Bug 97138 - cross-BB vectorization opportunity
Summary: cross-BB vectorization opportunity
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: tree-optimization (show other bugs)
Version: 11.0
: P3 normal
Target Milestone: ---
Assignee: Not yet assigned to anyone
URL:
Keywords: missed-optimization
Depends on:
Blocks: vectorizer
  Show dependency tree / graph
 
Reported: 2020-09-21 10:21 UTC by Richard Biener
Modified: 2020-12-08 09:52 UTC (History)
1 user (show)

See Also:
Host:
Target:
Build:
Known to work:
Known to fail:
Last reconfirmed:

Attachments
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Richard Biener 2020-09-21 10:21:12 UTC 
double a[1024], b[1024];
void bar();
void foo(double x)
{
  double a0 = a[0];
  double a1 = a[1];
  for (int i = 0; i < 511;)
    {
      b[2*i] = a0 + x;
      b[2*i + 1] = a1 + x;
      i++;
      bar ();
      a0 = a[2*i];
      a1 = a[2*i+1];
    }
}

can be BB vectorized but that needs handling of PHI nodes and multiple BBs
for an optimal result.
Comment 1 Richard Biener 2020-12-08 09:52:48 UTC 
Fixed for GCC11:

  <bb 2> [local count: 10737416]:
  _8 = {x_14(D), x_14(D)};
  vect_a0_12.8_24 = MEM <vector(2) double> [(double *)&a];

  <bb 3> [local count: 1063004409]:
  # vect_a0_26.9_23 = PHI <vect_a0_19.5_25(3), vect_a0_12.8_24(2)>
  # ivtmp.23_16 = PHI <ivtmp.23_15(3), 0(2)>
  vect__2.10_22 = _8 + vect_a0_26.9_23;
  MEM <vector(2) double> [(double *)&b + ivtmp.23_16 * 1] = vect__2.10_22;
  bar ();
  vect_a0_19.5_25 = MEM <vector(2) double> [(double *)&a + 16B + ivtmp.23_16 * 1];
  ivtmp.23_15 = ivtmp.23_16 + 16;
  if (ivtmp.23_15 != 8176)
    goto <bb 3>; [98.99%]
  else
    goto <bb 4>; [1.01%]

  <bb 4> [local count: 10737416]:
  return;