hello there! I've been struggling for almost half a day with a specific problem concerning SSE/AVX

[tender hare]

here's the starter code

// Compile with: gcc -mavx -O2 operation8x8.c -o operation8x8
#include <stdio.h>
#include <stdlib.h>
#include <immintrin.h>

int main() {

float *a,*b;
__m128 *pa;    // sse (128-bit) pointer
__m256 *pb;    // avx (256-bit) pointer

int i;

  // allocate test arrays - request alignment at 32 bytes
  i = posix_memalign((void **)&a,32,8*8*sizeof(float));
  if (i!=0) exit(1);
  i = posix_memalign((void **)&b,32,8*8*sizeof(float));
  
  //initialize float input test array a
  for (i=0;i<8*8;i++) {
    a[i] = i;
  }
  
  // alias the sse/avx pointers to input and output test arrays
  pa = (__m128 *)a; pb = (__m256 *)b;
  
  // operation, from a to b
  for (int i=0;i<2;i++) {
    __m256 t1 = _mm256_insertf128_ps(_mm256_castps128_ps256(pa[0+i]),pa[8+i],1);
    __m256 t2 = _mm256_insertf128_ps(_mm256_castps128_ps256(pa[2+i]),pa[10+i],1);
    __m256 u1 = _mm256_unpacklo_ps(t1,t2);
    __m256 u2 = _mm256_unpackhi_ps(t1,t2);
  
    __m256 t3 = _mm256_insertf128_ps(_mm256_castps128_ps256(pa[4+i]),pa[12+i],1);
    __m256 t4 = _mm256_insertf128_ps(_mm256_castps128_ps256(pa[6+i]),pa[14+i],1);
    __m256 u3 = _mm256_unpacklo_ps(t3,t4);
    __m256 u4 = _mm256_unpackhi_ps(t3,t4);
  
    pb[4*i+0] = _mm256_shuffle_ps(u1,u3,0x44);
    pb[4*i+1] = _mm256_shuffle_ps(u1,u3,0xee);
    pb[4*i+2] = _mm256_shuffle_ps(u2,u4,0x44);
    pb[4*i+3] = _mm256_shuffle_ps(u2,u4,0xee);
  }
  
  // free arrays
  free(b); free(a); 
  return 0;
}```

[gray iron]

When your question is answered use !solved to mark the question as resolved.

Remember to ask specific questions, provide necessary details, and reduce your question to its simplest form. For tips on how to ask a good question use !howto ask.

[tender hare]

The challenge is making this work with n x n float arrays instead of just 8 x 8 (n % 8 will always be 0 tho)

The gist of it is as such: a[i] = i, b[i] = a[i * 8 % len(a)], or at least that's what I understood when printing the results

Some context for the avx instructions here https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_insertf128_ps&ig_expand=3842, https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_unpacklo_ps&ig_expand=3842,7054, https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_unpackhi_ps&ig_expand=3842,7054,6997

It's been a while since I last used C, so I am forgetting some of the basics. In this case, I would probably need to split my float arrays into subarrays of 64 elements in order to use the avx operations

[tender hare]

Ok, I came up with an answer, here it is:

.

!solved