Vectorized version ?

Using a state vector containing the information of 4 seeds, we could provide a new API where we get 4 random numbers at a time (using SIMD instructions for example).

The state vector would have to be "interleaved" to optimize cache usage, so that when reading or writing 4 elements, we read or write contiguous memory.

(I use the number 4 but it could be a different number depending on the Variate type (to build a double we need 2 Word32), and on the vectorization primitives available.)

I'm not familiar with haskell vector primitives but looking at GHC.Exts I saw some there, so I guess it is possible to go this way... What do you think?

Start by reviewing the library's existing random-number API and state representation, then investigate the GHC.Exts vector primitives mentioned in the issue. Define the vector width, interleaved state layout, and supported Variate types before determining what tests and benchmarks would demonstrate a complete implementation.

Written by the indexing model from the issue text.