[patch] PR33299 (vectorizer wrong code)

[Dorit Nuzman <DORIT at il dot ibm dot com>]

Hi,

When vectorizing reductions we need to replace the uses of the original
reduction variable with the new (vectorized) one. To do this we look for
the corresponding loop-closed-ssa-form phi-node at the loop-exit, and
replace the uses of its PHI_RESULT with the new def. However, the code that
does that expected to find only one such loop-exit phi-node per reduction
variable, which does not have to be the case, as in this example:

<bb 7>:
  # s_4 = PHI <s_3(3)>
  # s_10 = PHI <s_3(3)>
  D.1368_15 = *x_14(D);
  if (D.1368_15 < 0.0)
    goto <bb 8>;
  else
    goto <bb 9>;

<bb 8>:
  s_16 = -s_10;

<bb 9>:
  # s_1 = PHI <s_4(7), s_16(8)>
  return s_1;

s_3 has two loop-closed-ssa-form phi-nodes at the loop-exit, but the
vectorizer fixed the uses just for one of them (s_10) and not the other
(s_4).

We fix this by applying the necessary update to all relevant loop-exit
phis, and not only the first one that we encounter.

Bootstrapped with vectorization enabled on i386-linux. Bootstrapped on
powerpc64-linux. Tested on the vectorizer testcases on both platforms. Will
commit once full regression testing passes.

dorit

ChangeLog:

      * tree-vect-transform.c (vect_create_epilog_for_reduction): Update
      uses for all relevant loop-exit phis, not just the first.

Patch:

(See attached file: pr33299fix.txt)

Attachment: pr33299fix.txt
Description: Text document