static
void
gf_w16_bytwo_b_sse_region_2_noxor(gf_region_data *rd, struct gf_w16_bytwo_data *btd)
{
#ifdef   INTEL_SSE4
  int i;
  uint8_t *d8, *s8, tb;
  __m128i pp, m1, m2, t1, t2, va, vb;

  s8 = (uint8_t *) rd->s_start;
  d8 = (uint8_t *) rd->d_start;

  pp = _mm_set1_epi16(btd->prim_poly&0xffff);
  m1 = _mm_set1_epi16((btd->mask1)&0xffff);
  m2 = _mm_set1_epi16((btd->mask2)&0xffff);

  while (d8 < (uint8_t *) rd->d_top) {
    va = _mm_load_si128 ((__m128i *)(s8));
    SSE_AB2(pp, m1, m2, va, t1, t2);
    _mm_store_si128((__m128i *)d8, va);
    d8 += 16;
    s8 += 16;
  }
#endif
}

static
void
gf_w16_bytwo_b_sse_region_2_xor(gf_region_data *rd, struct gf_w16_bytwo_data *btd)
{
#ifdef   INTEL_SSE4
  int i;
  uint8_t *d8, *s8, tb;
  __m128i pp, m1, m2, t1, t2, va, vb;

  s8 = (uint8_t *) rd->s_start;
  d8 = (uint8_t *) rd->d_start;

  pp = _mm_set1_epi16(btd->prim_poly&0xffff);
  m1 = _mm_set1_epi16((btd->mask1)&0xffff);
  m2 = _mm_set1_epi16((btd->mask2)&0xffff);

  while (d8 < (uint8_t *) rd->d_top) {
    va = _mm_load_si128 ((__m128i *)(s8));
    SSE_AB2(pp, m1, m2, va, t1, t2);
    vb = _mm_load_si128 ((__m128i *)(d8));
    vb = _mm_xor_si128(vb, va);
    _mm_store_si128((__m128i *)d8, vb);
    d8 += 16;
    s8 += 16;
  }
#endif
}


static
void 
gf_w16_bytwo_b_sse_multiply_region(gf_t *gf, void *src, void *dest, gf_val_32_t val, int bytes, int xor)
{
#ifdef   INTEL_SSE4
  int itb;
  uint8_t *d8, *s8;
  __m128i pp, m1, m2, t1, t2, va, vb;
  struct gf_w16_bytwo_data *btd;
  gf_region_data rd;
    
  if (val == 0) { gf_multby_zero(dest, bytes, xor); return; }
  if (val == 1) { gf_multby_one(gf, src, dest, bytes, xor); return; }

  gf_set_region_data(&rd, gf, src, dest, bytes, val, xor, 16);
  gf_do_initial_region_alignment(&rd);

  btd = (struct gf_w16_bytwo_data *) ((gf_internal_t *) (gf->scratch))->private;

  if (val == 2) {
    if (xor) {
      gf_w16_bytwo_b_sse_region_2_xor(&rd, btd);
    } else {
      gf_w16_bytwo_b_sse_region_2_noxor(&rd, btd);
    }
    gf_do_final_region_alignment(&rd);
    return;
  }

  s8 = (uint8_t *) rd.s_start;
  d8 = (uint8_t *) rd.d_start;

  pp = _mm_set1_epi16(btd->prim_poly&0xffff);
  m1 = _mm_set1_epi16((btd->mask1)&0xffff);
  m2 = _mm_set1_epi16((btd->mask2)&0xffff);

  while (d8 < (uint8_t *) rd.d_top) {
    va = _mm_load_si128 ((__m128i *)(s8));
    vb = (!xor) ? _mm_setzero_si128() : _mm_load_si128 ((__m128i *)(d8));
    itb = val;
    while (1) {
      if (itb & 1) vb = _mm_xor_si128(vb, va);
      itb >>= 1;
      if (itb == 0) break;
      SSE_AB2(pp, m1, m2, va, t1, t2);
    }
    _mm_store_si128((__m128i *)d8, vb);
    d8 += 16;
    s8 += 16;
  }

  gf_do_final_region_alignment(&rd);
#endif
}