From the flash attention paper:

We also extend FlashAttention to block-sparse attention, yielding an approximate attention algorithm that is faster than any existing approximate attention method.

So I assume they are using the approximate version as they also have an exact version.