Change code so it uses 2 adds instead of one FFABS. About 1% faster ff_ac3_bit_alloc_calc_psd on Intel Atom, overall speedup not measurable though. Should have a bigger effect on systems without cmov or with very slow cmov.