sse inner product

dot

        vmovss  (%rax), %xmm1
        vmulss  (%rdx), %xmm1, %xmm1
        vaddss  %xmm1, %xmm0, %xmm0
        addq    $4, %rdx
        addq    $4, %rax
        decq    %rcx
        jne     LBB0_2

gistfile1.cpp

float dot(const vector<float>& x,
          const vector<float>& y) {
  float s = 0;
  for (size_t i = 0, size = x.size(); i < size; ++i) {
    s += x[i] * y[i];
  }
  return s;
}

float dot_sse(const vector<float>& x,
              const vector<float>& y) {
  float s1 = 0, s2 = 0, s3 = 0, s4 = 0;
  for (size_t i = 0, size = x.size(); i < size; i += 4) {
    s1 += x[i] * y[i];
    s2 += x[i + 1] * y[i + 1];
    s3 += x[i + 2] * y[i + 2];
    s4 += x[i + 3] * y[i + 3];
  }
  return s1 + s2 + s3 + s4;
}

sse_dot

        vmovss  (%rax,%rsi,4), %xmm1
        vinsertps       $16, 12(%rax,%rsi,4), %xmm1, %xmm1
        vinsertps       $32, 8(%rax,%rsi,4), %xmm1, %xmm1
        vinsertps       $48, 4(%rax,%rsi,4), %xmm1, %xmm1
        vmovss  (%rdx,%rsi,4), %xmm2
        vinsertps       $16, 12(%rdx,%rsi,4), %xmm2, %xmm2
        vinsertps       $32, 8(%rdx,%rsi,4), %xmm2, %xmm2
        vinsertps       $48, 4(%rdx,%rsi,4), %xmm2, %xmm2
        vmulps  %xmm2, %xmm1, %xmm1
        vaddps  %xmm1, %xmm0, %xmm0
        addq    $4, %rsi
        cmpq    %rcx, %rsi
        jb      LBB2_2