PHP SSE Compiler Intrinsics - Performance Arithmetic

[AceInfinity]

Here's an example I wrote using compiler intrinsics in GCC for the SSE instruction set to operate on multiple data simultaneously.

[NO-PARSE]#include <cstdio>
#include <cassert>
#include <immintrin.h>

void insert(__m64 &m, int value, unsigned int index)
{
  assert(index < 4);
  switch (index)
  {
    case 0: m = _mm_insert_pi16(m, value, 0); break;
    case 1: m = _mm_insert_pi16(m, value, 1); break;
    case 2: m = _mm_insert_pi16(m, value, 2); break;
    case 3: m = _mm_insert_pi16(m, value, 3); break;
  }
}

int extract(const __m64 &m, unsigned int index)
{
  assert(index < 4);
  switch (index)
  {
    case 0:  return _mm_extract_pi16(m, 0);
    case 1:  return _mm_extract_pi16(m, 1);
    case 2:  return _mm_extract_pi16(m, 2);
    default: return _mm_extract_pi16(m, 3);
  }
}

int main()
{
  // 32 byte aligned memory block for 8 int elements
  const int n         =  8;
  const int mem_block = 32;
  int *arr = (int *)_mm_malloc(n * sizeof(int), mem_block);

  // initialize array data
  for (int i = 0; i < n; ++i) arr[i] = i;

  __m64 a, b; // variables for SSE

  // insert values at selector
  for (unsigned int i = 0; i < 4; ++i)
  {
    int k = i * 2;
    insert(a, arr[k], i);
    insert(b, arr[k + 1] * (arr[k] + 1), i);
  }

  _mm_free(arr); // free allocated memory

  // retrieve results for all averages
  __m64 result = _mm_avg_pu16(a, b);
  for (unsigned int i = 0; i < 4; ++i)
    printf("__m64 average a[%u] : b[%u] = %d : %d = %d\n",
           i, i, extract(a, i), extract(b, i),
           extract(result, i)
          );
}[/NO-PARSE]

Result:

[NO-PARSE]__m64 average a[0] : b[0] = 0 : 1 = 1
__m64 average a[1] : b[1] = 2 : 9 = 6
__m64 average a[2] : b[2] = 4 : 25 = 15
__m64 average a[3] : b[3] = 6 : 49 = 28[/NO-PARSE]