Hi All, This is my first time turning to the gcc mailing lists. I apologize if I have chosen the wrong one. I have a member function that is multiplying one array of complex numbers (stored as 2 arrays) by 3 other arrays of complex numbers (stored similarly) and outputting the results of those 3 sets of multiplications into different arrays. The arrays are all 64-byte aligned (even the ones in the shared_ptrs,) the array lengths are all powers of 2 and larger than 16, and I don’t do any aliasing. I have done my best to tell the compiler this via __restrict__, __builtin_assume_aligned, __builtin_unreachable, and assert. I know the code is vectorizable because clang did it off the bat, though the __builtin_assume_aligned does make it generate aligned load instructions instead of unaligned… Anyways, the code is here: https://godbolt.org/z/sd1KoYabf So, my question is: What am I missing that will make the autovectorizer issue vector instructions instead of scalar? Thank you all for your time, sorry to bug you with what I am sure is a relatively simple question. Cheers, James
