perf.log

       │ 80:   mov    %rsi,%rdx                                                                                                                                                                                     ▒
       │       xor    %ecx,%ecx                                                                                                                                                                                     ▒
       │       nop                                                                                                                                                                                                  ▒
  1.70 │ 88:   vaddps (%rdx),%xmm3,%xmm0                                                                                                                                                                            ▒
 55.13 │       add    $0x1,%ecx                                                                                                                                                                                     ▒
  0.82 │       add    $0x60,%rdx                                                                                                                                                                                    ▒
  1.59 │       vmovap %xmm0,-0x60(%rdx)                                                                                                                                                                             ▒
  4.75 │       vaddps -0x50(%rdx),%xmm2,%xmm0                                                                                                                                                                       ▒
  6.60 │       vmovap %xmm0,-0x50(%rdx)                                                                                                                                                                             ▒
  3.22 │       cmp    $0x61a8,%ecx                                                                                                                                                                                  ▒
       │     ↑ jne    88                                                                                                                                                                                            ▒
       │       mov    %rdi,%rcx                                                                                                                                                                                     ▒
       │       xor    %edx,%edx                                                                                                                                                                                     ▒
       │     ↓ jmp    c7                                                                                                                                                                                            ▒
       │       nop                                                                                                                                                                                                  ▒
  4.06 │ b8:   add    $0x1,%edx                                                                                                                                                                                     ▒
       │       add    $0x60,%rcx                                                                                                                                                                                    ▒
       │       cmp    $0x61a8,%edx                                                                                                                                                                                  ▒
       │     ↓ je     de                                                                                                                                                                                            ▒
  3.10 │ c7:   test   $0x1,%dl                                                                                                                                                                                      ▒
       │     ↑ jne    b8                                                                                                                                                                                            ▒
  4.45 │       add    $0x1,%edx                                                                                                                                                                                     ▒
       │       prefet (%rcx)                                                                                                                                                                                        ▒
 14.57 │       add    $0x60,%rcx                                                                                                                                                                                    ▒
       │       cmp    $0x61a8,%edx                                                                                                                                                                                  ▒
       │     ↑ jne    c7                                                                                                                                                                                            ▒
  0.01 │ de:   sub    $0x1,%eax                                                                                                                                                                                     ▒
       │     ↑ jne    80                                                                                                                                                                                            ▒
       │       vmovap %xmm2,0x10(%rsp)                                                                                                                                                                              ▒
       │       vmovap %xmm3,(%rsp)                                                                                                                                                                                  ▒
       │     → callq  uv_hrtime                                                                                                                                                                                     ▒
       │       vmovap (%rsp),%xmm3                                                                                                                                                                                  ▒
       │       sub    %rbp,%rax                                                                                                                                                                                     ▒
       │       vmovap 0x10(%rsp),%xmm2                                                                                                                                                                              ◆
       │       mov    %rax,%rbp

test.cpp

struct particle_group {
  __m128 pos_x;
  __m128 pos_y;
  __m128 force_x;
  __m128 force_y;
  __m128 vel_x;
  __m128 vel_y;
};



void add_force(float x, float y) {
  __m128 force_x = _mm_set_ps(x, x, x, x);
  __m128 force_y = _mm_set_ps(y, y, y, y);

  int const prefetch_count = power_of_two_below(CACHE_LINE_SIZE / sizeof(particles[0]));
  __m128 tmp_x;
  __m128 tmp_y;
  for(int i = 0; i < GROUP_COUNT; ++i){

    if(i % 256 == 0) {
      // _mm_prefetch(particles + i + 1, _MM_HINT_T0);
      // _mm_prefetch(particles + i + prefetch_count, _MM_HINT_T0);
    }
    tmp_x = particles[i].force_x;
    tmp_y = particles[i].force_y;

    particles[i].force_x = _mm_add_ps(tmp_x, force_x);
    particles[i].force_y = _mm_add_ps(tmp_y, force_y);

  }
}