http://gcc.gnu.org/viewcvs/trunk/gcc/config/m68k/m68k.md#rev117181
> * config/m68k/m68k.md (negsf2, negdf2, negxf2): Use
> -2147483647 - 1 instead of 0x80000000.
Fixes "internal compiler error: in do_SUBST" on compiling
floating point ops with -msoft-float or -m68010 on LP64 hosts,
and may also close PR toolchain/38359. Tested on alpha.