+ movh m0, [r2 + 16] + movh m1, [r2 + 16 + 2 * r3] + movh m2, [r4 + 16] + movh m3, [r4 + 16 + 2 * r5] + + punpcklqdq m0, m1 + punpcklqdq m2, m3 Can't this be replaced by movh + movhps? Jason