Testcase:
#define vector __attribute__((vector_size(16) ))
typedef vector int t;

t f(t a)
{
  int b = *(int*)&a;
  return (vector int){b,b,b,b};
}

----------------- CUT --------
This should just produce :
        vspltw v2,v2,0
        blr

But right now we do:
_f:
        addi r2,r1,-16
        stvewx v2,0,r2
        lvewx v2,0,r2
        vspltw v2,v2,0
        blr

Which is not good at all, as it causes a load hit store hazzard to happen on
the Cell and it also causes the dispatch group to be rejected for the
power4/power5/970.

This is related to PR 32107.


-- 
           Summary: vector "extract" and then vector "splat" is not
                    optimized
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
GCC target triplet: powerpc*-*-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32110

Reply via email to