bartmassey / bingomatic Goto Github PK

Right now, cards are played sequentially. An arbitrary number of cards should be playable in parallel.

Since "all cards are equal" statistically, there is no need to shuffle up a new card for each new game. Random draws should be sufficient to get our statistics. This would improve performance substantially, since shuffling up a card is way more expensive than running a game.

The marker bitboards (or counters once we move to counters in issue #5) will still have to be cleared for each game.

The "rdrand" random number stuff in this repo should be moved into its own repo. Also, should be fixed in the non-RDRAND case to use a faster and more reliable generator than lrand48(): preferably something with a large state.

Right now, RDRAND is the limiting factor in performance. Its output is not being used very efficiently, but still it is about 5x slower than multiple calls to lrand48(). Its quality still justifies having it, I guess.

It would be nice to use the on-chip SIMD support for integers bigger than 64 bits. There are some nice things that could be done with this.

The current implementation of randrange() is a hacked-up version of somebody's blog post code. An effort should be made to verify that it is correct.

As stated in issue #2, right now the performance bottleneck is the RNG. Once that is fixed, the bitboards should be replaced with counters to improve performance.

An array of counters per card can be used along with a per-marker array of counter indices to bump. This should reduce the per-card win test cost quite a bit.

The center "free square" is currently ignored. Some sort of handling should be introduced.