从评论来看，似乎最好的方法是这样的...（未检测）

__m256i veca = _mm256_set1_epi16(a);           // a is 16bit value you're looking for...
__m256i cmp = _mm256_cmpeq_epi16(veca, vecb);  // vecb is the vector you want to search
uint32_t mask = _mm256_movemask_epi8(cmp);     // cmp elements are 0 or 0xffff if no-match or match resp.
uint32_t index = 0;                            // mask is most significant bit of each byte in cmp
if (mask != 0) {
  index = (30 - lzcnt32(mask))/2;              // index is 0-15, 0 on right (higher address).
} else {
  ..no match.
}