[Bug tree-optimization/27469] New: zero extension not eliminated

[Bug tree-optimization/27469] New: zero extension not eliminated

[falk at debian dot org]

unsigned parity(unsigned x) {
    x ^= x >> 16;
    x ^= x >> 8;
    x ^= x >> 4;
    x &= 0xf;
    return (0x6996 >> x) & 1;
}

gcc 4.2.0 20060506 produces:
        extwl   a0,0x2,t2
        lda     v0,27030
        xor     t2,a0,t2
        zapnot  t2,0xf,t1 # redundant zero-extension
        srl     t1,0x8,t1
        xor     t1,t2,t1
        zapnot  t1,0xf,t0 # redundant zero-extension
        srl     t0,0x4,t0
        xor     t0,t1,t0
        and     t0,0xf,t0
        sra     v0,t0,v0
        and     v0,0x1,v0

-fsee doesn't change anything here.


-- 
           Summary: zero extension not eliminated
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: falk at debian dot org
 GCC build triplet: alphaev68-unknown-linux-gnu
  GCC host triplet: alphaev68-unknown-linux-gnu
GCC target triplet: alphaev68-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469

[Bug rtl-optimization/27469] zero extension not eliminated

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2006-05-07 18:59 -------
Confirmed.
It does try to do it:
(insn 18 16 19 2 (set (reg:DI 79 [ x.25 ])
        (zero_extend:DI (subreg/s:SI (reg/v:DI 71 [ x.25 ]) 0))) 52
{zero_extendsidi2} (insn_list:REG_DEP_TRUE 16 (nil))
    (nil))

into;
(insn 56 16 54 2 (set (subreg:SI (reg:DI 79 [ x.25 ]) 0)
        (subreg/s:SI (reg/v:DI 71 [ x.25 ]) 0)) -1 (nil)
    (nil))

(insn 54 56 19 2 (set (reg:DI 79 [ x.25 ])
        (zero_extend:DI (subreg:SI (reg:DI 79 [ x.25 ]) 0))) 52
{zero_extendsidi2} (nil)
    (nil))


-- 

pinskia 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         |2006-05-07 18:59:35
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469

[Bug rtl-optimization/27469] zero extension not eliminated

[mattst88 at gmail dot com]

------- Comment #2 from mattst88 at gmail dot com  2009-04-18 15:25 -------
(In reply to comment #0)
> unsigned parity(unsigned x) {
>     x ^= x >> 16;
>     x ^= x >> 8;
>     x ^= x >> 4;
>     x &= 0xf;
>     return (0x6996 >> x) & 1;
> }
> 
> gcc 4.2.0 20060506 produces:
>         extwl   a0,0x2,t2
>         lda     v0,27030
>         xor     t2,a0,t2
>         zapnot  t2,0xf,t1 # redundant zero-extension
>         srl     t1,0x8,t1
>         xor     t1,t2,t1
>         zapnot  t1,0xf,t0 # redundant zero-extension
>         srl     t0,0x4,t0
>         xor     t0,t1,t0
>         and     t0,0xf,t0
>         sra     v0,t0,v0
>         and     v0,0x1,v0
> 
> -fsee doesn't change anything here.
> 

Compaq C generates:
0000000000000000 <parity>:
   0:   82 16 02 4a     srl     a0,0x10,t1
   4:   96 69 bf 20     lda     t4,27030
   8:   02 08 02 46     xor     a0,t1,t1
   c:   83 16 41 48     srl     t1,0x8,t2
  10:   02 08 43 44     xor     t1,t2,t1
  14:   84 96 40 48     srl     t1,0x4,t3
  18:   02 08 44 44     xor     t1,t3,t1
  1c:   02 f0 41 44     and     t1,0xf,t1
  20:   82 07 a2 48     sra     t4,t1,t1
  24:   00 30 40 44     and     t1,0x1,v0
  28:   01 80 fa 6b     ret


-- 

mattst88 at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mattst88 at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469

[Bug rtl-optimization/27469] zero extension not eliminated

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469

[Bug rtl-optimization/27469] zero extension not eliminated

[rth at gcc dot gnu dot org]

------- Comment #3 from rth at gcc dot gnu dot org  2009-12-03 00:32 -------
Gcc 4.5 has regressed this even further:

        extwl $16,2,$1
        xor $1,$16,$16
        addl $31,$16,$16   << signextend
        zapnot $16,15,$1   << zeroextend of signextend
        srl $1,8,$1

Neither extension is needed, of course.


-- 

rth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2006-05-07 18:59:35         |2009-12-03 00:32:32
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469

[Bug rtl-optimization/27469] zero extension not eliminated

[rth at gcc dot gnu dot org]

------- Comment #4 from rth at gcc dot gnu dot org  2009-12-08 00:55 -------
4.5 regression fixed in pr42269.

I don't know that we'll ever be able to fix the original PR.

The problem is that we need to work backward from the "x &= 0xf"
through the xor, and see that the srl doesn't actually use the
garbage high bits that the srl might have shifted in.  And then
again back to the first srl.

Combine can't handle this because there are 4 insns involved,
and it only ever handles merging 3 at a time.  So we'd need some
sort of new pass that would be able to see these extensions are
unneeded.


-- 

rth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27469