Alignment of data for SSE/AVX does not seem to matter

[Martinsos]

Documentation says that data which will be loaded/stored by SSE must be memory aligned to 16/32 bytes, or loading/storing will be much slower and possibly cause exceptions.

I do that like this:

int myArray[4] __attribute__((aligned(16)));

However, if I remove alignment, or align it to some different, wrong value, nothing changes in speed!
I am not sure if loads/stores actually become slower but it does not affect overall speed of execution of opal, or am I actually doing something else wrong -> maybe I am aligning wrong all the time? I should investigate this, to make sure I am doing this correctly.