[Bug rtl-optimization/74585] powerpc64: Very poor code generation for homogeneous vector aggregates passed in registers

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=74585

--- Comment #9 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
We do optimize things well for the following:

typedef struct
{
  __vector double vx0;
  __vector double vx1;
  __vector double vx2;
  __vector double vx3;
} vdoublex8_t;

vdoublex8_t
test_vecd8_rotate_left (vdoublex8_t a)
{
  vdoublex8_t b;
  b.vx0 = a.vx0;
  b.vx1 = a.vx1;
  b.vx2 = a.vx2;
  b.vx3 = a.vx3;
  return b;
}

At expansion time we have similar code as above, with all the stack stores and
reloads for the argument and the return value, but the optimizer is able to
remove all of that.  So the problem only occurs when referencing part of a
vector.