// compile
// gcc -Icontrib/simde -mavx -o avx-test avx.c && ./avx-test

// #include <immintrin.h> // native instructions
#include <simde/x86/avx512.h>
#include <stdio.h>

int main() {

  /* Initialize the two argument vectors */
  __m256 evens = _mm256_set_ps(2.0, 4.0, 6.0, 8.0, 10.0, 12.0, 14.0, 17.0);
  __m256 odds = _mm256_set_ps(1.0, 3.0, 5.0, 7.0, 9.0, 11.0, 13.0, 15.0);
  /* Display the elements of the result vector */
  __m256 result = _mm256_sub_ps(evens, odds);
  float* f = (float *)&result;
  printf("float  %f %f %f %f %f %f %f %f\n",
    f[0], f[1], f[2], f[3], f[4], f[5], f[6], f[7]);

  // try doubles
  __m256d evens1 = _mm256_set_pd(2.0, 4.0, 6.0, 8.0);
  __m256d odds1 = _mm256_set_pd(1.0, 3.0, 5.0, 6.0);
  /* Compute the difference between the two vectors */
  __m256d result1 = _mm256_sub_pd(evens1, odds1);

  /* Display the elements of the result vector */
  double* d = (double*)&result1;
  printf("double %f %f %f %f\n",
    d[0], d[1], d[2], d[3]);

  // initialize from RAM
  double l1[] = { 2.0, 4.0, 6.0, 8.8 };
  double l2[] = { 0.0, 3.0, 5.0, 7.0 };
  __m256d evens3 = _mm256_loadu_pd(l1); // pd — “packed double” — 4 doubles
  __m256d odds3 = _mm256_loadu_pd(l2);
  __m256d result3 = _mm256_sub_pd(evens3, odds3);
  double* d3 = (double*)&result3;
  printf("double %f %f %f %f\n",
    d3[0], d3[1], d3[2], d3[3]);

  simde__m256d evens4 = simde_mm256_loadu_pd(l1); // pd — “packed double” — 4 doubles
  simde__m256d odds4 = simde_mm256_loadu_pd(l2);
  simde__m256d result4 = simde_mm256_sub_pd(evens3, odds3);
  double* d4 = (double*)&result4;
  printf("double %f %f %f %f\n",
    d4[0], d4[1], d4[2], d4[3]);

  return 0;
}