[Bug tree-optimization/37579] New: Vectorizer is causing code bloat and worse performance than unrolling would for a loop in SPEC 2k's eon

[pinskia at gcc dot gnu dot org]

Hi,
The following reduced testcase shows the issue:
struct f
{
  float a[8];
};

void set(struct f *a, float b)
{
  int i = 0;
  for(i=0;i<8;i++)
    a->a[i] = b;
}
--- CUT ---
Currently we vectorize this loop when really unrolling would perform better on
SPU, Altivec (PPC) and SSE (x86).
For Cell's PPC, it would cause a LHS as we have to transfer between the
floating point registers and the vector registers.


-- 
           Summary: Vectorizer is causing code bloat and worse performance
                    than unrolling would for a loop in SPEC 2k's eon
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37579