[Bug rtl-optimization/49095] Horrible code generation for trivial decrement with test

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49095

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*, i?86-*-*
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization
   Last reconfirmed|                            |2011.05.21 09:52:49
          Component|other                       |rtl-optimization
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-05-21 09:52:49 UTC ---
Confirmed (not using decq is because it is slower for some archs).  On the
tree level we cannot do better than

  D.2722_2 = *argv_1(D);
  D.2723_3 = D.2722_2 + -1;
  *argv_1(D) = D.2723_3;
  if (D.2723_3 == 0B)

because we lack anything like direct operations on memory (and that's good).
On the RTL side combine tries to do

Trying 7, 8 -> 9:
Failed to match this instruction:
(parallel [
        (set (mem/f:DI (reg/v/f:DI 63 [ argv ]) [2 *argv_1(D)+0 S8 A64])
            (plus:DI (mem/f:DI (reg/v/f:DI 63 [ argv ]) [2 *argv_1(D)+0 S8
A64])
                (const_int -1 [0xffffffffffffffff])))
        (set (reg/f:DI 60 [ D.2723 ])
            (plus:DI (mem/f:DI (reg/v/f:DI 63 [ argv ]) [2 *argv_1(D)+0 S8
A64])
                (const_int -1 [0xffffffffffffffff])))
    ])

because we have a use of the decrement result in the comparison.  It doesn't
try to combine this with the comparison though.

So this case is really special ;)  Without the use of the decremented
value we get the desired subq $1, (%rsi).

Manually sinking the store to *argv into the if and the else yields

        movq    (%rsi), %rbx
        subq    $1, %rbx
        je      L4
L2:
        movq    %rbx, (%rsi)
...

L4:
LCFI4:
        movq    $0, (%rsi)
        movq    %rsi, %rdi
        movq    %rsi, 8(%rsp)
        call    _fncall

so at least we then can combine the decrement with the test ...

As usual combine doesn't like stores.