If statements with comparison SSE in C
c, sse
Solution
You're close:
const __m128 P2f = _mm_set1_ps(2.0f);
const __m128 M2f = _mm_set1_ps(-2.0f);
for (int i = 0; i < n; i += 4)
{
__m128 xv = _mm_load_ps(x + i);
__m128 av = _mm_load_ps(a + i);
__m128 c1v = _mm_cmpgt_ps(xv, P2f);
__m128 c2v = _mm_cmplt_ps(xv, M2f);
__m128 cv = _mm_or_ps(c1v, c2v);
xv = _mm_and_ps(xv, cv);
av = _mm_add_ps(av, xv);
_mm_store_ps(a + i, av);
}
The trick is to `OR` the two comparison results and then use this combined result as a mask to zero out the X values which do not pass the test using a bitwise `AND` operation. You then add the masked X vector, which will add 0 or the original X value to each element of A according to the mask.
For the alternate version as mentioned in your comment below you'd do this:
const __m128 P2f = _mm_set1_ps(2.0f);
const __m128 M2f = _mm_set1_ps(-2.0f);
for (int i = 0; i < n; i += 4)
{
__m128 xv = _mm_load_ps(x + i);
__m128 av = _mm_load_ps(a + i);
__m128 c1v = _mm_cmpgt_ps(xv, P2f);
__m128 c2v = _mm_cmplt_ps(xv, M2f);
__m128 cv = _mm_or_ps(c1v, c2v);
xv = _mm_and_ps(P2f, cv); // <<< change this line to get a[i] += 2.0f
// instead of a[i] += x[i]
av = _mm_add_ps(av, xv);
_mm_store_ps(a + i, av);
}
For the third version you mention in later comments below (`a[i] *= 2.0`) it's slightly trickier, but you can do it by thinking of the expression as `a[i] += a[i]`:
const __m128 P2f = _mm_set1_ps(2.0f);
const __m128 M2f = _mm_set1_ps(-2.0f);
for (int i = 0; i < n; i += 4)
{
__m128 xv = _mm_load_ps(x + i);
__m128 av = _mm_load_ps(a + i);
__m128 c1v = _mm_cmpgt_ps(xv, P2f);
__m128 c2v = _mm_cmplt_ps(xv, M2f);
__m128 cv = _mm_or_ps(c1v, c2v);
xv = _mm_and_ps(av, cv)); // <<< change this line to get a[i] *= 2.0f (a[i] += a[i])
// instead of a[i] += x[i]
av = _mm_add_ps(av, xv);
_mm_store_ps(a + i, av);
}
Problem
I want to achieve this: ``` for (int i=0;i<n,i++){ if (x[i] > 2.0f || x[i] < -2.0f) a[i] += x[i]; } ``` I have gone this far but don't know what to do next: ``` __m128 P2f = _mm_set1_ps(2.0f); __m128 M2f = _mm_set1_ps(-2.0f); for(int i=0;i<n,i+=4){ __m128 xv = _mm_load_ps(x+i); __m128 av = _mm_load_ps(a+i); __m128 c1 = _mm_cmpgt_ps(xv, P2f); __m128 c2 = _mm_cmplt_ps(xv, M2f); __m128 or = _mm_or_ps(c1,c2); =???== av = _mm_add_ps(av, xv); _mm_store_ps(a+i, av); } ```