This was fun to track down

[Torbjorn Granlund]

  Note the following comment in convert_modes; it may apply to any attempt
  to use gen_lowpart:

    /* There is one case that we must handle specially: If we are converting
       a CONST_INT into a mode whose size is twice HOST_BITS_PER_WIDE_INT and
       we are to interpret the constant as unsigned, gen_lowpart will do
       the wrong if the constant appears negative.  What we want to do is
       make the high-order word of the constant zero, not all ones.  */

An `unsigned' constant that appears negative should never be generated.
Such constants should use const_double.

I think the gen_lowpart code connected to that comment is wrong.  When that
code went in, I suggested that the bug in the caller that relied on that
bogus behaviour be fixed instead.

Now gcc's constant representation is ambiguous.