Problems with AVX computations: can I run avx2 codes?

Question

I've been using Intel's SSE instructions with good performance gains and, recently, I tried to use AVX instructions.

The problem is: I can compile my avx instructions, but I cannot run them. The instructions are:

__m256i* avx1 = (__m256i*)vct1;
__m256i* avx2 = (__m256i*)vct2;
__m256i* avx3 = (__m256i*)vct3;

__m256i va0, va1, va2;            

va2 = _mm256_and_si256(va0, va1);

The only way I can compile these instructions is by having -mavx2 flag. Without this flag I can't compile the code and I received this error:

"sseAND.cpp:124:33: error: ‘_mm256_and_si256’ was not declared in this scope
  va2 = _mm256_and_si256(va0, va1); 
"

Ok, after compilling with -mavx2 and running the code, I received this error messagem:

"Illegal Instruction".

Running /proc/cpuinfo, I saw my Sandy Bridge CPU has support only for AVX instructions, not AVX2.

The question Is: is there anything wrong with my code or with my gcc flags?

Can I run the avx instructions I put below in my Sandybridge?

Thanks for your help.

Answer 1

Sandy Bridge only has AVX support. You're out of luck there unless you buy a Haswell CPU.

Bochs has support for AVX2 emulation since v2.5, but it will be much slower, you only can test the correctness of your code.