Vectorized unsafe IP address parsing v3

#include <stdio.h>
#include <time.h>
#include <vector>
#include <string>
#include <algorithm>
#include <smmintrin.h>
#include <stdint.h>
#include <string.h>
#ifdef _MSC_VER
    #define ALIGN(n) __declspec(align(n))
#else
    #define ALIGN(n) __attribute__((aligned(n)))
#endif

typedef unsigned int UINT32;

//=============================================== original =================================================

inline UINT32 GetIP(const char *p)
{
    UINT32 dwIP=0,dwIP_Part=0;
    while(true)
    {
        if(p[0] == 0)
        {
            dwIP = (dwIP << 8) | dwIP_Part;
            break;
        }
        if(p[0]=='.')
        {
            dwIP = (dwIP << 8) | dwIP_Part;
            dwIP_Part = 0;
           p++;
        }
        dwIP_Part = (dwIP_Part*10)+(p[0]-'0');
        p++;
    }
    return dwIP;
}

//=============================================== stgatilov =================================================

__m128i shuffleTable[65536];    //can be reduced 256x times, see solution by @IwillnotexistIdonotexist below
void MyInit() {
    memset(shuffleTable, -1, sizeof(shuffleTable));
    int len[4];
    for (len[0] = 1; len[0] <= 3; len[0]++)
        for (len[1] = 1; len[1] <= 3; len[1]++)
            for (len[2] = 1; len[2] <= 3; len[2]++)
                for (len[3] = 1; len[3] <= 3; len[3]++) {
                    int slen = len[0] + len[1] + len[2] + len[3] + 4;
                    int rem = 16 - slen;
                    for (int rmask = 0; rmask < 1<<rem; rmask++) {
//                    { int rmask = (1<<rem)-1;    //note: only maximal rmask is possible if strings are zero-padded
                        int mask = 0;
                        char shuf[16] = {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1};
                        int pos = 0;
                        for (int i = 0; i < 4; i++) {
                            for (int j = 0; j < len[i]; j++) {
                                shuf[(3-i) * 4 + (len[i]-1-j)] = pos;
                                pos++;
                            }
                            mask ^= (1<<pos);
                            pos++;
                        }
                        mask ^= (rmask<<slen);
                        _mm_store_si128(&shuffleTable[mask], _mm_loadu_si128((__m128i*)shuf));
                    }
                }
}
inline UINT32 MyGetIP(const char *str) {
    __m128i input = _mm_lddqu_si128((const __m128i*)str);
//  for (int i = 0; i < 16; i++) printf("%c ", input.m128i_u8[i]); printf("\n");
    input = _mm_sub_epi8(input, _mm_set1_epi8('0'));
//  for (int i = 0; i < 16; i++) printf("%d ", int(input.m128i_u8[i])); printf("\n");
    __m128i cmp = input;
//  for (int i = 0; i < 16; i++) printf("%c", cmp.m128i_i8[i] < 0 ? 'X' : '.'); printf("\n");
    uint64_t mask = _mm_movemask_epi8(cmp);
//  printf("%d\n", mask);
    __m128i shuf = shuffleTable[mask];
//  for (int i = 0; i < 16; i++) printf("%d ", int(shuf.m128i_i8[i])); printf("\n");
    __m128i arr = _mm_shuffle_epi8(input, shuf);
//  for (int i = 0; i < 16; i++) printf("%d ", int(arr.m128i_u8[i])); printf("\n");
    __m128i coeffs = _mm_set_epi8(0, 100, 10, 1, 0, 100, 10, 1, 0, 100, 10, 1, 0, 100, 10, 1);
    __m128i prod = _mm_maddubs_epi16(coeffs, arr);
//  for (int i = 0; i < 8; i++) printf("%d ", int(prod.m128i_u16[i])); printf("\n");
    prod = _mm_hadd_epi16(prod, prod);
//  for (int i = 0; i < 4; i++) printf("%d ", int(prod.m128i_u16[i])); printf("\n");
    prod = _mm_shuffle_epi8(prod, _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 6, 4, 2, 0));
//  for (int i = 0; i < 4; i++) printf("%d ", int(prod.m128i_u8[i])); printf("\n");
    return _mm_extract_epi32(prod, 0);
//  return (UINT32(_mm_extract_epi16(prod, 1)) << 16) + UINT32(_mm_extract_epi16(prod, 0));
}

//==================================== @IwillnotexistIdonotexist ===============================================
//lookup table is compressed to 4Kb
//taken from http://pastebin.com/XJ7w9nq1/* DEEP MAGIC BEGINS HERE. */
    return (unsigned)(v*0x008d981a)>>24;
}

inline UINT32 MyGetIP_comp(const char *str) {
  __m128i input = _mm_lddqu_si128((const __m128i*)str);   //"192.167.1.3"
  uint64_t zm = _mm_movemask_epi8(_mm_cmpeq_epi8(input, _mm_setzero_si128()));
  zm ^= (zm-1);                                           //(1<<12) - 1
  input = _mm_sub_epi8(input, _mm_set1_epi8('0'));        //1 9 2 254 1 6 7 254 1 254 3 208 245 0 8 40
  __m128i cmp = input;                                    //...X...X.X.XX...  (signs)
  uint64_t mask = _mm_movemask_epi8(cmp);                 //6792 - magic index
  mask &= zm;
  uint64_t hashmask = perfecthash(mask);
  __m128i shuf = ((const __m128i*)TBL)[hashmask];         //10 -1 -1 -1 8 -1 -1 -1 6 5 4 -1 2 1 0 -1
  __m128i arr = _mm_shuffle_epi8(input, shuf);            //3 0 0 0 | 1 0 0 0 | 7 6 1 0 | 2 9 1 0
  __m128i coeffs = _mm_set_epi8(0, 100, 10, 1, 0, 100, 10, 1, 0, 100, 10, 1, 0, 100, 10, 1);
  __m128i prod = _mm_maddubs_epi16(coeffs, arr);          //3 0 | 1 0 | 67 100 | 92 100
  prod = _mm_hadd_epi16(prod, prod);                      //3 | 1 | 167 | 192 | ? | ? | ? | ?
  __m128i imm = _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 6, 4, 2, 0);
  prod = _mm_shuffle_epi8(prod, imm);                     //3 1 167 192 0 0 0 0 0 0 0 0 0 0 0 0
  return _mm_extract_epi32(prod, 0);
}

//========================================== TESTING =================================================

static const int CNT = 16<<10;

int main() {
  MyInit();
  printf("%08X\n", MyGetIP("192.167.1.3"));

  std::vector<std::string> samples;
  for (int i = 0; i < CNT; i++) {
    std::string str;
    for (int j = 0; j < 4; j++) {
      if (j) str += '.';
      int x = rand() % 256;
      char qq[16];
      sprintf(qq, "%d", x);
      str += qq;
    }
    samples.push_back(str);

//    printf("%s\n", samples[i].c_str());
    auto a = GetIP(samples.back().c_str());
    auto b = MyGetIP(samples.back().c_str());
  auto c = MyGetIP_comp(samples.back().c_str());
    if (a != b || a != c || b != c)
      printf("%s: %08X vs %08X vs %08X\n", samples[i].c_str(), a, b, c);

  }


{
  int start = clock();
  int sum = 0;
  for (int i = 0; i < 1<<27; i++) {
      const char *input = samples[(i+0) & (CNT-1)].c_str();
      sum += MyGetIP(input);
  }
  int elapsed = clock() - start;
  printf("Time = %0.3lf   (%d)\n", double(elapsed) / CLOCKS_PER_SEC, sum);
}
{
  int start = clock();
  int sum = 0;
  for (int i = 0; i < 1<<27; i++) {
      const char *input = samples[(i+0) & (CNT-1)].c_str();
      sum += MyGetIP_comp(input);
  }
  int elapsed = clock() - start;
  printf("Time = %0.3lf   (%d)\n", double(elapsed) / CLOCKS_PER_SEC, sum);
}
{
  int start = clock();
  int sum = 0;
  for (int i = 0; i < 1<<27; i++) {
      const char *input = samples[i & (CNT-1)].c_str();
      sum += GetIP(input);
  }
  int elapsed = clock() - start;
  printf("Time = %0.3lf   (%d)\n", double(elapsed) / CLOCKS_PER_SEC, sum);
}

  return 0;
}