[Bug middle-end/110089] sub-optimal code for attempting to produce JNA (jump on CF or ZF)

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110089

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |uros at gcc dot gnu.org

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
That said, writing it the way the vectorizer currently emits it is

void bar ();
void foo (unsigned int n, unsigned s)
{
  unsigned oldn;
  do
    {
      bar (n % s);
      oldn = n;
      n = n - s;
    }
  while (oldn > s);
}

which results in (on x86):

.L2:
        movl    %ebx, %eax
        xorl    %edx, %edx
        divl    %ebp
        xorl    %eax, %eax
        movl    %edx, %edi
        call    bar
        movl    %ebx, %eax
        subl    %ebp, %ebx
        cmpl    %eax, %ebp
        jb      .L2

where the sub and the cmp should be combinable.  combine sees the following
but it doesn't try 12 -> 13 (since we set 84 in 13 which is used in 12)
and thus also not 12 -> 13 -> 15.  Also there's no "combine" incentive
for this since the 15 doesn't use 84 (12 -> 15 fails).

(insn 12 11 13 3 (set (reg/v:SI 83 [ n ])
        (reg/v:SI 84 [ n ])) 89 {*movsi_internal}
     (nil))      
(insn 13 12 15 3 (parallel [
            (set (reg/v:SI 84 [ n ])
                (minus:SI (reg/v:SI 84 [ n ])
                    (reg/v:SI 85 [ s ])))
            (clobber (reg:CC 17 flags))
        ]) "t.c":9:9 330 {*subsi_1}
     (expr_list:REG_UNUSED (reg:CC 17 flags)
        (nil)))  
(insn 15 13 16 3 (set (reg:CC 17 flags)
        (compare:CC (reg/v:SI 85 [ s ])
            (reg/v:SI 83 [ n ]))) "t.c":11:15 discrim 1 11 {*cmpsi_1}
     (expr_list:REG_DEAD (reg/v:SI 83 [ n ])
        (nil)))
(jump_insn 16 15 17 3 (set (pc)
        (if_then_else (ltu (reg:CC 17 flags)
                (const_int 0 [0]))
            (label_ref:DI 14)
            (pc))) "t.c":11:15 discrim 1 1002 {*jcc}
     (expr_list:REG_DEAD (reg:CC 17 flags)
        (int_list:REG_BR_PROB 955630228 (nil)))

cmpelim sees basically the same IL:

(insn 12 11 13 3 (set (reg/v:SI 0 ax [orig:83 n ] [83])
        (reg/v:SI 3 bx [orig:84 n ] [84])) 89 {*movsi_internal}
     (nil))
(insn 13 12 15 3 (parallel [
            (set (reg/v:SI 3 bx [orig:84 n ] [84])
                (minus:SI (reg/v:SI 3 bx [orig:84 n ] [84])
                    (reg/v:SI 6 bp [orig:85 s ] [85])))
            (clobber (reg:CC 17 flags))
        ]) "t.c":9:9 330 {*subsi_1}
     (nil))
(insn 15 13 16 3 (set (reg:CC 17 flags)
        (compare:CC (reg/v:SI 6 bp [orig:85 s ] [85])
            (reg/v:SI 0 ax [orig:83 n ] [83]))) "t.c":11:15 discrim 1 11
{*cmpsi_1}
     (nil))
(jump_insn 16 15 25 3 (set (pc)
        (if_then_else (ltu (reg:CC 17 flags)
                (const_int 0 [0]))
            (label_ref:DI 14)
            (pc))) "t.c":11:15 discrim 1 1002 {*jcc}