_mm256_loadu2_m128i intrinsic not available under g++?

Question

I'm trying to use the AVX2 intrinsic _mm256_loadu2_m128i, but it seems g++ 4.8.2 doesn't have it.

Is there any way to get it?

Answer 1

I have the same problem in GCC and Clang. However it compiles in ICC. You can test this with GCC, Clang, and ICC at http://gcc.godbolt.org/

Note, this is a AVX intrinsic not AVX2. Most of the 256-bit integer load and store intrinsics only need AVX. AVX2 provides some gather and mask loads but everything else only needs AVX.

Since Haswell can load two 128-bit values at once you can achieve the same effect as _mm256_loadu2_m128i using _mm256_inserti128_si256. Something like this

#include <immintrin.h>

int main() {
    int low[4];
    int high[4];
    _mm256_inserti128_si256(_mm256_castsi128_si256(
        _mm_loadu_si128((__m128i*)low)),
        _mm_loadu_si128((__m128i*)high),1);
}

Answer 2

Guess there is no such intrinsic named _mm256_loadu_m128i. All I can find with _mm256_load are _mm256_load_pd, _mm256_load_ps, _mm256_load_si256, _mm256_loadu_pd, _mm256_loadu_psand _mm256_loadu_si256. These all come with AVX and the gcc header is immintrin.h.