how invert __m128 into ints

Question

float a[4] = {1,2,3,4}, b[4] = {4,3,2,1};
uint32_t c[4];

int main() {

    __m128 pa = _mm_loadu_ps(a);

    __m128 pb = _mm_loadu_ps(b);
    __m128 pc = _mm_cmpgt_ps(pa, pb);
    _mm_storeu_ps((float*)c, pc);
    for (int i = 0;i  < 4; ++i) printf("%u\n", c[i]);
    return 0;
}

what is the correct instruction of _mm_storeu_ps((float*)c, pc)? here, c is a integer array... I don't think this way is good, any better?

Answer 1

There are two instructions to convert __m128 (float vector) into __m128i (int32_t vector) in SSE2: _mm_cvtps_epi32(with rounding) and _mm_cvttps_epi32(with truncation).

__m128i vi = _mm_cvttps_epi32(pc);
_mm_storeu_si128((__m128i *)c, vi);

If you can't use SSE2, you should convert float array to int array after storing pc into float array.

float d[4];
_mm_storeu_ps(d, pc);
c[0] = (int)d[0]; c[1] = (int)d[1]; c[2] = (int)d[2]; c[3] = (int)d[3];