Denormalized DOUBLE-s can't be properly handled by old MIPS processors.
So we need to enable specific mode for them so IRIX will do use
software round to handle such numbers. 


sql/mysqld.cc:
  Bug #29085 A small double precision number becomes zero.
  
  reset_floating_point_exeption() renamed as set_proper_floating_point_mode()
  #ifdef __sgi code added to enable denormalized DOUBLE-s on IRIX