I was recently researching implementations of vectorized exponential functions. It's actually very simple to find code for single-precision, but very hard to find code for double-precision. I haven't been able to find any implementation of double-precision AVX exponentiation. The code from SSE can almost be adapted but the trick that is generally used to multiply by 2^n does not seem to be adaptable to AVX.
Does anyone know of such an implementation ?
If you are using Intel's ICC compiler then there is _mm256_exp_pd in the SVML library.
Alternatively Agner Fog's vector class library supports exp for double precision vectors.