AES-NI vectorization improvements

!30 (merged) didn't implement an SSE-vectorized _mm_cvtepu64_pd equivalent because the stackoverflow solution didn't work. That turned out to be due to a bad optimization in GCC 5+ in fast-math mode. None of the other compilers (Clang, Intel, MSVC) have that issue, so we just disable fast-math for that function.

Also, we now use fused multiply-add if available.

pystencils/include/aesni_rand.h

@@ -49,8 +49,8 @@ QUALIFIERS __m128 _my_cvtepu32_ps(const __m128i v)
 #endif
 }
 
-#if !defined(__AVX512VL__) && defined(__FAST_MATH__) && defined(__GNUC__) && __GNUC__ >= 5
-__attribute__((optimize("no-fast-math")))
+#if !defined(__AVX512VL__) && defined(__GNUC__) && __GNUC__ >= 5
+__attribute__((optimize("no-associative-math")))
 #endif
 QUALIFIERS __m128d _my_cvtepu64_pd(const __m128i x)
 {