[Bug middle-end/54400] recognize vector reductions

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54400

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, with vectorizer support we get

  <bb 2> [local count: 1073741824]:
  _7 = .REDUC_PLUS (v_3(D)); [tail call]
  return _7;

but

f:
.LFB5669:
        .cfi_startproc
        movapd  %xmm0, %xmm1
        unpckhpd        %xmm0, %xmm0
        addpd   %xmm1, %xmm0
        ret

(note avoiding hadd in the reduc pattern was intended).  Not sure if
two element reduction vectorization is worthwhile - my current patch
bails w/o -fassociative-math (which is supposedly unnecessary for
two elements).

But mine anyway.