SSE指令集: http://blog.youkuaiyun.com/xieqidong/archive/2008/07/04/2612847.aspx
_mm_insert_ps: http://msdn.microsoft.com/en-us/library/bb514071.aspx
__m128 a, b
a 1.0 -1.0 1.5 105.5
b -5.0 10 -325.0625 81.125
const int sel = 0xD9
D
1101 11(b3) 01(r1) // The D means that b3 will be stored in r1
9
1001 1(r0) 0(r1) 0(r2) 1(r3) // The 9 means that r0 and r3 will be set to 0
__m128 res = _mm_insert_ps(a, b, sel);
r0 r1 r2 r3
res 0.000000 81.125000 1.500000 0.000000
0 b.m128_f32[3] a.m128_f32[2] 0
_MM_MK_INSERTPS_NDX(3,0,0)
c 11 00 b3 -> r0
0 0 0 0 0 r: b3, a2, a3, a4
_MM_MK_INSERTPS_NDX(3,1,0)
d 11 01 b3 -> r1
0 0 0 0 0 r: a1, b3, a3, a4
_MM_MK_INSERTPS_NDX(3,2,0)
e 11 10 b3 -> r2
0 0 0 0 0 r: a1, a2, b3, a4
_MM_MK_INSERTPS_NDX(3,3,0)
f 11 11 b3 -> r3
0 0 0 0 0 r: a1, a2, a3, b3