emulate Class — pytorch Architecture

Source Code

aten/src/ATen/cpu/vec/vec256/vec256_int.h lines 1133–1154

template <typename op_t>
Vectorized<int64_t> inline emulate(
    const Vectorized<int64_t>& a,
    const Vectorized<int64_t>& b,
    const op_t& op) {
  int64_t a0 = _mm256_extract_epi64(a, 0);
  int64_t a1 = _mm256_extract_epi64(a, 1);
  int64_t a2 = _mm256_extract_epi64(a, 2);
  int64_t a3 = _mm256_extract_epi64(a, 3);

  int64_t b0 = _mm256_extract_epi64(b, 0);
  int64_t b1 = _mm256_extract_epi64(b, 1);
  int64_t b2 = _mm256_extract_epi64(b, 2);
  int64_t b3 = _mm256_extract_epi64(b, 3);

  int64_t c0 = op(a0, b0);
  int64_t c1 = op(a1, b1);
  int64_t c2 = op(a2, b2);
  int64_t c3 = op(a3, b3);

  return _mm256_set_epi64x(c3, c2, c1, c0);
}