This patch fixes current implementations of the floating-point fused multiply-add/subtract long to/from accumulator instructions.
FMLAL <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Tb>
FMLAL2 <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Tb>
FMLSL <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Tb>
FMLSL2 <Vd>.<Ta>, <Vn>.<Tb>, <Vm>.<Tb>Issue: #2626