[Bug tree-optimization/71761] missing tailcall optimization

["glisse at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71761

--- Comment #3 from Marc Glisse <glisse at gcc dot gnu.org> ---
For clang, this seems based on size: up to size 16 they get jmp, starting from
17 they get a call.

For gcc, we give up on anything more complicated than a "register":

  /* If the LHS of our call is not just a simple register, we can't
     transform this into a tail or sibling call.  This situation happens,
     in (e.g.) "*p = foo()" where foo returns a struct.  In this case
     we won't have a temporary here, but we need to carry out the side
     effect anyway, so tailcall is impossible.

     ??? In some situations (when the struct is returned in memory via
     invisible argument) we could deal with this, e.g. by passing 'p'
     itself as that argument to foo, but it's too early to do this here,
     and expand_call() will not handle it anyway.  If it ever can, then
     we need to revisit this here, to allow that situation.  */
  if (ass_var && !is_gimple_reg (ass_var))
    return;

The ??? comment is exactly your case.