使用SSE指令优化的数学函数(整理)

float _SSE_cos( float x)
{
 float temp;
 __asm
 {
  movss xmm0, x
  movss xmm1, _ps_am_inv_sign_mask
  andps xmm0, xmm1
  addss xmm0, _ps_am_pi_o_2
  mulss xmm0, _ps_am_2_o_pi

  cvttss2si ecx, xmm0
  movss xmm5, _ps_am_1
  mov  edx, ecx
  shl  edx, (31 - 1)
  cvtsi2ss xmm1, ecx
  and  edx, 0x80000000
  and  ecx, 0x1

  subss xmm0, xmm1
  movss xmm6, _sincos_masks[ecx * 4]
  minss xmm0, xmm5

  movss xmm1, _ps_sincos_p3
  subss xmm5, xmm0

  andps xmm5, xmm6
  movss xmm7, _ps_sincos_p2
  andnps xmm6, xmm0
  mov  temp, edx
  orps xmm5, xmm6
  movss xmm0, xmm5

  mulss xmm5, xmm5
  movss xmm4, _ps_sincos_p1
  movss xmm2, xmm5
  mulss xmm5, xmm1
  movss xmm1, _ps_sincos_p0
  addss xmm5, xmm7
  mulss xmm5, xmm2
  movss xmm3, temp
  addss xmm5, xmm4
  mulss xmm5, xmm2
  orps xmm0, xmm3
  addss xmm5, xmm1
  mulss xmm0, xmm5
 
  movss   x,    xmm0

 }

 return x;
}

float _SSE2_cos(float x) 
{
 __asm
 {
  movss xmm0, x
  movss xmm1, _ps_am_inv_sign_mask
  movss xmm2, _ps_am_pi_o_2
  movss xmm3, _ps_am_2_o_pi
  andps xmm0, xmm1
  addss xmm0, xmm2
  mulss xmm0, xmm3

  pxor xmm3, xmm3
  movd xmm5, _epi32_1
  movss xmm4, _ps_am_1
  cvttps2dq xmm2, xmm0
  pand xmm5, xmm2
  movd xmm1, _epi32_2
  pcmpeqd xmm5, xmm3
  cvtdq2ps xmm6, xmm2
  pand xmm2, xmm1
  pslld xmm2, (31 - 1)

  subss xmm0, xmm6
  movss xmm3, _ps_sincos_p3
  minss xmm0, xmm4
  subss xmm4, xmm0
  andps xmm0, xmm5
  andnps xmm5, xmm4
  orps xmm0, xmm5

  movaps xmm1, xmm0
  movss xmm4, _ps_sincos_p2
  mulss xmm0, xmm0
  movss xmm5, _ps_sincos_p1
  orps xmm1, xmm2
  movaps xmm7, xmm0
  mulss xmm0, xmm3
  movss xmm6, _ps_sincos_p0
  addss xmm0, xmm4
  mulss xmm0, xmm7
  addss xmm0, xmm5
  mulss xmm0, xmm7
  addss xmm0, xmm6
  mulss xmm0, xmm1
  movss   x,    xmm0
 }

 return x;
}

float _SSE_Sqrt(float x)
{

 float root = 0.f;
 _asm
 {
  sqrtss  xmm0, x
  movss  root, xmm0
 }

 return root;
}