Hi, Richard,
It seems that the bug is in alpha_align_insns. There is
else if (ofs + len > align)
{
int nop_count = (align - ofs) / 4;
.....
do
emit_insn_before ((*next_nop)(&prev_in_use), where);
while (--nop_count);
But it doesn't handle the cae of nop_count == 0. In that case,
ofs == align == 8 and len == 4.
--
H.J. Lu (hjl@gnu.org)