这是它的描述:
The main difference between executing CCMATMPY and the above sequence is that
saturation is only performed once at the end and intermediate precision is kept at 34
bits
sat(tmp0_e + tmp1_e)->dst_0
sat(tmp0_o + tmp1_o)->dst_1
sat(tmp2_e + tmp3_e)->dst_2
sat(tmp2_o + tmp3_o)->dst_3
查了好多， 还是弄不明白这个意思.
