Yes, you're right, I am calling the AES round once per `next()` call. But note that this his could be rewritten as
if(counter_ >= 4) {
counter_ = 0;
u_.words_[0] = _mm_aesenc_si128(u_.words_[0], key_);
}
return u_.state_[counter_++];
Which only uses one AES call every 4 words (and only one block of storage). Instead, I chose to avoid the `if` and compute blocks ahead of time, which makes for more predictable performance.
You are, of course, right about the actual point you're making. And calling _mm_aesenc_si128() once per 4 calls to next() may well suffice to pass a statistical test. Then again, even an LSFR passes most statistical tests...