcc dot gnu dot org  2009-04-22 15:19 -------
WONTFIX on the 4.3 branch.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
      Known to fail|                            |4.3.3
      Known to work|                            |4.4.0
         Resolution|                            |FIXED
   Target Milestone|4.3.4                       |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38533